[AMDGPU] Match signed dot4/8 pattern.

[FarhanaAleen]

Summary

This patch matches signed dot4 and dot8 pattern.

[msearles]

Can the tests be reduced/made more flexible? E.g., the tests previously used FileCheck variables ( [[FF:s[0-9]+]] .

[FarhanaAleen]

In D52520#1251236, @msearles wrote:

Can the tests be reduced/made more flexible? E.g., the tests previously used FileCheck variables ( [[FF:s[0-9]+]] .

Hi Mark, thanks for looking into this.

Yes, we used variables in some of the previous files. Later we came to a decision that it's better to just have the pure/untouched output of update_llc_test_checks (as you can see the later test files are written following that) due to two reasons: 1) updating reduced test-checks could be cumbersome when there is a change in the instructions 2). Writing regexes on every assembly instructions for four different architectures is very time consuming since this is no automated way of doing it.

[msearles]

LGTM

lib/Target/AMDGPU/VOP3PInstructions.td

 let ClampLo = 0, ClampHi = 1 in {
 def V_FMA_MIXHI_F16 : VOP3_VOP3PInst<"v_fma_mixhi_f16", VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>, 1>;
 }
 }
 
 defm : MadFmaMixPats<fma, V_FMA_MIX_F32, V_FMA_MIXLO_F16, V_FMA_MIXHI_F16>;
 }
 
-class Srl<int N> : PatFrag<(ops node:$src),
-                           (srl node:$src, (i32 N))>;
+// Defines patterns that extract signed 4bit from each Idx[0].
+foreach Idx = [[0,28],[4,24],[8,20],[12,16],[16,12],[20,8],[24,4]] in
+  def ExtractSigned4bit_#Idx[0] : PatFrag<(ops node:$src),
+                                          (sra (shl node:$src, (i32 Idx[1])), (i32 28))>;
 
-foreach Bits = 1-7 in
-  def srl#!shl(Bits, 2) : Srl<!shl(Bits, 2)>;
-
-class Extract_U<int FromBitIndex, int BitMask> : PatFrag<
+// Defines code pattern that extracts U(unsigned/signed) 4/8bit from FromBitIndex.
+class Extract<int FromBitIndex, int BitMask, bit U>: PatFrag<
   (ops node:$src),
-  !if (!or (!and (!eq (BitMask, 255), !eq (FromBitIndex, 24)),
-            !and (!eq (BitMask, 15), !eq (FromBitIndex, 28))), // last element
-       (!cast<Srl>("srl"#FromBitIndex) node:$src),
+  !if (!or (!and (!eq (BitMask, 255), !eq (FromBitIndex, 24)), !eq (FromBitIndex, 28)), // last element
+       !if (U, (srl node:$src, (i32 FromBitIndex)), (sra node:$src, (i32 FromBitIndex))),
        !if (!eq (FromBitIndex, 0), // first element
-            (and node:$src, (i32 BitMask)),
-            (and (!cast<Srl>("srl"#FromBitIndex) node:$src), (i32 BitMask))))>;
+            !if (U, (and node:$src, (i32 BitMask)),
+                 !if (!eq (BitMask, 15), (!cast<PatFrag>("ExtractSigned4bit_"#FromBitIndex) node:$src),
+                                         (sext_inreg node:$src, i8))),
+            !if (U, (and (srl node:$src, (i32 FromBitIndex)), (i32 BitMask)),
+                 !if (!eq (BitMask, 15), (!cast<PatFrag>("ExtractSigned4bit_"#FromBitIndex) node:$src),
+                      (sext_inreg (srl node:$src, (i32 FromBitIndex)), i8)))))>;
+
 
+foreach Type = ["I", "U"] in
   foreach Index = 0-3 in {
     // Defines patterns that extract each Index'ed 8bit from an unsigned
     // 32bit scalar value;
-  def U#Index#"_8bit" : Extract_U<!shl(Index, 3),
-                                  255>;
+    def #Type#Index#"_8bit" : Extract<!shl(Index, 3), 255, !if (!eq (Type, "U"), 1, 0)>;
 
     // Defines multiplication patterns where the multiplication is happening on each
     // Index'ed 8bit of a 32bit scalar value.
-  def MulU_Elt#Index : PatFrag<
+
+    def Mul#Type#_Elt#Index : PatFrag<
       (ops node:$src0, node:$src1),
-    (AMDGPUmul_u24_oneuse (!cast<Extract_U>("U"#Index#"_8bit") node:$src0),
-                          (!cast<Extract_U>("U"#Index#"_8bit") node:$src1))>;
+      (!cast<HasOneUseBinOp>(!if (!eq (Type, "I"), AMDGPUmul_i24_oneuse, AMDGPUmul_u24_oneuse))
+                            (!cast<Extract>(#Type#Index#"_8bit") node:$src0),
+                            (!cast<Extract>(#Type#Index#"_8bit") node:$src1))>;
   }
 
 // Different variants of dot8 patterns cause a huge increase in the compile time.
 // Define non-associative/commutative add/mul to prevent permutation in the dot8
 // pattern.
 def NonACAdd        : SDNode<"ISD::ADD"       , SDTIntBinOp>;
 def NonACAdd_oneuse : HasOneUseBinOp<NonACAdd>;
 
 def NonACAMDGPUmul_u24        : SDNode<"AMDGPUISD::MUL_U24"       , SDTIntBinOp>;
 def NonACAMDGPUmul_u24_oneuse : HasOneUseBinOp<NonACAMDGPUmul_u24>;
 
+def NonACAMDGPUmul_i24        : SDNode<"AMDGPUISD::MUL_I24"       , SDTIntBinOp>;
+def NonACAMDGPUmul_i24_oneuse : HasOneUseBinOp<NonACAMDGPUmul_i24>;
+
+foreach Type = ["I", "U"] in
   foreach Index = 0-7 in {
     // Defines patterns that extract each Index'ed 4bit from an unsigned
     // 32bit scalar value;
-  def U#Index#"_4bit" : Extract_U<!shl(Index, 2),
-                                  15>;
+    def #Type#Index#"_4bit" : Extract<!shl(Index, 2), 15, !if (!eq (Type, "U"), 1, 0)>;
 
     // Defines multiplication patterns where the multiplication is happening on each
     // Index'ed 8bit of a 32bit scalar value.
-  def MulU#Index#"_4bit" : PatFrag<
+    def Mul#Type#Index#"_4bit" : PatFrag<
       (ops node:$src0, node:$src1),
-    (NonACAMDGPUmul_u24_oneuse (!cast<Extract_U>("U"#Index#"_4bit") node:$src0),
-                               (!cast<Extract_U>("U"#Index#"_4bit") node:$src1))>;
+      (!cast<HasOneUseBinOp>(!if (!eq (Type, "I"), NonACAMDGPUmul_i24_oneuse, NonACAMDGPUmul_u24_oneuse))
+                             (!cast<Extract>(#Type#Index#"_4bit") node:$src0),
+                             (!cast<Extract>(#Type#Index#"_4bit") node:$src1))>;
   }
 
 class UDot2Pat<Instruction Inst> : GCNPat <
   (add (add_oneuse (AMDGPUmul_u24_oneuse (srl i32:$src0, (i32 16)),
                                          (srl i32:$src1, (i32 16))), i32:$src2),
        (AMDGPUmul_u24_oneuse (and i32:$src0, (i32 65535)),
                              (and i32:$src1, (i32 65535)))
    ),
   (Inst (i32 8), $src0, (i32 8), $src1, (i32 8), $src2, (i1 0))

 defm : DotPats<int_amdgcn_sdot4, V_DOT4_I32_I8>;
 defm : DotPats<int_amdgcn_udot4, V_DOT4_U32_U8>;
 defm : DotPats<int_amdgcn_sdot8, V_DOT8_I32_I4>;
 defm : DotPats<int_amdgcn_udot8, V_DOT8_U32_U4>;
 
 def : UDot2Pat<V_DOT2_U32_U16>;
 def : SDot2Pat<V_DOT2_I32_I16>;
 
+foreach Type = ["U", "I"] in
   def : GCNPat <
     !cast<dag>(!foldl((i32 i32:$src2), [0, 1, 2, 3], lhs, y,
-                    (add_oneuse lhs, (!cast<PatFrag>("MulU_Elt"#y) i32:$src0, i32:$src1)))),
-  (V_DOT4_U32_U8 (i32 8), $src0, (i32 8), $src1, (i32 8), $src2, (i1 0))
->;
+                      (add_oneuse lhs, (!cast<PatFrag>("Mul"#Type#"_Elt"#y) i32:$src0, i32:$src1)))),
+    (!cast<VOP3PInst>("V_DOT4_"#Type#"32_"#Type#8) (i32 8), $src0, (i32 8), $src1, (i32 8), $src2, (i1 0))>;
 
+foreach Type = ["U", "I"] in
   def : GCNPat <
-  !cast<dag>(!foldl((add_oneuse i32:$src2, (MulU0_4bit i32:$src0, i32:$src1)), [1, 2, 3, 4, 5, 6, 7], lhs, y,
-                    (NonACAdd_oneuse lhs, (!cast<PatFrag>("MulU"#y#"_4bit") i32:$src0, i32:$src1)))),
-  (V_DOT8_U32_U4 (i32 8), $src0, (i32 8), $src1, (i32 8), $src2, (i1 0))
->;
+    !cast<dag>(!foldl((add_oneuse i32:$src2, (!cast<PatFrag>("Mul"#Type#"0_4bit") i32:$src0, i32:$src1)),
+                      [1, 2, 3, 4, 5, 6, 7], lhs, y,
+                      (NonACAdd_oneuse lhs, (!cast<PatFrag>("Mul"#Type#y#"_4bit") i32:$src0, i32:$src1)))),
+    (!cast<VOP3PInst>("V_DOT8_"#Type#"32_"#Type#4) (i32 8), $src0, (i32 8), $src1, (i32 8), $src2, (i1 0))>;
 
 } // End SubtargetPredicate = HasDLInsts
 
 multiclass VOP3P_Real_vi<bits<10> op> {
   def _vi : VOP3P_Real<!cast<VOP3_Pseudo>(NAME), SIEncodingFamily.VI>,
             VOP3Pe <op, !cast<VOP3_Pseudo>(NAME).Pfl> {
     let AssemblerPredicates = [HasVOP3PInsts];
     let DecoderNamespace = "VI";

test/CodeGen/AMDGPU/idot4.ll

-; RUN: llc -mtriple=amdgcn -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX789 %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX789 %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX789 %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx906 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-DL %s
-
-; GCN-LABEL: {{^}}udot4_acc32:
-; GCN:       ; %bb.0: ; %entry
-; GCN-NEXT:       s_load_dwordx4 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}}
-; GCN-NEXT:       s_load_dwordx2 s{{\[}}[[SRC2_LO:[0-9]+]]:[[SRC2_HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0xd|0x34}}
-
-; GFX789-NEXT:    s_movk_i32 s{{[0-9]+}}, 0xff
-; GFX789:         s_load_dword s{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}}, 0x0
-; GFX789-NEXT:    s_load_dword [[S1:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x0
-; GFX789-NEXT:    s_load_dword [[S2:s[0-9]+]], s{{\[}}[[SRC2_LO]]:[[SRC2_HI]]{{\]}}, 0x0
-; GFX789-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX789-NEXT:    s_and_b32 [[V1E1:s[0-9]+]], s{{[0-9]+}}, s{{[0-9]+}}
-; GFX789-NEXT:    s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
-; GFX789-NEXT:    s_bfe_u32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80008
-; GFX789-NEXT:    v_mov_b32_e32 [[V2E1:v[0-9]+]], s{{[0-9]+}}
-; GFX789-NEXT:    v_mov_b32_e32 [[SRC2:v[0-9]+]], s{{[0-9]+}}
-; GFX789-NEXT:    s_bfe_u32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80010
-; GFX789-NEXT:    v_mad_u32_u24 [[MAD1:v[0-9]+]], [[V1E1]], [[V2E1]], [[SRC2]]
-
-; GFX789-NEXT:    s_bfe_u32 [[V1E2:s[0-9]+]], s{{[0-9]+}}, 0x80008
-; GFX789-NEXT:    v_mov_b32_e32 [[V2E2:v[0-9]+]], s{{[0-9]+}}
-; GFX789-NEXT:    s_bfe_u32 [[V1E3:s[0-9]+]], s{{[0-9]+}}, 0x80010
-; GFX789-NEXT:    v_mad_u32_u24 [[MAD2:v[0-9]+]], [[V1E2]], [[V2E2]], [[MAD1]]
-
-; GFX789-NEXT:    v_mov_b32_e32 [[V2E3:v[0-9]+]], s{{[0-9]+}}
-; GFX789-NEXT:    s_lshr_b32 s{{[0-9]+}}, s{{[0-9]+}}, 24
-; GFX789-NEXT:    v_mad_u32_u24 [[MAD3:v[0-9]+]], [[V1E3]], [[V2E3]], [[MAD2]]
-
-; GFX789-NEXT:    s_lshr_b32 [[V1E4:s[0-9]+]], s{{[0-9]+}}, 24
-; GFX789-NEXT:    v_mov_b32_e32 [[V2E4:v[0-9]+]], s{{[0-9]+}}
-; GFX789-NEXT:    v_mad_u32_u24 [[RES:v[0-9]+]], [[V1E4]], [[V2E4]], [[MAD3]]
-; GFX789:         {{buffer|flat|global}}_store_dword
-; GFX789-NEXT:    s_endpgm
-
-; GCN-DL:         s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_load_dword [[SRC0:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x0
-; GCN-DL-NEXT:    s_load_dword [[S1:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x0
-; GCN-DL-NEXT:    s_load_dword [[S2:s[0-9]+]], s{{\[}}[[SRC2_LO]]:[[SRC2_HI]]{{\]}}, 0x0
-; GCN-DL-NEXT:    v_mov_b32_e32 v[[STLO:[0-9]+]], s[[SRC2_LO]]
-; GCN-DL-NEXT:    v_mov_b32_e32 v[[STHI:[0-9]+]], s[[SRC2_HI]]
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    v_mov_b32_e32 [[SRC1:v[0-9]+]], [[S1]]
-; GCN-DL-NEXT:    v_mov_b32_e32 [[SRC2:v[0-9]+]], [[S2]]
-; GCN-DL-NEXT:    v_dot4_u32_u8 [[DOT:v[0-9]+]], [[SRC0]], [[SRC1]], [[SRC2]]
-; GCN-DL-NEXT:    global_store_dword v{{\[}}[[STLO]]:[[STHI]]{{\]}}, [[DOT]], off
-; GCN-DL-NEXT:    s_endpgm
-
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=amdgcn -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX7 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX8 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9-NODL %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx906 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9-DL %s
 
 define amdgpu_kernel void @udot4_acc32(<4 x i8> addrspace(1)* %src1,
+; GFX7-LABEL: udot4_acc32:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_movk_i32 s8, 0xff
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_load_dword s6, s[0:1], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_and_b32 s7, s4, s8
+; GFX7-NEXT:    s_and_b32 s8, s5, s8
+; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80008
+; GFX7-NEXT:    v_mov_b32_e32 v0, s8
+; GFX7-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7-NEXT:    s_bfe_u32 s12, s5, 0x80010
+; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v0, v1
+; GFX7-NEXT:    s_bfe_u32 s9, s4, 0x80008
+; GFX7-NEXT:    v_mov_b32_e32 v1, s10
+; GFX7-NEXT:    s_bfe_u32 s11, s4, 0x80010
+; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s12
+; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v1, v0
+; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX7-NEXT:    v_mov_b32_e32 v1, s5
+; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
+; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: udot4_acc32:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_movk_i32 s2, 0xff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX8-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s6, s3, s2
+; GFX8-NEXT:    s_and_b32 s2, s4, s2
+; GFX8-NEXT:    s_bfe_u32 s8, s4, 0x80008
+; GFX8-NEXT:    v_mov_b32_e32 v0, s2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s5
+; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x80010
+; GFX8-NEXT:    v_mad_u32_u24 v0, s6, v0, v1
+; GFX8-NEXT:    s_bfe_u32 s7, s3, 0x80008
+; GFX8-NEXT:    v_mov_b32_e32 v1, s8
+; GFX8-NEXT:    s_bfe_u32 s9, s3, 0x80010
+; GFX8-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s10
+; GFX8-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX8-NEXT:    v_mad_u32_u24 v0, s9, v1, v0
+; GFX8-NEXT:    s_lshr_b32 s3, s3, 24
+; GFX8-NEXT:    v_mov_b32_e32 v1, s4
+; GFX8-NEXT:    v_mad_u32_u24 v2, s3, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: udot4_acc32:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_movk_i32 s2, 0xff
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_and_b32 s6, s3, s2
+; GFX9-NODL-NEXT:    s_and_b32 s2, s4, s2
+; GFX9-NODL-NEXT:    s_bfe_u32 s8, s4, 0x80008
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s5
+; GFX9-NODL-NEXT:    s_bfe_u32 s10, s4, 0x80010
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s6, v0, v1
+; GFX9-NODL-NEXT:    s_bfe_u32 s7, s3, 0x80008
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s8
+; GFX9-NODL-NEXT:    s_bfe_u32 s9, s3, 0x80010
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s10
+; GFX9-NODL-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s9, v1, v0
+; GFX9-NODL-NEXT:    s_lshr_b32 s3, s3, 24
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s4
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s3, v1, v0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: udot4_acc32:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s3
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-DL-NEXT:    v_dot4_u32_u8 v2, s2, v2, v3
+; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                        <4 x i8> addrspace(1)* %src2,
                                        i32 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
   %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
 
   %v1e0 = extractelement <4 x i8> %vec1, i64 0
   %cv1e0 = zext i8 %v1e0 to i32
entry:
   %mad3 = add i32 %mad2, %mul3
   %mad4 = add i32 %mad3, %mul4
 
   store i32 %mad4, i32 addrspace(1)* %dst, align 4
   ret void
 }
 
 define amdgpu_kernel void @udot4_acc16(<4 x i8> addrspace(1)* %src1,
-; GCN-LABEL: udot4_acc16:
-; GCN:       ; %bb.0: ; %entry
-; GCN-NEXT:       s_load_dwordx4 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}}
-; GCN-NEXT:       s_load_dwordx2 s{{\[}}[[SRC2_LO:[0-9]+]]:[[SRC2_HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0xd|0x34}}
-
-; GFX789:         {{buffer|flat|global}}_load_ushort [[SRC2:v[0-9]+]]
-; GFX789:         s_load_dword
-; GFX789:         s_waitcnt lgkmcnt(0)
-; GFX789:         s_and_b32
-; GFX789:         s_bfe_u32 [[V1E2:s[0-9]+]], s{{[0-9]+}}, 0x80008
-; GFX789:         s_bfe_u32
-; GFX789:         s_bfe_u32
-; GFX789-NEXT:    v_mov_b32_e32 [[V2E2:v[0-9]+]], s{{[0-9]+}}
-; GFX789-NEXT:    s_bfe_u32 [[V1E3:s[0-9]+]], s{{[0-9]+}}, 0x80010
-; GFX789-NEXT:    s_lshr_b32 [[V1E4:s[0-9]+]], s{{[0-9]+}}, 24
-; GFX789-NEXT:    v_mov_b32_e32 [[V2E3:v[0-9]+]]
-; GFX789-NEXT:    s_lshr_b32 [[V1E4:s[0-9]+]], s{{[0-9]+}}, 24
-; GFX789-NEXT:    s_waitcnt vmcnt(0)
-; GFX789-NEXT:    v_mad_u32_u24 [[MAD1:v[0-9]+]], {{s[0-9]+}}, {{v[0-9]+}}, [[SRC2]]
-; GFX789-NEXT:    v_mad_u32_u24 [[MAD2:v[0-9]+]], {{s[0-9]+}}, [[V2E2]], [[MAD1]]
-; GFX789-NEXT:    v_mad_u32_u24 [[MAD3:v[0-9]+]], [[V1E3]], [[V2E3]], [[MAD2]]
-; GFX789-NEXT:    v_mov_b32_e32 [[V2E4:v[0-9]+]], s{{[0-9]+}}
-; GFX789-NEXT:    v_mad_u32_u24 [[MAD4:v[0-9]+]], [[V1E4]], [[V2E4]], [[MAD3]]
-; GFX789-NEXT:    {{buffer|flat|global}}_store_short
-; GFX789-NEXT:    s_endpgm
-
-; GCN-DL:         s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_load_dword [[SRC0:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x0
-; GCN-DL-NEXT:    s_load_dword [[S1:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x0
-; GCN-DL-NEXT:    v_mov_b32_e32 v[[STLO:[0-9]+]], s[[SRC2_LO]]
-; GCN-DL-NEXT:    v_mov_b32_e32 v[[STHI:[0-9]+]], s[[SRC2_HI]]
-; GCN-DL-NEXT:    global_load_ushort [[SRC2:v[0-9]+]], v{{\[}}[[STLO]]:[[STHI]]{{\]}}, off
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    v_mov_b32_e32 [[SRC1:v[0-9]+]], [[S1]]
-; GCN-DL-NEXT:    s_waitcnt vmcnt(0)
-; GCN-DL-NEXT:    v_dot4_u32_u8 [[DOT:v[0-9]+]], [[SRC0]], [[SRC1]], [[SRC2]]
-; GCN-DL-NEXT:    global_store_short v{{\[}}[[STLO]]:[[STHI]]{{\]}}, [[DOT]], off
-; GCN-DL-NEXT:    s_endpgm
+; GFX7-LABEL: udot4_acc16:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_movk_i32 s8, 0xff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    buffer_load_ushort v0, off, s[0:3], 0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_and_b32 s7, s4, s8
+; GFX7-NEXT:    s_bfe_u32 s9, s4, 0x80008
+; GFX7-NEXT:    s_and_b32 s6, s5, s8
+; GFX7-NEXT:    s_bfe_u32 s8, s5, 0x80008
+; GFX7-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80010
+; GFX7-NEXT:    v_mov_b32_e32 v2, s8
+; GFX7-NEXT:    s_bfe_u32 s11, s4, 0x80010
+; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
+; GFX7-NEXT:    v_mov_b32_e32 v3, s10
+; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v2, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v3, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s5
+; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
+; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: udot4_acc16:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ushort v2, v[0:1]
+; GFX8-NEXT:    s_load_dword s1, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s3, s1, s0
+; GFX8-NEXT:    s_and_b32 s0, s2, s0
+; GFX8-NEXT:    s_bfe_u32 s5, s2, 0x80008
+; GFX8-NEXT:    v_mov_b32_e32 v3, s0
+; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x80010
+; GFX8-NEXT:    s_bfe_u32 s4, s1, 0x80008
+; GFX8-NEXT:    v_mov_b32_e32 v4, s5
+; GFX8-NEXT:    s_bfe_u32 s6, s1, 0x80010
+; GFX8-NEXT:    s_lshr_b32 s2, s2, 24
+; GFX8-NEXT:    v_mov_b32_e32 v5, s7
+; GFX8-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mad_u32_u24 v2, s3, v3, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s4, v4, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s1, v3, v2
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: udot4_acc16:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_load_ushort v2, v[0:1], off
+; GFX9-NODL-NEXT:    s_load_dword s1, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_movk_i32 s0, 0xff
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_and_b32 s3, s1, s0
+; GFX9-NODL-NEXT:    s_and_b32 s0, s2, s0
+; GFX9-NODL-NEXT:    s_bfe_u32 s5, s2, 0x80008
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s0
+; GFX9-NODL-NEXT:    s_bfe_u32 s7, s2, 0x80010
+; GFX9-NODL-NEXT:    s_bfe_u32 s4, s1, 0x80008
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s5
+; GFX9-NODL-NEXT:    s_bfe_u32 s6, s1, 0x80010
+; GFX9-NODL-NEXT:    s_lshr_b32 s2, s2, 24
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s7
+; GFX9-NODL-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s3, v3, v2
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s4, v4, v2
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s2
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s1, v3, v2
+; GFX9-NODL-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: udot4_acc16:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ushort v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_dot4_u32_u8 v2, s2, v3, v2
+; GFX9-DL-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                        <4 x i8> addrspace(1)* %src2,
                                        i16 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
   %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
 
   %v1e0 = extractelement <4 x i8> %vec1, i64 0
   %cv1e0 = zext i8 %v1e0 to i16
entry:
   %mad3 = add i16 %mad2, %mul3
   %mad4 = add i16 %mad3, %mul4
 
   store i16 %mad4, i16 addrspace(1)* %dst, align 2
   ret void
 }
 
 define amdgpu_kernel void @udot4_acc8(<4 x i8> addrspace(1)* %src1,
-; GCN-LABEL: udot4_acc8:
-; GCN:       ; %bb.0: ; %entry
-; GCN-NEXT:     s_load_dwordx4 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}}
-; GCN-NEXT:     s_load_dwordx2 s{{\[}}[[SRC2_LO:[0-9]+]]:[[SRC2_HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0xd|0x34}}
-; GFX789:       s_movk_i32 s{{[0-9]+}}, 0xff
-; GFX789:       s_waitcnt lgkmcnt(0)
-; GFX789:       {{buffer|flat|global}}_load_ubyte [[SRC2:v[0-9]+]]
-; GFX789:       s_load_dword s{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}}
-; GFX789:       s_waitcnt lgkmcnt(0)
-; GFX789:       s_bfe_u32 [[V1E2:s[0-9]+]], s{{[0-9]+}}, 0x80008
-; GFX789:       s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
-; GFX789:       s_bfe_u32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80008
-; GFX789:       v_mov_b32_e32 [[V2E1:v[0-9]+]]
-; GFX789:       s_bfe_u32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80010
-; GFX789-NEXT:  v_mov_b32_e32 [[V2E2:v[0-9]+]]
-; GFX789-NEXT:  s_bfe_u32 [[V1E3:s[0-9]+]], s{{[0-9]+}}, 0x80010
-; GFX789-NEXT:  s_lshr_b32 s{{[0-9]+}}, s{{[0-9]+}}, 24
-; GFX789-NEXT:  v_mov_b32_e32 [[V2E3:v[0-9]+]]
-; GFX789-NEXT:  s_lshr_b32 [[V1E4:s[0-9]+]], s{{[0-9]+}}, 24
-; GFX789-NEXT:  s_waitcnt vmcnt(0)
-; GFX789-NEXT:  v_mad_u32_u24 [[MAD1:v[0-9]+]], s{{[0-9]+}}, [[V2E1]], [[SRC2]]
-; GFX789-NEXT:  v_mad_u32_u24 [[MAD2:v[0-9]+]], [[V1E2]], [[V2E2]], [[MAD1]]
-; GFX789-NEXT:  v_mad_u32_u24 [[MAD3:v[0-9]+]], [[V1E3]], [[V2E3]], [[MAD2]]
-; GFX789-NEXT:  v_mov_b32_e32 [[V2E4:v[0-9]+]]
-; GFX789-NEXT:  v_mad_u32_u24 [[MAD4:v[0-9]+]], [[V1E4]], [[V2E4]], [[MAD3]]
-; GFX789-NEXT:  {{buffer|flat|global}}_store_byte
-; GFX789-NEXT:  s_endpgm
-
-; GCN-DL:       s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:  s_load_dword [[SRC0:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}
-; GCN-DL-NEXT:  s_load_dword [[S1:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}
-; GCN-DL-NEXT:  v_mov_b32_e32 v[[STLO:[0-9]+]], s[[SRC2_LO]]
-; GCN-DL-NEXT:  v_mov_b32_e32 v[[STHI:[0-9]+]], s[[SRC2_HI]]
-; GCN-DL-NEXT:  global_load_ubyte [[SRC2:v[0-9]+]], v{{\[}}[[STLO]]:[[STHI]]{{\]}}, off
-; GCN-DL-NEXT:  s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:  v_mov_b32_e32 [[SRC1:v[0-9]+]], [[S1]]
-; GCN-DL-NEXT:  s_waitcnt vmcnt(0)
-; GCN-DL-NEXT:  v_dot4_u32_u8 [[DOT:v[0-9]+]], [[SRC0]], [[SRC1]], [[SRC2]]
-; GCN-DL-NEXT:  global_store_byte v{{\[}}[[STLO]]:[[STHI]]{{\]}}, [[DOT]], off
-; GCN-DL-NEXT:  s_endpgm
+; GFX7-LABEL: udot4_acc8:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_movk_i32 s8, 0xff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_and_b32 s7, s4, s8
+; GFX7-NEXT:    s_bfe_u32 s9, s4, 0x80008
+; GFX7-NEXT:    s_and_b32 s6, s5, s8
+; GFX7-NEXT:    s_bfe_u32 s8, s5, 0x80008
+; GFX7-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80010
+; GFX7-NEXT:    v_mov_b32_e32 v2, s8
+; GFX7-NEXT:    s_bfe_u32 s11, s4, 0x80010
+; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
+; GFX7-NEXT:    v_mov_b32_e32 v3, s10
+; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v2, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v3, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s5
+; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
+; GFX7-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: udot4_acc8:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_movk_i32 s2, 0xff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ubyte v2, v[0:1]
+; GFX8-NEXT:    s_load_dword s0, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_bfe_u32 s4, s0, 0x80008
+; GFX8-NEXT:    s_and_b32 s3, s1, s2
+; GFX8-NEXT:    s_bfe_u32 s5, s1, 0x80008
+; GFX8-NEXT:    s_and_b32 s2, s0, s2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s3
+; GFX8-NEXT:    s_bfe_u32 s6, s1, 0x80010
+; GFX8-NEXT:    v_mov_b32_e32 v4, s5
+; GFX8-NEXT:    s_bfe_u32 s7, s0, 0x80010
+; GFX8-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX8-NEXT:    v_mov_b32_e32 v5, s6
+; GFX8-NEXT:    s_lshr_b32 s0, s0, 24
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s4, v4, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX8-NEXT:    flat_store_byte v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: udot4_acc8:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_movk_i32 s2, 0xff
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-NODL-NEXT:    s_load_dword s0, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_bfe_u32 s4, s0, 0x80008
+; GFX9-NODL-NEXT:    s_and_b32 s3, s1, s2
+; GFX9-NODL-NEXT:    s_bfe_u32 s5, s1, 0x80008
+; GFX9-NODL-NEXT:    s_and_b32 s2, s0, s2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-NODL-NEXT:    s_bfe_u32 s6, s1, 0x80010
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s5
+; GFX9-NODL-NEXT:    s_bfe_u32 s7, s0, 0x80010
+; GFX9-NODL-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s6
+; GFX9-NODL-NEXT:    s_lshr_b32 s0, s0, 24
+; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s4, v4, v2
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX9-NODL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: udot4_acc8:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_dot4_u32_u8 v2, s2, v3, v2
+; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                       <4 x i8> addrspace(1)* %src2,
                                       i8 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
   %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
 
   %v1e0 = extractelement <4 x i8> %vec1, i64 0
   %v2e0 = extractelement <4 x i8> %vec2, i64 0
entry:
   %mad4 = add i8 %mad3, %mul4
 
   store i8 %mad4, i8 addrspace(1)* %dst, align 2
   ret void
 }
 
 ; TODO: Generate udot4?
 define amdgpu_kernel void @udot2_8(<4 x i8> addrspace(1)* %src1,
-; GCN-LABEL: udot2_8:
-; GCN-NEXT:   ; %bb.0: ; %entry
-; GCN-NEXT:    s_load_dwordx4 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}}
-; GCN-NEXT:    s_load_dwordx2 s{{\[}}[[SRC2_LO:[0-9]+]]:[[SRC2_HI:[0-9]+]]{{\]}}
-; GCN:         s_movk_i32 [[FF:s[0-9]+]], 0xff
-; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN:         s_load_dword [[V1:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x0
-; GCN:         s_load_dword [[V2:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x0
-; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN:         s_and_b32 [[V1E1:s[0-9]+]], [[V1]], [[FF]]
-; GCN:         s_bfe_u32 [[VE2:s[0-9]+]], {{s[0-9]+}}, 0x80008
-; GCN:         s_bfe_u32 [[V1E2:s[0-9]+]], {{s[0-9]+}}, 0x80008
-; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    v_mad_u32_u24 [[MAD1:v[0-9]+]], [[V1E1]]
-; GCN-NEXT:    v_mov_b32_e32 [[V2E2:v[0-9]+]]
-; GCN-NEXT:    v_mad_u32_u24 [[MAD2:v[0-9]+]], {{s[0-9]+}}, [[V2E2]], [[MAD1]]
-; GCN-NEXT:    {{buffer|flat|global}}_store_byte
-; GCN-NEXT:    s_endpgm
+; GFX7-LABEL: udot2_8:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_movk_i32 s8, 0xff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_and_b32 s7, s4, s8
+; GFX7-NEXT:    s_bfe_u32 s4, s4, 0x80008
+; GFX7-NEXT:    s_and_b32 s6, s5, s8
+; GFX7-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7-NEXT:    s_bfe_u32 s5, s5, 0x80008
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s5
+; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
+; GFX7-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: udot2_8:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_movk_i32 s2, 0xff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ubyte v2, v[0:1]
+; GFX8-NEXT:    s_load_dword s0, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s3, s1, s2
+; GFX8-NEXT:    s_and_b32 s2, s0, s2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s3
+; GFX8-NEXT:    s_bfe_u32 s1, s1, 0x80008
+; GFX8-NEXT:    s_bfe_u32 s0, s0, 0x80008
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX8-NEXT:    flat_store_byte v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: udot2_8:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_movk_i32 s2, 0xff
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-NODL-NEXT:    s_load_dword s0, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_and_b32 s3, s1, s2
+; GFX9-NODL-NEXT:    s_and_b32 s2, s0, s2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-NODL-NEXT:    s_bfe_u32 s1, s1, 0x80008
+; GFX9-NODL-NEXT:    s_bfe_u32 s0, s0, 0x80008
+; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX9-NODL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: udot2_8:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_movk_i32 s2, 0xff
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-DL-NEXT:    s_load_dword s0, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_and_b32 s3, s1, s2
+; GFX9-DL-NEXT:    s_and_b32 s2, s0, s2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-DL-NEXT:    s_bfe_u32 s1, s1, 0x80008
+; GFX9-DL-NEXT:    s_bfe_u32 s0, s0, 0x80008
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                    <4 x i8> addrspace(1)* %src2,
                                    i8 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
   %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
 
   %v1e0 = extractelement <4 x i8> %vec1, i64 0
   %v2e0 = extractelement <4 x i8> %vec2, i64 0
   %mul1 = mul nuw nsw i8 %v1e0, %v2e0
 
   %v1e1 = extractelement <4 x i8> %vec1, i64 1
   %v2e1 = extractelement <4 x i8> %vec2, i64 1
   %mul2 = mul nuw nsw i8 %v1e1, %v2e1
 
   %acc = load i8, i8 addrspace(1)* %dst, align 2
   %mad1 = add i8 %mul1, %acc
   %mad2 = add i8 %mad1, %mul2
   store i8 %mad2, i8 addrspace(1)* %dst, align 2
   ret void
 }
 
 define amdgpu_kernel void @udot4_CommutationInsideMAD(<4 x i8> addrspace(1)* %src1,
-; GCN-LABEL: udot4_CommutationInsideMAD:
-; GCN:       ; %bb.0: ; %entry
-; GCN-NEXT:    s_load_dwordx4 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}}
-; GCN-NEXT:    s_load_dwordx2 s{{\[}}[[SRC2_LO:[0-9]+]]:[[SRC2_HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0xd|0x34}}
-; GFX789:      s_waitcnt lgkmcnt(0)
-; GFX789:      {{buffer|flat|global}}_load_ubyte [[SRC2:v[0-9]+]]
-; GFX789:      s_load_dword s{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}}
-; GFX789:      s_waitcnt lgkmcnt(0)
-; GFX789:      s_bfe_u32
-; GFX789:      s_bfe_u32
-; GFX789-NEXT: s_bfe_u32 [[V1E2:s[0-9]+]], s{{[0-9]+}}, 0x80008
-; GFX789-NEXT: v_mov_b32_e32 [[V2E2:v[0-9]+]]
-; GFX789-NEXT: s_bfe_u32 [[V1E3:s[0-9]+]], s{{[0-9]+}}, 0x80010
-; GFX789-NEXT: s_lshr_b32 [[VE4:s[0-9]+]], s{{[0-9]+}}, 24
-; GFX789-NEXT: v_mov_b32_e32 [[V2E3:v[0-9]+]]
-; GFX789-NEXT: s_lshr_b32 [[V1E4:s[0-9]+]], s{{[0-9]+}}, 24
-; GFX789-NEXT: s_waitcnt vmcnt(0)
-
-; GFX789-NEXT: v_mad_u32_u24 [[MAD1:v[0-9]+]],  s{{[0-9]+}},  v{{[0-9]+}}, [[SRC2]]
-; GFX789-NEXT: v_mad_u32_u24 [[MAD2:v[0-9]+]], [[V1E2]],  [[V2E2]], [[MAD1]]
-; GFX789-NEXT: v_mad_u32_u24 [[MAD3:v[0-9]+]], [[V1E3]], [[V2E3]], [[MAD2]]
-; GFX789-NEXT: v_mov_b32_e32 [[V2E4:v[0-9]+]], [[VE4]]
-; GFX789-NEXT: v_mad_u32_u24 [[MAD4:v[0-9]+]], [[V1E4]], [[V2E4]], [[MAD3]]
-; GFX789-NEXT: {{buffer|flat|global}}_store_byte
-; GFX789-NEXT: s_endpgm
-
-; GCN-DL:      s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT: s_load_dword [[S1:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}
-; GCN-DL-NEXT: s_load_dword [[SRC0:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}
-; GCN-DL-NEXT: v_mov_b32_e32 v[[STLO:[0-9]+]], s[[SRC2_LO]]
-; GCN-DL-NEXT: v_mov_b32_e32 v[[STHI:[0-9]+]], s[[SRC2_HI]]
-; GCN-DL-NEXT: global_load_ubyte [[SRC2:v[0-9]+]], v{{\[}}[[STLO]]:[[STHI]]{{\]}}, off
-; GCN-DL-NEXT: s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT: v_mov_b32_e32 [[SRC1:v[0-9]+]], [[S1]]
-; GCN-DL-NEXT: s_waitcnt vmcnt(0)
-; GCN-DL-NEXT: v_dot4_u32_u8 [[DOT:v[0-9]+]], [[SRC0]], [[SRC1]], [[SRC2]]
-; GCN-DL-NEXT: global_store_byte v{{\[}}[[STLO]]:[[STHI]]{{\]}}, [[DOT]], off
-; GCN-DL-NEXT: s_endpgm
+; GFX7-LABEL: udot4_CommutationInsideMAD:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_movk_i32 s8, 0xff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_and_b32 s6, s4, s8
+; GFX7-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7-NEXT:    s_and_b32 s7, s5, s8
+; GFX7-NEXT:    s_bfe_u32 s8, s4, 0x80008
+; GFX7-NEXT:    s_bfe_u32 s10, s4, 0x80010
+; GFX7-NEXT:    s_bfe_u32 s9, s5, 0x80008
+; GFX7-NEXT:    v_mov_b32_e32 v2, s8
+; GFX7-NEXT:    s_bfe_u32 s11, s5, 0x80010
+; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX7-NEXT:    v_mov_b32_e32 v3, s10
+; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v2, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v3, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s4
+; GFX7-NEXT:    v_mad_u32_u24 v0, s5, v1, v0
+; GFX7-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: udot4_CommutationInsideMAD:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ubyte v2, v[0:1]
+; GFX8-NEXT:    s_load_dword s1, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s3, s1, s0
+; GFX8-NEXT:    s_bfe_u32 s5, s1, 0x80008
+; GFX8-NEXT:    s_and_b32 s0, s2, s0
+; GFX8-NEXT:    v_mov_b32_e32 v3, s3
+; GFX8-NEXT:    s_bfe_u32 s6, s1, 0x80010
+; GFX8-NEXT:    s_bfe_u32 s4, s2, 0x80008
+; GFX8-NEXT:    v_mov_b32_e32 v4, s5
+; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x80010
+; GFX8-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX8-NEXT:    v_mov_b32_e32 v5, s6
+; GFX8-NEXT:    s_lshr_b32 s2, s2, 24
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s4, v4, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX8-NEXT:    flat_store_byte v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: udot4_CommutationInsideMAD:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-NODL-NEXT:    s_load_dword s1, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_movk_i32 s0, 0xff
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_and_b32 s3, s1, s0
+; GFX9-NODL-NEXT:    s_bfe_u32 s5, s1, 0x80008
+; GFX9-NODL-NEXT:    s_and_b32 s0, s2, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-NODL-NEXT:    s_bfe_u32 s6, s1, 0x80010
+; GFX9-NODL-NEXT:    s_bfe_u32 s4, s2, 0x80008
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s5
+; GFX9-NODL-NEXT:    s_bfe_u32 s7, s2, 0x80010
+; GFX9-NODL-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s6
+; GFX9-NODL-NEXT:    s_lshr_b32 s2, s2, 24
+; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s4, v4, v2
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX9-NODL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: udot4_CommutationInsideMAD:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s2
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_dot4_u32_u8 v2, s3, v3, v2
+; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                                       <4 x i8> addrspace(1)* %src2,
                                                       i8 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
   %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
 
   %v1e0 = extractelement <4 x i8> %vec1, i64 0
   %v2e0 = extractelement <4 x i8> %vec2, i64 0
   %mul1 = mul nuw nsw i8 %v2e0, %v1e0
 
entry:
   %mad4 = add i8 %mul4, %mad3
 
   store i8 %mad4, i8 addrspace(1)* %dst, align 2
   ret void
 }
 
 ; TODO: Support commutation accross the adds.
 define amdgpu_kernel void @udot4_CommutationAccrossMADs(<4 x i8> addrspace(1)* %src1,
-; GCN-LABEL: udot4_CommutationAccrossMADs:
-; GCN:       ; %bb.0: ; %entry
-; GCN-NEXT:    s_load_dwordx4 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}}
-; GCN-NEXT:    s_load_dwordx2 s{{\[}}[[SRC2_LO:[0-9]+]]:[[SRC2_HI:[0-9]+]]{{\]}}
-; GCN:         s_waitcnt lgkmcnt(0)
-; GCN:         {{buffer|flat|global}}_load_ubyte [[SRC2:v[0-9]+]]
-; GCN:         s_load_dword [[V2:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x0
-; GCN:         s_waitcnt lgkmcnt(0)
-; GCN:         s_bfe_u32 {{s[0-9]+}}, {{s[0-9]+}}, 0x80008
-; GCN:         s_bfe_u32 {{s[0-9]+}}, {{s[0-9]+}}, 0x80008
-; GCN:         v_mov_b32_e32 [[V2E1:v[0-9]+]]
-; GCN:         s_bfe_u32 [[V1E3:s[0-9]+]], {{s[0-9]+}}, 0x80010
-; GCN:         s_lshr_b32 [[VE4:s[0-9]+]], {{s[0-9]+}}, 24
-; GCN-NEXT:    v_mov_b32_e32 [[V2E3:v[0-9]+]], {{s[0-9]+}}
-; GCN-NEXT:    s_lshr_b32 [[V1E4:s[0-9]+]], {{s[0-9]+}}, 24
-; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    v_mad_u32_u24 [[MAD1:v[0-9]+]], {{s[0-9]+}}, [[V2E1]], [[SRC2]]
-; GCN-NEXT:    v_mad_u32_u24 [[MAD2:v[0-9]+]], {{s[0-9]+}}, {{v[0-9]+}}, [[MAD1]]
-; GCN-NEXT:    v_mad_u32_u24 [[MAD3:v[0-9]+]], {{s[0-9]+}}, {{v[0-9]+}}, [[MAD2]]
-; GCN-NEXT:    v_mov_b32_e32 [[V2E4:v[0-9]+]], [[VE4]]
-; GCN-NEXT:    v_mad_u32_u24 [[MAD4:v[0-9]+]], [[V1E4]], [[V2E4]], [[MAD3]]
-; GCN-NEXT:    {{buffer|flat|global}}_store_byte
-; GCN-NEXT:    s_endpgm
+; GFX7-LABEL: udot4_CommutationAccrossMADs:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_movk_i32 s8, 0xff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_and_b32 s6, s4, s8
+; GFX7-NEXT:    s_bfe_u32 s10, s4, 0x80010
+; GFX7-NEXT:    s_and_b32 s7, s5, s8
+; GFX7-NEXT:    s_bfe_u32 s8, s4, 0x80008
+; GFX7-NEXT:    s_bfe_u32 s9, s5, 0x80008
+; GFX7-NEXT:    v_mov_b32_e32 v1, s8
+; GFX7-NEXT:    v_mov_b32_e32 v2, s6
+; GFX7-NEXT:    s_bfe_u32 s11, s5, 0x80010
+; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX7-NEXT:    v_mov_b32_e32 v3, s10
+; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v1, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v2, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v3, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s4
+; GFX7-NEXT:    v_mad_u32_u24 v0, s5, v1, v0
+; GFX7-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: udot4_CommutationAccrossMADs:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ubyte v2, v[0:1]
+; GFX8-NEXT:    s_load_dword s1, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_bfe_u32 s4, s1, 0x80008
+; GFX8-NEXT:    s_and_b32 s3, s2, s0
+; GFX8-NEXT:    s_and_b32 s0, s1, s0
+; GFX8-NEXT:    s_bfe_u32 s5, s2, 0x80008
+; GFX8-NEXT:    v_mov_b32_e32 v3, s4
+; GFX8-NEXT:    s_bfe_u32 s6, s1, 0x80010
+; GFX8-NEXT:    v_mov_b32_e32 v4, s0
+; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x80010
+; GFX8-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX8-NEXT:    v_mov_b32_e32 v5, s6
+; GFX8-NEXT:    s_lshr_b32 s2, s2, 24
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v3, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s3, v4, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX8-NEXT:    flat_store_byte v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: udot4_CommutationAccrossMADs:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-NODL-NEXT:    s_load_dword s1, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_movk_i32 s0, 0xff
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_bfe_u32 s4, s1, 0x80008
+; GFX9-NODL-NEXT:    s_and_b32 s3, s2, s0
+; GFX9-NODL-NEXT:    s_and_b32 s0, s1, s0
+; GFX9-NODL-NEXT:    s_bfe_u32 s5, s2, 0x80008
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-NODL-NEXT:    s_bfe_u32 s6, s1, 0x80010
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s0
+; GFX9-NODL-NEXT:    s_bfe_u32 s7, s2, 0x80010
+; GFX9-NODL-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s6
+; GFX9-NODL-NEXT:    s_lshr_b32 s2, s2, 24
+; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s5, v3, v2
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s3, v4, v2
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX9-NODL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: udot4_CommutationAccrossMADs:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-DL-NEXT:    s_load_dword s1, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_movk_i32 s0, 0xff
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_bfe_u32 s4, s1, 0x80008
+; GFX9-DL-NEXT:    s_and_b32 s3, s2, s0
+; GFX9-DL-NEXT:    s_and_b32 s0, s1, s0
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s2, 0x80008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s1, 0x80010
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s0
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x80010
+; GFX9-DL-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s6
+; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 24
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s5, v3, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s3, v4, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                                         <4 x i8> addrspace(1)* %src2,
                                                         i8 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
   %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
 
   %v1e0 = extractelement <4 x i8> %vec1, i64 0
   %v2e0 = extractelement <4 x i8> %vec2, i64 0
entry:
   %mad3 = add i8 %mad2, %mul3
   %mad4 = add i8 %mad3, %mul4
 
   store i8 %mad4, i8 addrspace(1)* %dst, align 2
   ret void
 }
 
 define amdgpu_kernel void @udot4_multiuse_mul1(<4 x i8> addrspace(1)* %src1,
-; GCN-LABEL: udot4_multiuse_mul1:
-; GCN:        ; %bb.0: ; %entry
-; GCN-NEXT:    s_load_dwordx4 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}}
-; GCN-NEXT:    s_load_dwordx2 s{{\[}}[[SRC2_LO:[0-9]+]]:[[SRC2_HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0xd|0x34}}
-; GCN-NEXT:    s_movk_i32 [[FF:s[0-9]+]], 0xff
-; GCN:         s_load_dword [[S0:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x0
-; GCN-NEXT:    s_load_dword [[S1:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x0
-; GCN-NEXT:    s_load_dword [[S2:s[0-9]+]], s{{\[}}[[SRC2_LO]]:[[SRC2_HI]]{{\]}}, 0x0
-; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_and_b32 [[V1E1:s[0-9]+]], [[S0]], [[FF]]
-; GCN-NEXT:    s_and_b32 [[SV2E1:s[0-9]+]], [[S1]], [[FF]]
-; GCN-NEXT:    s_bfe_u32 [[SV2E2:s[0-9]+]], [[S1]], 0x80008
-; GCN-NEXT:    v_mov_b32_e32 [[V2E1:v[0-9]+]], [[SV2E1]]
-; GCN-NEXT:    v_mov_b32_e32 [[SRC2:v[0-9]+]], [[S2]]
-; GCN-NEXT:    s_bfe_u32 [[V1E2:s[0-9]+]], [[S0]], 0x80008
-; GCN-NEXT:    v_mad_u32_u24 [[MAD1:v[0-9]+]], [[V1E1]], [[V2E1]], [[SRC2]]
-; GCN-NEXT:    v_mov_b32_e32 [[V2E2:v[0-9]+]], [[SV2E2]]
-; GCN-NEXT:    s_bfe_u32 [[VE4:s[0-9]+]], [[S1]], 0x80010
-; GCN-NEXT:    v_mad_u32_u24 [[MAD2:v[0-9]+]], [[V1E2]], [[V2E2]], [[MAD1]]
-; GCN-NEXT:    s_bfe_u32 [[V1E3:s[0-9]+]], [[S0]], 0x80010
-; GCN-NEXT:    v_mad_u32_u24 [[MAD3:v[0-9]+]], [[V1E1]], [[V2E1]], [[MAD2]]
-; GCN-NEXT:    v_mov_b32_e32 [[V2E3:v[0-9]+]], [[VE4]]
-; GCN-NEXT:    s_lshr_b32 [[VE4:s[0-9]+]], [[S1]], 24
-; GCN-NEXT:    v_mad_u32_u24 [[MAD4:v[0-9]+]], [[V1E3]], [[V2E3]], [[MAD3]]
-; GCN-NEXT:    s_lshr_b32 [[V1E4:s[0-9]+]], [[S0]], 24
-; GCN-NEXT:    v_mov_b32_e32 [[V2E4:v[0-9]+]]
-; GCN-NEXT:    v_mad_u32_u24 [[MAD5:v[0-9]+]], [[V1E4]], [[V2E4]], [[MAD4]]
-; GCN:         {{buffer|flat|global}}_store_dword
-; GCN-NEXT:    s_endpgm
-
-
+; GFX7-LABEL: udot4_multiuse_mul1:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_movk_i32 s8, 0xff
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_load_dword s6, s[0:1], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_and_b32 s7, s4, s8
+; GFX7-NEXT:    s_and_b32 s8, s5, s8
+; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80008
+; GFX7-NEXT:    v_mov_b32_e32 v0, s8
+; GFX7-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7-NEXT:    s_bfe_u32 s9, s4, 0x80008
+; GFX7-NEXT:    v_mad_u32_u24 v1, s7, v0, v1
+; GFX7-NEXT:    v_mov_b32_e32 v2, s10
+; GFX7-NEXT:    s_bfe_u32 s12, s5, 0x80010
+; GFX7-NEXT:    v_mad_u32_u24 v1, s9, v2, v1
+; GFX7-NEXT:    s_bfe_u32 s11, s4, 0x80010
+; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v0, v1
+; GFX7-NEXT:    v_mov_b32_e32 v1, s12
+; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v1, v0
+; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX7-NEXT:    v_mov_b32_e32 v1, s5
+; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
+; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: udot4_multiuse_mul1:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_movk_i32 s2, 0xff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX8-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s6, s3, s2
+; GFX8-NEXT:    s_and_b32 s2, s4, s2
+; GFX8-NEXT:    s_bfe_u32 s8, s4, 0x80008
+; GFX8-NEXT:    v_mov_b32_e32 v0, s2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s5
+; GFX8-NEXT:    s_bfe_u32 s7, s3, 0x80008
+; GFX8-NEXT:    v_mad_u32_u24 v1, s6, v0, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, s8
+; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x80010
+; GFX8-NEXT:    v_mad_u32_u24 v1, s7, v2, v1
+; GFX8-NEXT:    s_bfe_u32 s9, s3, 0x80010
+; GFX8-NEXT:    v_mad_u32_u24 v0, s6, v0, v1
+; GFX8-NEXT:    v_mov_b32_e32 v1, s10
+; GFX8-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX8-NEXT:    v_mad_u32_u24 v0, s9, v1, v0
+; GFX8-NEXT:    s_lshr_b32 s3, s3, 24
+; GFX8-NEXT:    v_mov_b32_e32 v1, s4
+; GFX8-NEXT:    v_mad_u32_u24 v2, s3, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: udot4_multiuse_mul1:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_movk_i32 s2, 0xff
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_and_b32 s6, s3, s2
+; GFX9-NODL-NEXT:    s_and_b32 s2, s4, s2
+; GFX9-NODL-NEXT:    s_bfe_u32 s8, s4, 0x80008
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s5
+; GFX9-NODL-NEXT:    s_bfe_u32 s7, s3, 0x80008
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v1, s6, v0, v1
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v2, s8
+; GFX9-NODL-NEXT:    s_bfe_u32 s10, s4, 0x80010
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v1, s7, v2, v1
+; GFX9-NODL-NEXT:    s_bfe_u32 s9, s3, 0x80010
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s6, v0, v1
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s10
+; GFX9-NODL-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s9, v1, v0
+; GFX9-NODL-NEXT:    s_lshr_b32 s3, s3, 24
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s4
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s3, v1, v0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: udot4_multiuse_mul1:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_movk_i32 s2, 0xff
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_and_b32 s6, s3, s2
+; GFX9-DL-NEXT:    s_and_b32 s2, s4, s2
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s4, 0x80008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s5
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s3, 0x80008
+; GFX9-DL-NEXT:    v_mad_u32_u24 v1, s6, v0, v1
+; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s8
+; GFX9-DL-NEXT:    s_bfe_u32 s10, s4, 0x80010
+; GFX9-DL-NEXT:    v_mad_u32_u24 v1, s7, v2, v1
+; GFX9-DL-NEXT:    s_bfe_u32 s9, s3, 0x80010
+; GFX9-DL-NEXT:    v_mad_u32_u24 v0, s6, v0, v1
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s10
+; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX9-DL-NEXT:    v_mad_u32_u24 v0, s9, v1, v0
+; GFX9-DL-NEXT:    s_lshr_b32 s3, s3, 24
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s4
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s3, v1, v0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                                <4 x i8> addrspace(1)* %src2,
                                                i32 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
   %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
 
   %v1e0 = extractelement <4 x i8> %vec1, i64 0
   %cv1e0 = zext i8 %v1e0 to i32
entry:
   %add3 = add i32 %add2, %mul3
   %add4 = add i32 %add3, %mul4
 
   store i32 %add4, i32 addrspace(1)* %dst, align 4
   ret void
 }
 
 define amdgpu_kernel void @udot4_multiuse_add1(<4 x i8> addrspace(1)* %src1,
-; GCN-LABEL: udot4_multiuse_add1:
-; GCN:       ; %bb.0: ; %entry
-; GCN-NEXT:    s_load_dwordx4 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}}
-; GCN-NEXT:    s_load_dwordx2 s{{\[}}[[SRC2_LO:[0-9]+]]:[[SRC2_HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0xd|0x34}}
-; GCN-NEXT:    s_movk_i32 [[FF:s[0-9]+]], 0xff
-; GCN:         s_load_dword [[S0:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x0
-; GCN-NEXT:    s_load_dword [[S1:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x0
-; GCN-NEXT:    s_load_dword [[S2:s[0-9]+]], s{{\[}}[[SRC2_LO]]:[[SRC2_HI]]{{\]}}, 0x0
-; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_and_b32 [[V1E2:s[0-9]+]], [[S0]], [[FF]]
-; GCN-NEXT:    s_bfe_u32 [[SV2E1:s[0-9]+]], [[S1]], 0x80008
-; GCN-NEXT:    s_and_b32 [[SV2E2:s[0-9]+]], [[S1]], [[FF]]
-; GCN-NEXT:    s_bfe_u32 [[V1E1:s[0-9]+]], [[S0]], 0x80008
-; GCN-NEXT:    v_mov_b32_e32 [[V2E1:v[0-9]+]], [[SV2E1]]
-; GCN-NEXT:    v_mov_b32_e32 [[SRC2:v[0-9]+]], [[S2]]
-; GCN-NEXT:    v_mad_u32_u24 [[MAD1:v[0-9]+]], [[V1E1]], [[V2E1]], [[SRC2]]
-; GCN-NEXT:    s_bfe_u32 [[SV2E3:s[0-9]+]], [[S1]], 0x80010
-; GCN-NEXT:    v_mov_b32_e32 [[V2E2:v[0-9]+]], [[SV2E2]]
-; GCN-NEXT:    s_bfe_u32 [[V1E3:s[0-9]+]], [[S0]], 0x80010
-; GCN-NEXT:    v_add_{{i|u}}32_e32 [[ADD1:v[0-9]+]]
-; GCN-NEXT:    v_mad_u32_u24 [[MAD2:v[0-9]+]], [[V1E2]], [[V2E2]], [[MAD1]]
-; GCN-NEXT:    v_mov_b32_e32 [[V2E3:v[0-9]+]], [[SV2E3]]
-; GCN-NEXT:    s_lshr_b32 [[SV2E4:s[0-9]+]], [[S1]], 24
-; GCN-NEXT:    v_mad_u32_u24 [[MAD3:v[0-9]+]], [[V1E3]], [[V2E3]], [[MAD2]]
-; GCN-NEXT:    s_lshr_b32 [[V1E4:s[0-9]+]], [[S0]], 24
-; GCN-NEXT:    v_mov_b32_e32 [[V2E4:v[0-9]+]], [[SV2E4]]
-; GCN-NEXT:    v_mad_u32_u24 [[MAD4:v[0-9]+]], [[V1E4]], [[V2E4]], [[MAD3]]
-; GCN-NEXT:    v_add_{{i|u}}32_e32 [[RES:v[0-9]+]]
-; GCN:         {{buffer|flat|global}}_store_dword
-; GCN-NEXT:    s_endpgm
+; GFX7-LABEL: udot4_multiuse_add1:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_movk_i32 s8, 0xff
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_load_dword s6, s[0:1], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_and_b32 s7, s4, s8
+; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80008
+; GFX7-NEXT:    s_and_b32 s8, s5, s8
+; GFX7-NEXT:    s_bfe_u32 s9, s4, 0x80008
+; GFX7-NEXT:    v_mov_b32_e32 v0, s10
+; GFX7-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v0, v1
+; GFX7-NEXT:    s_bfe_u32 s12, s5, 0x80010
+; GFX7-NEXT:    v_mov_b32_e32 v2, s8
+; GFX7-NEXT:    s_bfe_u32 s11, s4, 0x80010
+; GFX7-NEXT:    v_add_i32_e32 v1, vcc, s6, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v2, v0
+; GFX7-NEXT:    v_mov_b32_e32 v2, s12
+; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v2, v0
+; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX7-NEXT:    v_mov_b32_e32 v2, s5
+; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v2, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: udot4_multiuse_add1:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_movk_i32 s2, 0xff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX8-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s6, s3, s2
+; GFX8-NEXT:    s_bfe_u32 s8, s4, 0x80008
+; GFX8-NEXT:    s_and_b32 s2, s4, s2
+; GFX8-NEXT:    s_bfe_u32 s7, s3, 0x80008
+; GFX8-NEXT:    v_mov_b32_e32 v0, s8
+; GFX8-NEXT:    v_mov_b32_e32 v1, s5
+; GFX8-NEXT:    v_mad_u32_u24 v0, s7, v0, v1
+; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x80010
+; GFX8-NEXT:    v_mov_b32_e32 v2, s2
+; GFX8-NEXT:    s_bfe_u32 s9, s3, 0x80010
+; GFX8-NEXT:    v_add_u32_e32 v1, vcc, s5, v0
+; GFX8-NEXT:    v_mad_u32_u24 v0, s6, v2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v2, s10
+; GFX8-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX8-NEXT:    v_mad_u32_u24 v0, s9, v2, v0
+; GFX8-NEXT:    s_lshr_b32 s3, s3, 24
+; GFX8-NEXT:    v_mov_b32_e32 v2, s4
+; GFX8-NEXT:    v_mad_u32_u24 v0, s3, v2, v0
+; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v0, v1
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: udot4_multiuse_add1:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_movk_i32 s2, 0xff
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_and_b32 s6, s3, s2
+; GFX9-NODL-NEXT:    s_bfe_u32 s8, s4, 0x80008
+; GFX9-NODL-NEXT:    s_and_b32 s2, s4, s2
+; GFX9-NODL-NEXT:    s_bfe_u32 s7, s3, 0x80008
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s8
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s5
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s7, v0, v1
+; GFX9-NODL-NEXT:    s_bfe_u32 s10, s4, 0x80010
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-NODL-NEXT:    s_bfe_u32 s9, s3, 0x80010
+; GFX9-NODL-NEXT:    v_add_u32_e32 v1, s5, v0
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s6, v2, v0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v2, s10
+; GFX9-NODL-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s9, v2, v0
+; GFX9-NODL-NEXT:    s_lshr_b32 s3, s3, 24
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v2, s4
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s3, v2, v0
+; GFX9-NODL-NEXT:    v_add_u32_e32 v2, v0, v1
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: udot4_multiuse_add1:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_movk_i32 s2, 0xff
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_and_b32 s6, s3, s2
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s4, 0x80008
+; GFX9-DL-NEXT:    s_and_b32 s2, s4, s2
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s3, 0x80008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s8
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s5
+; GFX9-DL-NEXT:    v_mad_u32_u24 v0, s7, v0, v1
+; GFX9-DL-NEXT:    s_bfe_u32 s10, s4, 0x80010
+; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-DL-NEXT:    s_bfe_u32 s9, s3, 0x80010
+; GFX9-DL-NEXT:    v_add_u32_e32 v1, s5, v0
+; GFX9-DL-NEXT:    v_mad_u32_u24 v0, s6, v2, v0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s10
+; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX9-DL-NEXT:    v_mad_u32_u24 v0, s9, v2, v0
+; GFX9-DL-NEXT:    s_lshr_b32 s3, s3, 24
+; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s4
+; GFX9-DL-NEXT:    v_mad_u32_u24 v0, s3, v2, v0
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v0, v1
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                                <4 x i8> addrspace(1)* %src2,
                                                i32 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
   %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
 
   %v1e0 = extractelement <4 x i8> %vec1, i64 0
   %cv1e0 = zext i8 %v1e0 to i32
   %v2e0 = extractelement <4 x i8> %vec2, i64 0
   %cv2e0 = zext i8 %v2e0 to i32
entry:
   %add3 = add i32 %add2, %mul3
   %add4 = add i32 %add3, %mul4
   %res = add i32 %add4, %add
   store i32 %res, i32 addrspace(1)* %dst, align 4
   ret void
 }
 
 define amdgpu_kernel void @notdot4_mixedtypes(<4 x i8> addrspace(1)* %src1,
-; GCN-LABEL: notdot4_mixedtypes:
-; GCN:       ; %bb.0: ; %entry
-; GCN-NEXT:    s_load_dwordx4 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}}
-; GCN-NEXT:    s_load_dwordx2 s{{\[}}[[SRC2_LO:[0-9]+]]:[[SRC2_HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0xd|0x34}}
-; GCN:         s_mov_b32 [[FFFF:s[0-9]+]], 0xffff
-; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN:         {{buffer|flat|global}}_load_ushort [[SRC2:v[0-9]+]]
-; GCN:         s_load_dword [[S1:s[0-9]+]], s[6:7], 0x0
-; GCN:         s_waitcnt lgkmcnt(0)
-; GCN:         s_bfe_u32 [[SV2E1:s[0-9]+]], [[S1]], 0x80008
-; GCN:         v_mov_b32_e32 [[V2E1:v[0-9]+]], [[SV2E1]]
-; GCN:         s_bfe_u32 [[SV2E3:s[0-9]+]], [[S1]], 0x80010
-; GCN:         v_mov_b32_e32 [[V2E2:v[0-9]+]]
-; GCN:         s_bfe_u32 [[V1E3:s[0-9]+]], {{s[0-9]+}}, 0x80010
-; GCN-NEXT:    s_lshr_b32 [[SV2E4:s[0-9]+]], [[S1]], 24
-; GCN-NEXT:    v_mov_b32_e32 [[V2E3:v[0-9]+]], [[SV2E3]]
-; GCN-NEXT:    s_lshr_b32 [[V1E4:s[0-9]+]], {{s[0-9]+}}, 24
-; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    v_mad_u32_u24 [[MAD1:v[0-9]+]], {{s[0-9]+}}, [[V2E1]], [[SRC2]]
-; GCN-NEXT:    v_mad_u32_u24 [[MAD2:v[0-9]+]], {{s[0-9]+}}, [[V2E2]], [[MAD1]]
-; GCN-NEXT:    v_mad_u32_u24 [[MAD3:v[0-9]+]], [[V1E3]], [[V2E3]], [[MAD2]]
-; GCN-NEXT:    v_mov_b32_e32 [[V2E4:v[0-9]+]], [[SV2E4]]
-; GCN-NEXT:    v_mad_u32_u24 [[MAD4:v[0-9]+]], [[V1E4]], [[V2E4]], [[MAD3]]
-; GCN-NEXT:    {{buffer|flat|global}}_store_short
-; GCN-NEXT:    s_endpgm
+; GFX7-LABEL: notdot4_mixedtypes:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b32 s8, 0xffff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    buffer_load_ushort v0, off, s[0:3], 0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_sext_i32_i8 s6, s4
+; GFX7-NEXT:    s_bfe_u32 s10, s4, 0x80008
+; GFX7-NEXT:    s_sext_i32_i8 s7, s5
+; GFX7-NEXT:    s_bfe_u32 s9, s5, 0x80008
+; GFX7-NEXT:    s_and_b32 s7, s7, s8
+; GFX7-NEXT:    v_mov_b32_e32 v1, s9
+; GFX7-NEXT:    s_bfe_u32 s11, s5, 0x80010
+; GFX7-NEXT:    s_and_b32 s6, s6, s8
+; GFX7-NEXT:    v_mov_b32_e32 v3, s7
+; GFX7-NEXT:    s_bfe_u32 s12, s4, 0x80010
+; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
+; GFX7-NEXT:    v_mov_b32_e32 v2, s11
+; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_mad_u32_u24 v0, s10, v1, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s6, v3, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s12, v2, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s5
+; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
+; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: notdot4_mixedtypes:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_mov_b32 s2, 0xffff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ushort v2, v[0:1]
+; GFX8-NEXT:    s_load_dword s0, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_bfe_i32 s3, s0, 0x80000
+; GFX8-NEXT:    s_bfe_u32 s6, s1, 0x80008
+; GFX8-NEXT:    s_bfe_i32 s4, s1, 0x80000
+; GFX8-NEXT:    s_and_b32 s3, s2, s3
+; GFX8-NEXT:    s_and_b32 s2, s2, s4
+; GFX8-NEXT:    s_bfe_u32 s5, s0, 0x80008
+; GFX8-NEXT:    v_mov_b32_e32 v3, s6
+; GFX8-NEXT:    s_bfe_u32 s8, s1, 0x80010
+; GFX8-NEXT:    v_mov_b32_e32 v5, s2
+; GFX8-NEXT:    s_bfe_u32 s7, s0, 0x80010
+; GFX8-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX8-NEXT:    v_mov_b32_e32 v4, s8
+; GFX8-NEXT:    s_lshr_b32 s0, s0, 24
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v3, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s3, v5, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v4, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: notdot4_mixedtypes:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_mov_b32 s2, 0xffff
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_load_ushort v2, v[0:1], off
+; GFX9-NODL-NEXT:    s_load_dword s0, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_bfe_i32 s3, s0, 0x80000
+; GFX9-NODL-NEXT:    s_bfe_u32 s6, s1, 0x80008
+; GFX9-NODL-NEXT:    s_bfe_i32 s4, s1, 0x80000
+; GFX9-NODL-NEXT:    s_and_b32 s3, s2, s3
+; GFX9-NODL-NEXT:    s_and_b32 s2, s2, s4
+; GFX9-NODL-NEXT:    s_bfe_u32 s5, s0, 0x80008
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s6
+; GFX9-NODL-NEXT:    s_bfe_u32 s8, s1, 0x80010
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s2
+; GFX9-NODL-NEXT:    s_bfe_u32 s7, s0, 0x80010
+; GFX9-NODL-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s8
+; GFX9-NODL-NEXT:    s_lshr_b32 s0, s0, 24
+; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s5, v3, v2
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s3, v5, v2
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s7, v4, v2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX9-NODL-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: notdot4_mixedtypes:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_mov_b32 s2, 0xffff
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ushort v2, v[0:1], off
+; GFX9-DL-NEXT:    s_load_dword s0, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_bfe_i32 s3, s0, 0x80000
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s1, 0x80008
+; GFX9-DL-NEXT:    s_bfe_i32 s4, s1, 0x80000
+; GFX9-DL-NEXT:    s_and_b32 s3, s2, s3
+; GFX9-DL-NEXT:    s_and_b32 s2, s2, s4
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s0, 0x80008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s6
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s1, 0x80010
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s2
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s0, 0x80010
+; GFX9-DL-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s8
+; GFX9-DL-NEXT:    s_lshr_b32 s0, s0, 24
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s5, v3, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s3, v5, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s7, v4, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX9-DL-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                               <4 x i8> addrspace(1)* %src2,
                                               i16 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
   %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
 
   %v1e0 = extractelement <4 x i8> %vec1, i64 0
   %cv1e0 = sext i8 %v1e0 to i16
entry:
   %add1 = add i16 %mul2, %acc
   %add2 = add i16 %add1, %mul1
   %add3 = add i16 %add2, %mul3
   %add4 = add i16 %add3, %mul4
 
   store i16 %add4, i16 addrspace(1)* %dst, align 2
   ret void
 }
+
+define amdgpu_kernel void @idot4_acc32(<4 x i8> addrspace(1)* %src1,
+; GFX7-LABEL: idot4_acc32:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_load_dword s6, s[0:1], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_sext_i32_i8 s7, s4
+; GFX7-NEXT:    s_sext_i32_i8 s8, s5
+; GFX7-NEXT:    s_bfe_i32 s10, s5, 0x80008
+; GFX7-NEXT:    v_mov_b32_e32 v0, s8
+; GFX7-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7-NEXT:    s_bfe_i32 s12, s5, 0x80010
+; GFX7-NEXT:    v_mad_i32_i24 v0, s7, v0, v1
+; GFX7-NEXT:    s_bfe_i32 s9, s4, 0x80008
+; GFX7-NEXT:    v_mov_b32_e32 v1, s10
+; GFX7-NEXT:    s_bfe_i32 s11, s4, 0x80010
+; GFX7-NEXT:    v_mad_i32_i24 v0, s9, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s12
+; GFX7-NEXT:    s_ashr_i32 s5, s5, 24
+; GFX7-NEXT:    v_mad_i32_i24 v0, s11, v1, v0
+; GFX7-NEXT:    s_ashr_i32 s4, s4, 24
+; GFX7-NEXT:    v_mov_b32_e32 v1, s5
+; GFX7-NEXT:    v_mad_i32_i24 v0, s4, v1, v0
+; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: idot4_acc32:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX8-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_sext_i32_i8 s0, s2
+; GFX8-NEXT:    s_sext_i32_i8 s1, s3
+; GFX8-NEXT:    s_bfe_i32 s6, s3, 0x80008
+; GFX8-NEXT:    v_mov_b32_e32 v2, s1
+; GFX8-NEXT:    v_mov_b32_e32 v3, s4
+; GFX8-NEXT:    s_bfe_i32 s8, s3, 0x80010
+; GFX8-NEXT:    v_mad_i32_i24 v2, s0, v2, v3
+; GFX8-NEXT:    s_bfe_i32 s5, s2, 0x80008
+; GFX8-NEXT:    v_mov_b32_e32 v3, s6
+; GFX8-NEXT:    s_bfe_i32 s7, s2, 0x80010
+; GFX8-NEXT:    v_mad_i32_i24 v2, s5, v3, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s8
+; GFX8-NEXT:    s_ashr_i32 s3, s3, 24
+; GFX8-NEXT:    v_mad_i32_i24 v2, s7, v3, v2
+; GFX8-NEXT:    s_ashr_i32 s2, s2, 24
+; GFX8-NEXT:    v_mov_b32_e32 v3, s3
+; GFX8-NEXT:    v_mad_i32_i24 v2, s2, v3, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: idot4_acc32:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_sext_i32_i8 s0, s2
+; GFX9-NODL-NEXT:    s_sext_i32_i8 s1, s3
+; GFX9-NODL-NEXT:    s_bfe_i32 s6, s3, 0x80008
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-NODL-NEXT:    s_bfe_i32 s8, s3, 0x80010
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v2, s0, v2, v3
+; GFX9-NODL-NEXT:    s_bfe_i32 s5, s2, 0x80008
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s6
+; GFX9-NODL-NEXT:    s_bfe_i32 s7, s2, 0x80010
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v2, s5, v3, v2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s8
+; GFX9-NODL-NEXT:    s_ashr_i32 s3, s3, 24
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v2, s7, v3, v2
+; GFX9-NODL-NEXT:    s_ashr_i32 s2, s2, 24
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v2, s2, v3, v2
+; GFX9-NODL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: idot4_acc32:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s3
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-DL-NEXT:    v_dot4_i32_i8 v2, s2, v2, v3
+; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                       <4 x i8> addrspace(1)* %src2,
+                                       i32 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
+  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
+
+  %v1e0 = extractelement <4 x i8> %vec1, i64 0
+  %cv1e0 = sext i8 %v1e0 to i32
+  %v2e0 = extractelement <4 x i8> %vec2, i64 0
+  %cv2e0 = sext i8 %v2e0 to i32
+  %mul1 = mul nuw nsw i32 %cv1e0, %cv2e0
+
+  %v1e1 = extractelement <4 x i8> %vec1, i64 1
+  %cv1e1 = sext i8 %v1e1 to i32
+  %v2e1 = extractelement <4 x i8> %vec2, i64 1
+  %cv2e1 = sext i8 %v2e1 to i32
+  %mul2 = mul nuw nsw i32 %cv1e1, %cv2e1
+
+  %v1e2 = extractelement <4 x i8> %vec1, i64 2
+  %cv1e2 = sext i8 %v1e2 to i32
+  %v2e2 = extractelement <4 x i8> %vec2, i64 2
+  %cv2e2 = sext i8 %v2e2 to i32
+  %mul3 = mul nuw nsw i32 %cv1e2, %cv2e2
+
+  %v1e3 = extractelement <4 x i8> %vec1, i64 3
+  %cv1e3 = sext i8 %v1e3 to i32
+  %v2e3 = extractelement <4 x i8> %vec2, i64 3
+  %cv2e3 = sext i8 %v2e3 to i32
+  %mul4 = mul nuw nsw i32 %cv1e3, %cv2e3
+
+  %acc = load i32, i32 addrspace(1)* %dst, align 4
+  %add1 = add i32 %mul1, %acc
+  %add2 = add i32 %add1, %mul2
+  %add3 = add i32 %add2, %mul3
+  %add4 = add i32 %add3, %mul4
+  store i32 %add4, i32 addrspace(1)* %dst, align 4
+  ret void
+}
+
+; TODO: Currently, vector elements{0 and 3} get zero_extended from i16 to i32 which should
+; be sign_extended directly to i32; prevents the pattern recognizer to recognize this pattern.
+define amdgpu_kernel void @idot4_acc16(<4 x i8> addrspace(1)* %src1,
+; GFX7-LABEL: idot4_acc16:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b32 s8, 0xffff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    buffer_load_ushort v0, off, s[0:3], 0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_sext_i32_i8 s6, s4
+; GFX7-NEXT:    s_bfe_i32 s9, s4, 0x80008
+; GFX7-NEXT:    s_sext_i32_i8 s7, s5
+; GFX7-NEXT:    s_bfe_i32 s10, s5, 0x80008
+; GFX7-NEXT:    s_and_b32 s7, s7, s8
+; GFX7-NEXT:    s_bfe_i32 s12, s5, 0x80010
+; GFX7-NEXT:    s_and_b32 s10, s10, s8
+; GFX7-NEXT:    s_and_b32 s6, s6, s8
+; GFX7-NEXT:    v_mov_b32_e32 v1, s7
+; GFX7-NEXT:    s_bfe_i32 s11, s4, 0x80010
+; GFX7-NEXT:    s_ashr_i32 s5, s5, 24
+; GFX7-NEXT:    s_and_b32 s12, s12, s8
+; GFX7-NEXT:    s_and_b32 s9, s9, s8
+; GFX7-NEXT:    v_mov_b32_e32 v2, s10
+; GFX7-NEXT:    s_ashr_i32 s4, s4, 24
+; GFX7-NEXT:    s_and_b32 s11, s11, s8
+; GFX7-NEXT:    s_and_b32 s5, s5, s8
+; GFX7-NEXT:    v_mov_b32_e32 v3, s12
+; GFX7-NEXT:    s_and_b32 s4, s4, s8
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_mad_u32_u24 v0, s6, v1, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v2, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v3, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s5
+; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
+; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: idot4_acc16:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ushort v2, v[0:1]
+; GFX8-NEXT:    s_load_dword s1, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s5, s1, 8
+; GFX8-NEXT:    s_lshr_b32 s6, s2, 8
+; GFX8-NEXT:    s_sext_i32_i8 s4, s2
+; GFX8-NEXT:    s_bfe_i32 s5, s5, 0x80000
+; GFX8-NEXT:    s_bfe_i32 s6, s6, 0x80000
+; GFX8-NEXT:    s_bfe_i32 s8, s2, 0x80010
+; GFX8-NEXT:    s_lshr_b32 s2, s2, 24
+; GFX8-NEXT:    v_mov_b32_e32 v3, s4
+; GFX8-NEXT:    s_sext_i32_i8 s3, s1
+; GFX8-NEXT:    s_bfe_i32 s7, s1, 0x80010
+; GFX8-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX8-NEXT:    s_and_b32 s4, s0, s5
+; GFX8-NEXT:    s_and_b32 s5, s0, s6
+; GFX8-NEXT:    s_bfe_i32 s1, s1, 0x80000
+; GFX8-NEXT:    s_bfe_i32 s2, s2, 0x80000
+; GFX8-NEXT:    v_mov_b32_e32 v5, s5
+; GFX8-NEXT:    s_and_b32 s1, s0, s1
+; GFX8-NEXT:    v_mov_b32_e32 v4, s8
+; GFX8-NEXT:    s_and_b32 s0, s0, s2
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mad_i32_i24 v2, s3, v3, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s4, v5, v2
+; GFX8-NEXT:    v_mad_i32_i24 v2, s7, v4, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s0
+; GFX8-NEXT:    v_mad_u32_u24 v2, s1, v3, v2
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: idot4_acc16:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_load_ushort v2, v[0:1], off
+; GFX9-NODL-NEXT:    s_load_dword s1, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_lshr_b32 s5, s1, 8
+; GFX9-NODL-NEXT:    s_lshr_b32 s6, s2, 8
+; GFX9-NODL-NEXT:    s_sext_i32_i8 s4, s2
+; GFX9-NODL-NEXT:    s_bfe_i32 s5, s5, 0x80000
+; GFX9-NODL-NEXT:    s_bfe_i32 s6, s6, 0x80000
+; GFX9-NODL-NEXT:    s_bfe_i32 s8, s2, 0x80010
+; GFX9-NODL-NEXT:    s_lshr_b32 s2, s2, 24
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-NODL-NEXT:    s_sext_i32_i8 s3, s1
+; GFX9-NODL-NEXT:    s_bfe_i32 s7, s1, 0x80010
+; GFX9-NODL-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX9-NODL-NEXT:    s_and_b32 s4, s0, s5
+; GFX9-NODL-NEXT:    s_and_b32 s5, s0, s6
+; GFX9-NODL-NEXT:    s_bfe_i32 s1, s1, 0x80000
+; GFX9-NODL-NEXT:    s_bfe_i32 s2, s2, 0x80000
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s5
+; GFX9-NODL-NEXT:    s_and_b32 s1, s0, s1
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s8
+; GFX9-NODL-NEXT:    s_and_b32 s0, s0, s2
+; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v2, s3, v3, v2
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s4, v5, v2
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v2, s7, v4, v2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s0
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s1, v3, v2
+; GFX9-NODL-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: idot4_acc16:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ushort v2, v[0:1], off
+; GFX9-DL-NEXT:    s_load_dword s1, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_lshr_b32 s5, s1, 8
+; GFX9-DL-NEXT:    s_lshr_b32 s6, s2, 8
+; GFX9-DL-NEXT:    s_sext_i32_i8 s4, s2
+; GFX9-DL-NEXT:    s_bfe_i32 s5, s5, 0x80000
+; GFX9-DL-NEXT:    s_bfe_i32 s6, s6, 0x80000
+; GFX9-DL-NEXT:    s_bfe_i32 s8, s2, 0x80010
+; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 24
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-DL-NEXT:    s_sext_i32_i8 s3, s1
+; GFX9-DL-NEXT:    s_bfe_i32 s7, s1, 0x80010
+; GFX9-DL-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX9-DL-NEXT:    s_and_b32 s4, s0, s5
+; GFX9-DL-NEXT:    s_and_b32 s5, s0, s6
+; GFX9-DL-NEXT:    s_bfe_i32 s1, s1, 0x80000
+; GFX9-DL-NEXT:    s_bfe_i32 s2, s2, 0x80000
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s5
+; GFX9-DL-NEXT:    s_and_b32 s1, s0, s1
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s8
+; GFX9-DL-NEXT:    s_and_b32 s0, s0, s2
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s3, v3, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s4, v5, v2
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s7, v4, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s0
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s1, v3, v2
+; GFX9-DL-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                       <4 x i8> addrspace(1)* %src2,
+                                       i16 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
+  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
+
+  %v1e0 = extractelement <4 x i8> %vec1, i64 0
+  %cv1e0 = sext i8 %v1e0 to i16
+  %v2e0 = extractelement <4 x i8> %vec2, i64 0
+  %cv2e0 = sext i8 %v2e0 to i16
+  %mul1 = mul nsw i16 %cv1e0, %cv2e0
+
+  %v1e1 = extractelement <4 x i8> %vec1, i64 1
+  %cv1e1 = sext i8 %v1e1 to i16
+  %v2e1 = extractelement <4 x i8> %vec2, i64 1
+  %cv2e1 = sext i8 %v2e1 to i16
+  %mul2 = mul nsw i16 %cv1e1, %cv2e1
+
+  %v1e2 = extractelement <4 x i8> %vec1, i64 2
+  %cv1e2 = sext i8 %v1e2 to i16
+  %v2e2 = extractelement <4 x i8> %vec2, i64 2
+  %cv2e2 = sext i8 %v2e2 to i16
+  %mul3 = mul nsw i16 %cv1e2, %cv2e2
+
+  %v1e3 = extractelement <4 x i8> %vec1, i64 3
+  %cv1e3 = sext i8 %v1e3 to i16
+  %v2e3 = extractelement <4 x i8> %vec2, i64 3
+  %cv2e3 = sext i8 %v2e3 to i16
+  %mul4 = mul nsw i16 %cv1e3, %cv2e3
+
+  %acc = load i16, i16 addrspace(1)* %dst, align 2
+  %add1 = add i16 %mul1, %acc
+  %add2 = add i16 %add1, %mul2
+  %add3 = add i16 %add2, %mul3
+  %add4 = add i16 %add3, %mul4
+  store i16 %add4, i16 addrspace(1)* %dst, align 2
+  ret void
+}
+
+define amdgpu_kernel void @idot4_acc8(<4 x i8> addrspace(1)* %src1,
+; GFX7-LABEL: idot4_acc8:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_movk_i32 s8, 0xff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_and_b32 s7, s4, s8
+; GFX7-NEXT:    s_bfe_u32 s9, s4, 0x80008
+; GFX7-NEXT:    s_and_b32 s6, s5, s8
+; GFX7-NEXT:    s_bfe_u32 s8, s5, 0x80008
+; GFX7-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80010
+; GFX7-NEXT:    v_mov_b32_e32 v2, s8
+; GFX7-NEXT:    s_bfe_u32 s11, s4, 0x80010
+; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
+; GFX7-NEXT:    v_mov_b32_e32 v3, s10
+; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v2, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v3, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s5
+; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
+; GFX7-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: idot4_acc8:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_movk_i32 s2, 0xff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ubyte v2, v[0:1]
+; GFX8-NEXT:    s_load_dword s0, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_bfe_u32 s4, s0, 0x80008
+; GFX8-NEXT:    s_and_b32 s3, s1, s2
+; GFX8-NEXT:    s_bfe_u32 s5, s1, 0x80008
+; GFX8-NEXT:    s_and_b32 s2, s0, s2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s3
+; GFX8-NEXT:    s_bfe_u32 s6, s1, 0x80010
+; GFX8-NEXT:    v_mov_b32_e32 v4, s5
+; GFX8-NEXT:    s_bfe_u32 s7, s0, 0x80010
+; GFX8-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX8-NEXT:    v_mov_b32_e32 v5, s6
+; GFX8-NEXT:    s_lshr_b32 s0, s0, 24
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s4, v4, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX8-NEXT:    flat_store_byte v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: idot4_acc8:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_movk_i32 s2, 0xff
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-NODL-NEXT:    s_load_dword s0, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_bfe_u32 s4, s0, 0x80008
+; GFX9-NODL-NEXT:    s_and_b32 s3, s1, s2
+; GFX9-NODL-NEXT:    s_bfe_u32 s5, s1, 0x80008
+; GFX9-NODL-NEXT:    s_and_b32 s2, s0, s2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-NODL-NEXT:    s_bfe_u32 s6, s1, 0x80010
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s5
+; GFX9-NODL-NEXT:    s_bfe_u32 s7, s0, 0x80010
+; GFX9-NODL-NEXT:    s_lshr_b32 s1, s1, 24
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s6
+; GFX9-NODL-NEXT:    s_lshr_b32 s0, s0, 24
+; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s4, v4, v2
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX9-NODL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: idot4_acc8:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_dot4_u32_u8 v2, s2, v3, v2
+; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                      <4 x i8> addrspace(1)* %src2,
+                                      i8 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
+  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
+
+  %v1e0 = extractelement <4 x i8> %vec1, i64 0
+  %v2e0 = extractelement <4 x i8> %vec2, i64 0
+  %mul1 = mul i8 %v1e0, %v2e0
+
+  %v1e1 = extractelement <4 x i8> %vec1, i64 1
+  %v2e1 = extractelement <4 x i8> %vec2, i64 1
+  %mul2 = mul i8 %v1e1, %v2e1
+
+  %v1e2 = extractelement <4 x i8> %vec1, i64 2
+  %v2e2 = extractelement <4 x i8> %vec2, i64 2
+  %mul3 = mul i8 %v1e2, %v2e2
+
+  %v1e3 = extractelement <4 x i8> %vec1, i64 3
+  %v2e3 = extractelement <4 x i8> %vec2, i64 3
+  %mul4 = mul i8 %v1e3, %v2e3
+
+  %acc = load i8, i8 addrspace(1)* %dst, align 2
+  %add1 = add i8 %mul1, %acc
+  %add2 = add i8 %add1, %mul2
+  %add3 = add i8 %add2, %mul3
+  %add4 = add nsw i8 %add3, %mul4
+  store i8 %add4, i8 addrspace(1)* %dst, align 2
+  ret void
+}
+
+define amdgpu_kernel void @idot4_multiuse_mul1(<4 x i8> addrspace(1)* %src1,
+; GFX7-LABEL: idot4_multiuse_mul1:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_load_dword s6, s[0:1], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_sext_i32_i8 s7, s4
+; GFX7-NEXT:    s_sext_i32_i8 s8, s5
+; GFX7-NEXT:    s_bfe_i32 s10, s5, 0x80008
+; GFX7-NEXT:    v_mov_b32_e32 v0, s8
+; GFX7-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7-NEXT:    s_bfe_i32 s9, s4, 0x80008
+; GFX7-NEXT:    v_mad_i32_i24 v1, s7, v0, v1
+; GFX7-NEXT:    v_mov_b32_e32 v2, s10
+; GFX7-NEXT:    s_bfe_i32 s12, s5, 0x80010
+; GFX7-NEXT:    v_mad_i32_i24 v1, s9, v2, v1
+; GFX7-NEXT:    s_bfe_i32 s11, s4, 0x80010
+; GFX7-NEXT:    v_mad_i32_i24 v0, s7, v0, v1
+; GFX7-NEXT:    v_mov_b32_e32 v1, s12
+; GFX7-NEXT:    s_ashr_i32 s5, s5, 24
+; GFX7-NEXT:    v_mad_i32_i24 v0, s11, v1, v0
+; GFX7-NEXT:    s_ashr_i32 s4, s4, 24
+; GFX7-NEXT:    v_mov_b32_e32 v1, s5
+; GFX7-NEXT:    v_mad_i32_i24 v0, s4, v1, v0
+; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: idot4_multiuse_mul1:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX8-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_sext_i32_i8 s0, s2
+; GFX8-NEXT:    s_sext_i32_i8 s1, s3
+; GFX8-NEXT:    s_bfe_i32 s6, s3, 0x80008
+; GFX8-NEXT:    v_mov_b32_e32 v2, s1
+; GFX8-NEXT:    v_mov_b32_e32 v3, s4
+; GFX8-NEXT:    s_bfe_i32 s5, s2, 0x80008
+; GFX8-NEXT:    v_mad_i32_i24 v3, s0, v2, v3
+; GFX8-NEXT:    v_mov_b32_e32 v4, s6
+; GFX8-NEXT:    s_bfe_i32 s8, s3, 0x80010
+; GFX8-NEXT:    v_mad_i32_i24 v3, s5, v4, v3
+; GFX8-NEXT:    s_bfe_i32 s7, s2, 0x80010
+; GFX8-NEXT:    v_mad_i32_i24 v2, s0, v2, v3
+; GFX8-NEXT:    v_mov_b32_e32 v3, s8
+; GFX8-NEXT:    s_ashr_i32 s3, s3, 24
+; GFX8-NEXT:    v_mad_i32_i24 v2, s7, v3, v2
+; GFX8-NEXT:    s_ashr_i32 s2, s2, 24
+; GFX8-NEXT:    v_mov_b32_e32 v3, s3
+; GFX8-NEXT:    v_mad_i32_i24 v2, s2, v3, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: idot4_multiuse_mul1:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_sext_i32_i8 s0, s2
+; GFX9-NODL-NEXT:    s_sext_i32_i8 s1, s3
+; GFX9-NODL-NEXT:    s_bfe_i32 s6, s3, 0x80008
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-NODL-NEXT:    s_bfe_i32 s5, s2, 0x80008
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v3, s0, v2, v3
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s6
+; GFX9-NODL-NEXT:    s_bfe_i32 s8, s3, 0x80010
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v3, s5, v4, v3
+; GFX9-NODL-NEXT:    s_bfe_i32 s7, s2, 0x80010
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v2, s0, v2, v3
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s8
+; GFX9-NODL-NEXT:    s_ashr_i32 s3, s3, 24
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v2, s7, v3, v2
+; GFX9-NODL-NEXT:    s_ashr_i32 s2, s2, 24
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v2, s2, v3, v2
+; GFX9-NODL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: idot4_multiuse_mul1:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_sext_i32_i8 s0, s2
+; GFX9-DL-NEXT:    s_sext_i32_i8 s1, s3
+; GFX9-DL-NEXT:    s_bfe_i32 s6, s3, 0x80008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-DL-NEXT:    s_bfe_i32 s5, s2, 0x80008
+; GFX9-DL-NEXT:    v_mad_i32_i24 v3, s0, v2, v3
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s6
+; GFX9-DL-NEXT:    s_bfe_i32 s8, s3, 0x80010
+; GFX9-DL-NEXT:    v_mad_i32_i24 v3, s5, v4, v3
+; GFX9-DL-NEXT:    s_bfe_i32 s7, s2, 0x80010
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s0, v2, v3
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s8
+; GFX9-DL-NEXT:    s_ashr_i32 s3, s3, 24
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s7, v3, v2
+; GFX9-DL-NEXT:    s_ashr_i32 s2, s2, 24
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s2, v3, v2
+; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                               <4 x i8> addrspace(1)* %src2,
+                                               i32 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
+  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
+
+  %v1e0 = extractelement <4 x i8> %vec1, i64 0
+  %cv1e0 = sext i8 %v1e0 to i32
+  %v2e0 = extractelement <4 x i8> %vec2, i64 0
+  %cv2e0 = sext i8 %v2e0 to i32
+  %mul1 = mul nuw nsw i32 %cv1e0, %cv2e0
+
+  %v1e1 = extractelement <4 x i8> %vec1, i64 1
+  %cv1e1 = sext i8 %v1e1 to i32
+  %v2e1 = extractelement <4 x i8> %vec2, i64 1
+  %cv2e1 = sext i8 %v2e1 to i32
+  %mul2 = mul nuw nsw i32 %cv1e1, %cv2e1
+
+  %v1e2 = extractelement <4 x i8> %vec1, i64 2
+  %cv1e2 = sext i8 %v1e2 to i32
+  %v2e2 = extractelement <4 x i8> %vec2, i64 2
+  %cv2e2 = sext i8 %v2e2 to i32
+  %mul3 = mul nuw nsw i32 %cv1e2, %cv2e2
+
+  %v1e3 = extractelement <4 x i8> %vec1, i64 3
+  %cv1e3 = sext i8 %v1e3 to i32
+  %v2e3 = extractelement <4 x i8> %vec2, i64 3
+  %cv2e3 = sext i8 %v2e3 to i32
+  %mul4 = mul nuw nsw i32 %cv1e3, %cv2e3
+
+  %acc = load i32, i32 addrspace(1)* %dst, align 4
+  %add = add i32 %mul1, %acc
+  %add1 = add i32 %mul2, %add
+  %add2 = add i32 %add1, %mul1
+  %add3 = add i32 %add2, %mul3
+  %add4 = add i32 %add3, %mul4
+
+  store i32 %add4, i32 addrspace(1)* %dst, align 4
+  ret void
+}
+
+; TODO: cleanup s_lshr_b32 and support this pattern.
+define amdgpu_kernel void @udot4_acc32_vecMul(<4 x i8> addrspace(1)* %src1,
+; GFX7-LABEL: udot4_acc32_vecMul:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_movk_i32 s12, 0xff
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_load_dword s6, s[0:1], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_lshr_b32 s7, s4, 24
+; GFX7-NEXT:    s_lshr_b32 s9, s5, 24
+; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80008
+; GFX7-NEXT:    s_bfe_u32 s13, s5, 0x80010
+; GFX7-NEXT:    s_and_b32 s5, s5, s12
+; GFX7-NEXT:    s_bfe_u32 s8, s4, 0x80008
+; GFX7-NEXT:    s_bfe_u32 s11, s4, 0x80010
+; GFX7-NEXT:    s_and_b32 s4, s4, s12
+; GFX7-NEXT:    v_mov_b32_e32 v0, s5
+; GFX7-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v0, v1
+; GFX7-NEXT:    v_mov_b32_e32 v1, s10
+; GFX7-NEXT:    v_mad_u32_u24 v0, s8, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s13
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s9
+; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
+; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: udot4_acc32_vecMul:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_movk_i32 s2, 0xff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX8-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s6, s3, 24
+; GFX8-NEXT:    s_bfe_u32 s8, s3, 0x80010
+; GFX8-NEXT:    v_lshrrev_b16_e64 v0, 8, s3
+; GFX8-NEXT:    s_and_b32 s3, s3, s2
+; GFX8-NEXT:    s_and_b32 s2, s4, s2
+; GFX8-NEXT:    v_mov_b32_e32 v2, s2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s5
+; GFX8-NEXT:    s_bfe_u32 s9, s4, 0x80010
+; GFX8-NEXT:    v_lshrrev_b16_e64 v1, 8, s4
+; GFX8-NEXT:    v_mad_u32_u24 v2, s3, v2, v3
+; GFX8-NEXT:    v_mad_u32_u24 v0, v0, v1, v2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s9
+; GFX8-NEXT:    s_lshr_b32 s7, s4, 24
+; GFX8-NEXT:    v_mad_u32_u24 v0, s8, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s7
+; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: udot4_acc32_vecMul:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_movk_i32 s2, 0xff
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_lshr_b32 s6, s3, 24
+; GFX9-NODL-NEXT:    s_bfe_u32 s8, s3, 0x80010
+; GFX9-NODL-NEXT:    v_lshrrev_b16_e64 v0, 8, s3
+; GFX9-NODL-NEXT:    s_and_b32 s3, s3, s2
+; GFX9-NODL-NEXT:    s_and_b32 s2, s4, s2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-NODL-NEXT:    s_bfe_u32 s9, s4, 0x80010
+; GFX9-NODL-NEXT:    v_lshrrev_b16_e64 v1, 8, s4
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s3, v2, v3
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, v0, v1, v2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s9
+; GFX9-NODL-NEXT:    s_lshr_b32 s7, s4, 24
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s8, v1, v0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s7
+; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s6, v1, v0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: udot4_acc32_vecMul:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_movk_i32 s2, 0xff
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s3, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_lshr_b32 s6, s3, 24
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s3, 0x80010
+; GFX9-DL-NEXT:    v_lshrrev_b16_e64 v0, 8, s3
+; GFX9-DL-NEXT:    s_and_b32 s3, s3, s2
+; GFX9-DL-NEXT:    s_and_b32 s2, s4, s2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-DL-NEXT:    s_bfe_u32 s9, s4, 0x80010
+; GFX9-DL-NEXT:    v_lshrrev_b16_e64 v1, 8, s4
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s3, v2, v3
+; GFX9-DL-NEXT:    v_mad_u32_u24 v0, v0, v1, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s9
+; GFX9-DL-NEXT:    s_lshr_b32 s7, s4, 24
+; GFX9-DL-NEXT:    v_mad_u32_u24 v0, s8, v1, v0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s7
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s6, v1, v0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                              <4 x i8> addrspace(1)* %src2,
+                                              i32 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
+  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
+
+  %cvec1 = zext <4 x i8> %vec1 to <4 x i32>
+  %cvec2 = zext <4 x i8> %vec2 to <4 x i32>
+
+  %mul = mul <4 x i32> %cvec1, %cvec2
+  %mul0 = extractelement <4 x i32> %mul, i64 0
+  %mul1 = extractelement <4 x i32> %mul, i64 1
+  %mul2 = extractelement <4 x i32> %mul, i64 2
+  %mul3 = extractelement <4 x i32> %mul, i64 3
+
+  %acc = load i32, i32 addrspace(1)* %dst, align 4
+  %add1 = add i32 %mul0, %acc
+  %add2 = add i32 %add1, %mul1
+  %add3 = add i32 %add2, %mul2
+  %add4 = add i32 %add3, %mul3
+
+  store i32 %add4, i32 addrspace(1)* %dst, align 4
+  ret void
+}
+
+; TODO: Support this pattern.
+define amdgpu_kernel void @idot4_acc32_vecMul(<4 x i8> addrspace(1)* %src1,
+; GFX7-LABEL: idot4_acc32_vecMul:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_load_dword s6, s[0:1], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_ashr_i32 s7, s4, 24
+; GFX7-NEXT:    s_ashr_i32 s10, s5, 24
+; GFX7-NEXT:    s_bfe_i32 s11, s5, 0x80010
+; GFX7-NEXT:    s_bfe_i32 s12, s5, 0x80008
+; GFX7-NEXT:    s_sext_i32_i8 s5, s5
+; GFX7-NEXT:    s_bfe_i32 s8, s4, 0x80010
+; GFX7-NEXT:    s_bfe_i32 s9, s4, 0x80008
+; GFX7-NEXT:    s_sext_i32_i8 s4, s4
+; GFX7-NEXT:    v_mov_b32_e32 v0, s5
+; GFX7-NEXT:    v_mov_b32_e32 v1, s6
+; GFX7-NEXT:    v_mad_i32_i24 v0, s4, v0, v1
+; GFX7-NEXT:    v_mov_b32_e32 v1, s12
+; GFX7-NEXT:    v_mad_i32_i24 v0, s9, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s11
+; GFX7-NEXT:    v_mad_i32_i24 v0, s8, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s10
+; GFX7-NEXT:    v_mad_i32_i24 v0, s7, v1, v0
+; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: idot4_acc32_vecMul:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX8-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b16_e64 v2, 8, s2
+; GFX8-NEXT:    v_lshrrev_b16_e64 v3, 8, s3
+; GFX8-NEXT:    s_ashr_i32 s5, s3, 24
+; GFX8-NEXT:    s_bfe_i32 s6, s3, 0x80010
+; GFX8-NEXT:    s_sext_i32_i8 s3, s3
+; GFX8-NEXT:    s_ashr_i32 s0, s2, 24
+; GFX8-NEXT:    s_bfe_i32 s1, s2, 0x80010
+; GFX8-NEXT:    s_sext_i32_i8 s2, s2
+; GFX8-NEXT:    v_mov_b32_e32 v4, s3
+; GFX8-NEXT:    v_mov_b32_e32 v5, s4
+; GFX8-NEXT:    v_bfe_i32 v2, v2, 0, 8
+; GFX8-NEXT:    v_bfe_i32 v3, v3, 0, 8
+; GFX8-NEXT:    v_mad_i32_i24 v4, s2, v4, v5
+; GFX8-NEXT:    v_mad_i32_i24 v2, v2, v3, v4
+; GFX8-NEXT:    v_mov_b32_e32 v3, s6
+; GFX8-NEXT:    v_mad_i32_i24 v2, s1, v3, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s5
+; GFX8-NEXT:    v_mad_i32_i24 v2, s0, v3, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: idot4_acc32_vecMul:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    v_lshrrev_b16_e64 v2, 8, s2
+; GFX9-NODL-NEXT:    v_lshrrev_b16_e64 v3, 8, s3
+; GFX9-NODL-NEXT:    s_ashr_i32 s5, s3, 24
+; GFX9-NODL-NEXT:    s_bfe_i32 s6, s3, 0x80010
+; GFX9-NODL-NEXT:    s_sext_i32_i8 s3, s3
+; GFX9-NODL-NEXT:    s_ashr_i32 s0, s2, 24
+; GFX9-NODL-NEXT:    s_bfe_i32 s1, s2, 0x80010
+; GFX9-NODL-NEXT:    s_sext_i32_i8 s2, s2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s3
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s4
+; GFX9-NODL-NEXT:    v_bfe_i32 v2, v2, 0, 8
+; GFX9-NODL-NEXT:    v_bfe_i32 v3, v3, 0, 8
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v4, s2, v4, v5
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v2, v2, v3, v4
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s6
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v2, s1, v3, v2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-NODL-NEXT:    v_mad_i32_i24 v2, s0, v3, v2
+; GFX9-NODL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: idot4_acc32_vecMul:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_lshrrev_b16_e64 v2, 8, s2
+; GFX9-DL-NEXT:    v_lshrrev_b16_e64 v3, 8, s3
+; GFX9-DL-NEXT:    s_ashr_i32 s5, s3, 24
+; GFX9-DL-NEXT:    s_bfe_i32 s6, s3, 0x80010
+; GFX9-DL-NEXT:    s_sext_i32_i8 s3, s3
+; GFX9-DL-NEXT:    s_ashr_i32 s0, s2, 24
+; GFX9-DL-NEXT:    s_bfe_i32 s1, s2, 0x80010
+; GFX9-DL-NEXT:    s_sext_i32_i8 s2, s2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s3
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s4
+; GFX9-DL-NEXT:    v_bfe_i32 v2, v2, 0, 8
+; GFX9-DL-NEXT:    v_bfe_i32 v3, v3, 0, 8
+; GFX9-DL-NEXT:    v_mad_i32_i24 v4, s2, v4, v5
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, v2, v3, v4
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s6
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s1, v3, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s0, v3, v2
+; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                              <4 x i8> addrspace(1)* %src2,
+                                              i32 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
+  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
+
+  %cvec1 = sext <4 x i8> %vec1 to <4 x i32>
+  %cvec2 = sext <4 x i8> %vec2 to <4 x i32>
+
+  %mul = mul <4 x i32> %cvec1, %cvec2
+  %mul0 = extractelement <4 x i32> %mul, i64 0
+  %mul1 = extractelement <4 x i32> %mul, i64 1
+  %mul2 = extractelement <4 x i32> %mul, i64 2
+  %mul3 = extractelement <4 x i32> %mul, i64 3
+
+  %acc = load i32, i32 addrspace(1)* %dst, align 4
+  %add1 = add i32 %mul0, %acc
+  %add2 = add i32 %add1, %mul1
+  %add3 = add i32 %add2, %mul2
+  %add4 = add i32 %add3, %mul3
+
+  store i32 %add4, i32 addrspace(1)* %dst, align 4
+  ret void
+}
+
+; TODO: This pattern should be recognized.
+define amdgpu_kernel void @udot4_acc16_vecMul(<4 x i8> addrspace(1)* %src1,
+; GFX7-LABEL: udot4_acc16_vecMul:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_movk_i32 s8, 0xff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    buffer_load_ushort v0, off, s[0:3], 0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_and_b32 s11, s4, s8
+; GFX7-NEXT:    s_bfe_u32 s6, s4, 0x80008
+; GFX7-NEXT:    s_bfe_u32 s9, s5, 0x80008
+; GFX7-NEXT:    s_lshr_b32 s10, s5, 24
+; GFX7-NEXT:    s_and_b32 s8, s5, s8
+; GFX7-NEXT:    v_mov_b32_e32 v4, s9
+; GFX7-NEXT:    s_lshr_b32 s7, s4, 24
+; GFX7-NEXT:    v_mov_b32_e32 v2, s10
+; GFX7-NEXT:    s_bfe_u32 s5, s5, 0x80010
+; GFX7-NEXT:    v_mov_b32_e32 v3, s8
+; GFX7-NEXT:    v_mul_u32_u24_e32 v2, s7, v2
+; GFX7-NEXT:    v_mul_u32_u24_e32 v4, s6, v4
+; GFX7-NEXT:    s_bfe_u32 s4, s4, 0x80010
+; GFX7-NEXT:    v_mov_b32_e32 v1, s5
+; GFX7-NEXT:    v_mul_u32_u24_e32 v1, s4, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_mul_u32_u24_e32 v3, s11, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_or_b32_e32 v2, v3, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v3, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
+; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: udot4_acc16_vecMul:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ushort v2, v[0:1]
+; GFX8-NEXT:    s_load_dword s1, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s6, s1, s0
+; GFX8-NEXT:    s_and_b32 s0, s2, s0
+; GFX8-NEXT:    v_mov_b32_e32 v5, s0
+; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x80010
+; GFX8-NEXT:    v_lshrrev_b16_e64 v3, 8, s2
+; GFX8-NEXT:    v_lshrrev_b16_e64 v4, 8, s1
+; GFX8-NEXT:    s_lshr_b32 s4, s2, 24
+; GFX8-NEXT:    s_bfe_u32 s5, s1, 0x80010
+; GFX8-NEXT:    v_mov_b32_e32 v6, s7
+; GFX8-NEXT:    s_lshr_b32 s3, s1, 24
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, v4, v3, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v6, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s4
+; GFX8-NEXT:    v_mad_u32_u24 v2, s3, v3, v2
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: udot4_acc16_vecMul:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, 0xffff
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_lshr_b32 s5, s2, 16
+; GFX9-NODL-NEXT:    s_lshr_b32 s7, s3, 16
+; GFX9-NODL-NEXT:    s_lshr_b32 s4, s2, 24
+; GFX9-NODL-NEXT:    v_and_b32_sdwa v4, v0, s5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NODL-NEXT:    s_lshr_b32 s6, s3, 24
+; GFX9-NODL-NEXT:    v_and_b32_sdwa v3, v0, s7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NODL-NEXT:    v_lshl_or_b32 v3, s6, 16, v3
+; GFX9-NODL-NEXT:    v_lshl_or_b32 v4, s4, 16, v4
+; GFX9-NODL-NEXT:    v_pk_mul_lo_u16 v3, v4, v3
+; GFX9-NODL-NEXT:    v_and_b32_sdwa v4, v0, s3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NODL-NEXT:    v_lshrrev_b16_e64 v2, 8, s3
+; GFX9-NODL-NEXT:    v_lshrrev_b16_e64 v1, 8, s2
+; GFX9-NODL-NEXT:    v_and_b32_sdwa v0, v0, s2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NODL-NEXT:    v_lshl_or_b32 v2, v2, 16, v4
+; GFX9-NODL-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-NODL-NEXT:    v_pk_mul_lo_u16 v2, v0, v2
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_load_ushort v4, v[0:1], off
+; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NODL-NEXT:    v_add_u32_e32 v4, v2, v4
+; GFX9-NODL-NEXT:    v_add_u32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NODL-NEXT:    v_add_u32_e32 v2, v2, v3
+; GFX9-NODL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NODL-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: udot4_acc16_vecMul:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, 0xffff
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_lshr_b32 s5, s2, 16
+; GFX9-DL-NEXT:    s_lshr_b32 s7, s3, 16
+; GFX9-DL-NEXT:    s_lshr_b32 s4, s2, 24
+; GFX9-DL-NEXT:    v_and_b32_sdwa v4, v0, s5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-DL-NEXT:    s_lshr_b32 s6, s3, 24
+; GFX9-DL-NEXT:    v_and_b32_sdwa v3, v0, s7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-DL-NEXT:    v_lshl_or_b32 v3, s6, 16, v3
+; GFX9-DL-NEXT:    v_lshl_or_b32 v4, s4, 16, v4
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v3, v4, v3
+; GFX9-DL-NEXT:    v_and_b32_sdwa v4, v0, s3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-DL-NEXT:    v_lshrrev_b16_e64 v2, 8, s3
+; GFX9-DL-NEXT:    v_lshrrev_b16_e64 v1, 8, s2
+; GFX9-DL-NEXT:    v_and_b32_sdwa v0, v0, s2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-DL-NEXT:    v_lshl_or_b32 v2, v2, 16, v4
+; GFX9-DL-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v2, v0, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ushort v4, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_add_u32_e32 v4, v2, v4
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v3
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                              <4 x i8> addrspace(1)* %src2,
+                                              i16 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
+  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
+
+  %cvec1 = zext <4 x i8> %vec1 to <4 x i16>
+  %cvec2 = zext <4 x i8> %vec2 to <4 x i16>
+
+  %mul = mul <4 x i16> %cvec1, %cvec2
+  %mul0 = extractelement <4 x i16> %mul, i64 0
+  %mul1 = extractelement <4 x i16> %mul, i64 1
+  %mul2 = extractelement <4 x i16> %mul, i64 2
+  %mul3 = extractelement <4 x i16> %mul, i64 3
+
+  %acc = load i16, i16 addrspace(1)* %dst, align 4
+  %add1 = add i16 %mul0, %acc
+  %add2 = add i16 %add1, %mul1
+  %add3 = add i16 %add2, %mul2
+  %add4 = add i16 %add3, %mul3
+
+  store i16 %add4, i16 addrspace(1)* %dst, align 4
+  ret void
+}
+
+define amdgpu_kernel void @idot4_acc16_vecMul(<4 x i8> addrspace(1)* %src1,
+; GFX7-LABEL: idot4_acc16_vecMul:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b32 s8, 0xffff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    buffer_load_ushort v0, off, s[0:3], 0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_sext_i32_i8 s6, s4
+; GFX7-NEXT:    s_bfe_i32 s7, s4, 0x80008
+; GFX7-NEXT:    s_sext_i32_i8 s10, s5
+; GFX7-NEXT:    s_bfe_i32 s11, s5, 0x80008
+; GFX7-NEXT:    s_bfe_i32 s12, s5, 0x80010
+; GFX7-NEXT:    s_ashr_i32 s5, s5, 24
+; GFX7-NEXT:    v_mov_b32_e32 v3, s11
+; GFX7-NEXT:    v_mov_b32_e32 v4, s10
+; GFX7-NEXT:    s_bfe_i32 s9, s4, 0x80010
+; GFX7-NEXT:    v_mov_b32_e32 v2, s12
+; GFX7-NEXT:    s_ashr_i32 s4, s4, 24
+; GFX7-NEXT:    v_mov_b32_e32 v1, s5
+; GFX7-NEXT:    v_mul_i32_i24_e32 v1, s4, v1
+; GFX7-NEXT:    v_mul_i32_i24_e32 v2, s9, v2
+; GFX7-NEXT:    v_mul_i32_i24_e32 v3, s7, v3
+; GFX7-NEXT:    v_mul_i32_i24_e32 v4, s6, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX7-NEXT:    v_and_b32_e32 v2, s8, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX7-NEXT:    v_and_b32_e32 v4, s8, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX7-NEXT:    v_or_b32_e32 v2, v4, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v3, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
+; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: idot4_acc16_vecMul:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ushort v2, v[0:1]
+; GFX8-NEXT:    s_load_dword s1, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s3, s1, 16
+; GFX8-NEXT:    s_bfe_i32 s6, s2, 0x80000
+; GFX8-NEXT:    s_lshr_b32 s4, s2, 16
+; GFX8-NEXT:    s_bfe_i32 s5, s1, 0x80000
+; GFX8-NEXT:    v_ashrrev_i16_e64 v4, 8, s1
+; GFX8-NEXT:    s_bfe_i32 s1, s3, 0x80000
+; GFX8-NEXT:    v_ashrrev_i16_e64 v6, 8, s3
+; GFX8-NEXT:    s_and_b32 s3, s0, s6
+; GFX8-NEXT:    v_ashrrev_i16_e64 v3, 8, s2
+; GFX8-NEXT:    s_bfe_i32 s2, s4, 0x80000
+; GFX8-NEXT:    v_ashrrev_i16_e64 v5, 8, s4
+; GFX8-NEXT:    s_and_b32 s4, s0, s5
+; GFX8-NEXT:    v_mov_b32_e32 v7, s3
+; GFX8-NEXT:    s_and_b32 s2, s0, s2
+; GFX8-NEXT:    s_and_b32 s0, s0, s1
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mad_u32_u24 v2, s4, v7, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, v4, v3, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s2
+; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, v6, v5, v2
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: idot4_acc16_vecMul:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, 0xffff
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_lshr_b32 s4, s2, 16
+; GFX9-NODL-NEXT:    s_lshr_b32 s5, s3, 16
+; GFX9-NODL-NEXT:    v_ashrrev_i16_e64 v3, 8, s5
+; GFX9-NODL-NEXT:    s_bfe_i32 s5, s5, 0x80000
+; GFX9-NODL-NEXT:    v_ashrrev_i16_e64 v2, 8, s4
+; GFX9-NODL-NEXT:    v_and_b32_e32 v5, s5, v4
+; GFX9-NODL-NEXT:    s_bfe_i32 s4, s4, 0x80000
+; GFX9-NODL-NEXT:    v_lshl_or_b32 v3, v3, 16, v5
+; GFX9-NODL-NEXT:    v_and_b32_e32 v5, s4, v4
+; GFX9-NODL-NEXT:    v_lshl_or_b32 v2, v2, 16, v5
+; GFX9-NODL-NEXT:    v_ashrrev_i16_e64 v1, 8, s3
+; GFX9-NODL-NEXT:    s_bfe_i32 s3, s3, 0x80000
+; GFX9-NODL-NEXT:    v_ashrrev_i16_e64 v0, 8, s2
+; GFX9-NODL-NEXT:    v_pk_mul_lo_u16 v2, v2, v3
+; GFX9-NODL-NEXT:    v_and_b32_e32 v3, s3, v4
+; GFX9-NODL-NEXT:    s_bfe_i32 s2, s2, 0x80000
+; GFX9-NODL-NEXT:    v_lshl_or_b32 v1, v1, 16, v3
+; GFX9-NODL-NEXT:    v_and_b32_e32 v3, s2, v4
+; GFX9-NODL-NEXT:    v_lshl_or_b32 v0, v0, 16, v3
+; GFX9-NODL-NEXT:    v_pk_mul_lo_u16 v3, v0, v1
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_load_ushort v4, v[0:1], off
+; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NODL-NEXT:    v_add_u32_e32 v4, v3, v4
+; GFX9-NODL-NEXT:    v_add_u32_sdwa v3, v4, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NODL-NEXT:    v_add_u32_e32 v3, v3, v2
+; GFX9-NODL-NEXT:    v_add_u32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NODL-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: idot4_acc16_vecMul:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, 0xffff
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_lshr_b32 s4, s2, 16
+; GFX9-DL-NEXT:    s_lshr_b32 s5, s3, 16
+; GFX9-DL-NEXT:    v_ashrrev_i16_e64 v3, 8, s5
+; GFX9-DL-NEXT:    s_bfe_i32 s5, s5, 0x80000
+; GFX9-DL-NEXT:    v_ashrrev_i16_e64 v2, 8, s4
+; GFX9-DL-NEXT:    v_and_b32_e32 v5, s5, v4
+; GFX9-DL-NEXT:    s_bfe_i32 s4, s4, 0x80000
+; GFX9-DL-NEXT:    v_lshl_or_b32 v3, v3, 16, v5
+; GFX9-DL-NEXT:    v_and_b32_e32 v5, s4, v4
+; GFX9-DL-NEXT:    v_lshl_or_b32 v2, v2, 16, v5
+; GFX9-DL-NEXT:    v_ashrrev_i16_e64 v1, 8, s3
+; GFX9-DL-NEXT:    s_bfe_i32 s3, s3, 0x80000
+; GFX9-DL-NEXT:    v_ashrrev_i16_e64 v0, 8, s2
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v2, v2, v3
+; GFX9-DL-NEXT:    v_and_b32_e32 v3, s3, v4
+; GFX9-DL-NEXT:    s_bfe_i32 s2, s2, 0x80000
+; GFX9-DL-NEXT:    v_lshl_or_b32 v1, v1, 16, v3
+; GFX9-DL-NEXT:    v_and_b32_e32 v3, s2, v4
+; GFX9-DL-NEXT:    v_lshl_or_b32 v0, v0, 16, v3
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v3, v0, v1
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ushort v4, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_add_u32_e32 v4, v3, v4
+; GFX9-DL-NEXT:    v_add_u32_sdwa v3, v4, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_add_u32_e32 v3, v3, v2
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                              <4 x i8> addrspace(1)* %src2,
+                                              i16 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
+  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
+
+  %cvec1 = sext <4 x i8> %vec1 to <4 x i16>
+  %cvec2 = sext <4 x i8> %vec2 to <4 x i16>
+
+  %mul = mul <4 x i16> %cvec1, %cvec2
+  %mul0 = extractelement <4 x i16> %mul, i64 0
+  %mul1 = extractelement <4 x i16> %mul, i64 1
+  %mul2 = extractelement <4 x i16> %mul, i64 2
+  %mul3 = extractelement <4 x i16> %mul, i64 3
+
+  %acc = load i16, i16 addrspace(1)* %dst, align 4
+  %add1 = add i16 %mul0, %acc
+  %add2 = add i16 %add1, %mul1
+  %add3 = add i16 %add2, %mul2
+  %add4 = add i16 %add3, %mul3
+
+  store i16 %add4, i16 addrspace(1)* %dst, align 4
+  ret void
+}
+
+; TODO: Support this pattern.
+define amdgpu_kernel void @udot4_acc8_vecMul(<4 x i8> addrspace(1)* %src1,
+; GFX7-LABEL: udot4_acc8_vecMul:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_movk_i32 s8, 0xff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX7-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0
+; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_bfe_u32 s6, s4, 0x80008
+; GFX7-NEXT:    s_lshr_b32 s7, s4, 16
+; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80008
+; GFX7-NEXT:    s_lshr_b32 s11, s5, 16
+; GFX7-NEXT:    s_lshr_b32 s12, s5, 24
+; GFX7-NEXT:    v_mov_b32_e32 v2, s11
+; GFX7-NEXT:    v_mov_b32_e32 v3, s10
+; GFX7-NEXT:    s_lshr_b32 s9, s4, 24
+; GFX7-NEXT:    v_mov_b32_e32 v1, s12
+; GFX7-NEXT:    s_mul_i32 s4, s4, s5
+; GFX7-NEXT:    v_mul_u32_u24_e32 v1, s9, v1
+; GFX7-NEXT:    v_mul_u32_u24_e32 v2, s7, v2
+; GFX7-NEXT:    v_mul_u32_u24_e32 v3, s6, v3
+; GFX7-NEXT:    s_and_b32 s4, s4, s8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
+; GFX7-NEXT:    v_and_b32_e32 v2, s8, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 8, v3
+; GFX7-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX7-NEXT:    v_or_b32_e32 v2, s4, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX7-NEXT:    v_and_b32_e32 v2, 0xffff, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 24, v1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v3, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
+; GFX7-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: udot4_acc8_vecMul:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ubyte v2, v[0:1]
+; GFX8-NEXT:    s_load_dword s1, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s3, s1, 24
+; GFX8-NEXT:    s_lshr_b32 s4, s2, 24
+; GFX8-NEXT:    s_and_b32 s6, s1, s0
+; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x80010
+; GFX8-NEXT:    s_and_b32 s0, s2, s0
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_mov_b32_e32 v4, s2
+; GFX8-NEXT:    v_mul_u32_u24_sdwa v3, v3, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:BYTE_1
+; GFX8-NEXT:    v_mov_b32_e32 v4, s0
+; GFX8-NEXT:    s_bfe_u32 s5, s1, 0x80010
+; GFX8-NEXT:    v_mov_b32_e32 v5, s7
+; GFX8-NEXT:    v_mov_b32_e32 v6, s4
+; GFX8-NEXT:    v_mov_b32_e32 v7, s3
+; GFX8-NEXT:    v_mul_u32_u24_e32 v4, s6, v4
+; GFX8-NEXT:    v_mul_u32_u24_e32 v5, s5, v5
+; GFX8-NEXT:    v_mul_u32_u24_sdwa v6, v7, v6 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v5, v5, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v3, v4, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v3, v3, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 8, v3
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v3
+; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v4, v2
+; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    flat_store_byte v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-NODL-LABEL: udot4_acc8_vecMul:
+; GFX9-NODL:       ; %bb.0: ; %entry
+; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-NODL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NODL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NODL-NEXT:    s_lshr_b32 s0, s2, 16
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s3
+; GFX9-NODL-NEXT:    s_lshr_b32 s1, s3, 16
+; GFX9-NODL-NEXT:    s_lshr_b32 s4, s3, 24
+; GFX9-NODL-NEXT:    v_mul_lo_u16_e32 v3, s2, v3
+; GFX9-NODL-NEXT:    v_mul_lo_u16_sdwa v4, s2, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:BYTE_1
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s1
+; GFX9-NODL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NODL-NEXT:    s_lshr_b32 s5, s2, 24
+; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s4
+; GFX9-NODL-NEXT:    v_mul_lo_u16_sdwa v4, s5, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NODL-NEXT:    v_mul_lo_u16_e32 v5, s0, v5
+; GFX9-NODL-NEXT:    v_or_b32_sdwa v4, v5, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NODL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX9-NODL-NEXT:    v_lshrrev_b32_e32 v4, 8, v3
+; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NODL-NEXT:    v_add_u32_e32 v2, v3, v2
+; GFX9-NODL-NEXT:    v_add_u32_e32 v2, v2, v4
+; GFX9-NODL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NODL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
+; GFX9-NODL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-NODL-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: udot4_acc8_vecMul:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_lshr_b32 s0, s2, 16
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s3
+; GFX9-DL-NEXT:    s_lshr_b32 s1, s3, 16
+; GFX9-DL-NEXT:    s_lshr_b32 s4, s3, 24
+; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v3, s2, v3
+; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v4, s2, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:BYTE_1
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s1
+; GFX9-DL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-DL-NEXT:    s_lshr_b32 s5, s2, 24
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s4
+; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v4, s5, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v5, s0, v5
+; GFX9-DL-NEXT:    v_or_b32_sdwa v4, v5, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-DL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v4, 8, v3
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v3, v2
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v4
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
+; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                             <4 x i8> addrspace(1)* %src2,
+                                             i8 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
+  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2
+
+  %mul = mul <4 x i8> %vec1, %vec2
+  %mul0 = extractelement <4 x i8> %mul, i64 0
+  %mul1 = extractelement <4 x i8> %mul, i64 1
+  %mul2 = extractelement <4 x i8> %mul, i64 2
+  %mul3 = extractelement <4 x i8> %mul, i64 3
+
+  %acc = load i8, i8 addrspace(1)* %dst, align 4
+  %add1 = add i8 %mul0, %acc
+  %add2 = add i8 %add1, %mul1
+  %add3 = add i8 %add2, %mul2
+  %add4 = add i8 %add3, %mul3
+
+  store i8 %add4, i8 addrspace(1)* %dst, align 4
+  ret void
+}

test/CodeGen/AMDGPU/idot8.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=amdgcn -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX7 %s
 ; RUN: llc -mtriple=amdgcn -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX8 %s
 ; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9 %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx906 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN-DL %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx906 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9-DL %s
 
 define amdgpu_kernel void @udot8_acc32(<8 x i4> addrspace(1)* %src1,
 ; GFX7-LABEL: udot8_acc32:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
 ; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
 ; GFX7-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX7-NEXT:    s_mov_b32 s6, -1

 ; GFX9-NEXT:    v_mad_u32_u24 v2, s6, v3, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s12
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s1, v3, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s11
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
 ; GFX9-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX9-NEXT:    s_endpgm
 ;
-; GCN-DL-LABEL: udot8_acc32:
-; GCN-DL:       ; %bb.0: ; %entry
-; GCN-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GCN-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
-; GCN-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
-; GCN-DL-NEXT:    s_load_dword s5, s[0:1], 0x0
-; GCN-DL-NEXT:    v_mov_b32_e32 v0, s0
-; GCN-DL-NEXT:    v_mov_b32_e32 v1, s1
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    v_mov_b32_e32 v2, s4
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s5
-; GCN-DL-NEXT:    v_dot8_u32_u4 v2, s2, v2, v3
-; GCN-DL-NEXT:    global_store_dword v[0:1], v2, off
-; GCN-DL-NEXT:    s_endpgm
+; GFX9-DL-LABEL: udot8_acc32:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s4
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-DL-NEXT:    v_dot8_u32_u4 v2, s2, v2, v3
+; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                        <8 x i4> addrspace(1)* %src2,
                                        i32 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
   %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
 
   %v1e0 = extractelement <8 x i4> %vec1, i64 0
   %cv1e0 = zext i4 %v1e0 to i32

 ; GFX9-NEXT:    v_mad_u32_u24 v2, s9, v7, v2
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s11, v8, v2
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s13, v9, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s14
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
 ; GFX9-NEXT:    global_store_short v[0:1], v2, off
 ; GFX9-NEXT:    s_endpgm
 ;
-; GCN-DL-LABEL: udot8_acc16:
-; GCN-DL:       ; %bb.0: ; %entry
-; GCN-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GCN-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
-; GCN-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
-; GCN-DL-NEXT:    v_mov_b32_e32 v0, s0
-; GCN-DL-NEXT:    v_mov_b32_e32 v1, s1
-; GCN-DL-NEXT:    global_load_ushort v2, v[0:1], off
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_and_b32 s0, s2, 15
-; GCN-DL-NEXT:    s_and_b32 s1, s4, 15
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s1
-; GCN-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s5
-; GCN-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
-; GCN-DL-NEXT:    s_bfe_u32 s5, s4, 0x40008
-; GCN-DL-NEXT:    s_bfe_u32 s8, s4, 0x40010
-; GCN-DL-NEXT:    s_bfe_u32 s10, s4, 0x40014
-; GCN-DL-NEXT:    s_bfe_u32 s12, s4, 0x40018
-; GCN-DL-NEXT:    s_lshr_b32 s14, s4, 28
-; GCN-DL-NEXT:    s_bfe_u32 s4, s4, 0x4000c
-; GCN-DL-NEXT:    s_bfe_u32 s6, s2, 0x40008
-; GCN-DL-NEXT:    v_mov_b32_e32 v5, s5
-; GCN-DL-NEXT:    s_bfe_u32 s7, s2, 0x4000c
-; GCN-DL-NEXT:    v_mov_b32_e32 v6, s4
-; GCN-DL-NEXT:    s_bfe_u32 s9, s2, 0x40010
-; GCN-DL-NEXT:    v_mov_b32_e32 v7, s8
-; GCN-DL-NEXT:    s_bfe_u32 s11, s2, 0x40014
-; GCN-DL-NEXT:    v_mov_b32_e32 v8, s10
-; GCN-DL-NEXT:    s_bfe_u32 s13, s2, 0x40018
-; GCN-DL-NEXT:    v_mov_b32_e32 v9, s12
-; GCN-DL-NEXT:    s_lshr_b32 s2, s2, 28
-; GCN-DL-NEXT:    s_waitcnt vmcnt(0)
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
-; GCN-DL-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s7, v6, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s9, v7, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s11, v8, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s13, v9, v2
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s14
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
-; GCN-DL-NEXT:    global_store_short v[0:1], v2, off
-; GCN-DL-NEXT:    s_endpgm
+; GFX9-DL-LABEL: udot8_acc16:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ushort v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_and_b32 s0, s2, 15
+; GFX9-DL-NEXT:    s_and_b32 s1, s4, 15
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s5
+; GFX9-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40008
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s4, 0x40010
+; GFX9-DL-NEXT:    s_bfe_u32 s10, s4, 0x40014
+; GFX9-DL-NEXT:    s_bfe_u32 s12, s4, 0x40018
+; GFX9-DL-NEXT:    s_lshr_b32 s14, s4, 28
+; GFX9-DL-NEXT:    s_bfe_u32 s4, s4, 0x4000c
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x40008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s5
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x4000c
+; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s4
+; GFX9-DL-NEXT:    s_bfe_u32 s9, s2, 0x40010
+; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s8
+; GFX9-DL-NEXT:    s_bfe_u32 s11, s2, 0x40014
+; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s10
+; GFX9-DL-NEXT:    s_bfe_u32 s13, s2, 0x40018
+; GFX9-DL-NEXT:    v_mov_b32_e32 v9, s12
+; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 28
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
+; GFX9-DL-NEXT:    v_and_b32_e32 v2, 0xffff, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s7, v6, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s9, v7, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s11, v8, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s13, v9, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s14
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX9-DL-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                        <8 x i4> addrspace(1)* %src2,
                                        i16 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
   %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
 
   %v1e0 = extractelement <8 x i4> %vec1, i64 0
   %cv1e0 = zext i4 %v1e0 to i16

 ; GFX9-NEXT:    v_mad_u32_u24 v2, s9, v7, v2
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s11, v8, v2
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s13, v9, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s14
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
 ; GFX9-NEXT:    global_store_byte v[0:1], v2, off
 ; GFX9-NEXT:    s_endpgm
 ;
-; GCN-DL-LABEL: udot8_acc8:
-; GCN-DL:       ; %bb.0: ; %entry
-; GCN-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GCN-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
-; GCN-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
-; GCN-DL-NEXT:    v_mov_b32_e32 v0, s0
-; GCN-DL-NEXT:    v_mov_b32_e32 v1, s1
-; GCN-DL-NEXT:    global_load_ubyte v2, v[0:1], off
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_and_b32 s0, s2, 15
-; GCN-DL-NEXT:    s_and_b32 s1, s4, 15
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s1
-; GCN-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s5
-; GCN-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
-; GCN-DL-NEXT:    s_bfe_u32 s5, s4, 0x40008
-; GCN-DL-NEXT:    s_bfe_u32 s8, s4, 0x40010
-; GCN-DL-NEXT:    s_bfe_u32 s10, s4, 0x40014
-; GCN-DL-NEXT:    s_bfe_u32 s12, s4, 0x40018
-; GCN-DL-NEXT:    s_lshr_b32 s14, s4, 28
-; GCN-DL-NEXT:    s_bfe_u32 s4, s4, 0x4000c
-; GCN-DL-NEXT:    s_bfe_u32 s6, s2, 0x40008
-; GCN-DL-NEXT:    v_mov_b32_e32 v5, s5
-; GCN-DL-NEXT:    s_bfe_u32 s7, s2, 0x4000c
-; GCN-DL-NEXT:    v_mov_b32_e32 v6, s4
-; GCN-DL-NEXT:    s_bfe_u32 s9, s2, 0x40010
-; GCN-DL-NEXT:    v_mov_b32_e32 v7, s8
-; GCN-DL-NEXT:    s_bfe_u32 s11, s2, 0x40014
-; GCN-DL-NEXT:    v_mov_b32_e32 v8, s10
-; GCN-DL-NEXT:    s_bfe_u32 s13, s2, 0x40018
-; GCN-DL-NEXT:    v_mov_b32_e32 v9, s12
-; GCN-DL-NEXT:    s_lshr_b32 s2, s2, 28
-; GCN-DL-NEXT:    s_waitcnt vmcnt(0)
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
-; GCN-DL-NEXT:    v_and_b32_e32 v2, 0xff, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s7, v6, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s9, v7, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s11, v8, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s13, v9, v2
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s14
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
-; GCN-DL-NEXT:    global_store_byte v[0:1], v2, off
-; GCN-DL-NEXT:    s_endpgm
+; GFX9-DL-LABEL: udot8_acc8:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_and_b32 s0, s2, 15
+; GFX9-DL-NEXT:    s_and_b32 s1, s4, 15
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s5
+; GFX9-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40008
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s4, 0x40010
+; GFX9-DL-NEXT:    s_bfe_u32 s10, s4, 0x40014
+; GFX9-DL-NEXT:    s_bfe_u32 s12, s4, 0x40018
+; GFX9-DL-NEXT:    s_lshr_b32 s14, s4, 28
+; GFX9-DL-NEXT:    s_bfe_u32 s4, s4, 0x4000c
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x40008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s5
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x4000c
+; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s4
+; GFX9-DL-NEXT:    s_bfe_u32 s9, s2, 0x40010
+; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s8
+; GFX9-DL-NEXT:    s_bfe_u32 s11, s2, 0x40014
+; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s10
+; GFX9-DL-NEXT:    s_bfe_u32 s13, s2, 0x40018
+; GFX9-DL-NEXT:    v_mov_b32_e32 v9, s12
+; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 28
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
+; GFX9-DL-NEXT:    v_and_b32_e32 v2, 0xff, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s7, v6, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s9, v7, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s11, v8, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s13, v9, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s14
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                       <8 x i4> addrspace(1)* %src2,
                                       i8 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
   %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
 
   %v1e0 = extractelement <8 x i4> %vec1, i64 0
   %cv1e0 = zext i4 %v1e0 to i8

 ; GFX9-NEXT:    v_mad_u32_u24 v2, s7, v8, v2
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s8, v9, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s4
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
 ; GFX9-NEXT:    v_and_b32_e32 v2, 15, v2
 ; GFX9-NEXT:    global_store_byte v[0:1], v2, off
 ; GFX9-NEXT:    s_endpgm
 ;
-; GCN-DL-LABEL: udot8_acc4:
-; GCN-DL:       ; %bb.0: ; %entry
-; GCN-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GCN-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
-; GCN-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
-; GCN-DL-NEXT:    v_mov_b32_e32 v0, s0
-; GCN-DL-NEXT:    v_mov_b32_e32 v1, s1
-; GCN-DL-NEXT:    global_load_ubyte v2, v[0:1], off
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_and_b32 s0, s2, 15
-; GCN-DL-NEXT:    s_and_b32 s1, s4, 15
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s1
-; GCN-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
-; GCN-DL-NEXT:    s_bfe_u32 s6, s4, 0x40008
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s6
-; GCN-DL-NEXT:    s_bfe_u32 s7, s2, 0x40008
-; GCN-DL-NEXT:    v_mov_b32_e32 v5, s5
-; GCN-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
-; GCN-DL-NEXT:    v_mul_u32_u24_e32 v4, s7, v4
-; GCN-DL-NEXT:    s_bfe_u32 s5, s4, 0x4000c
-; GCN-DL-NEXT:    v_and_b32_e32 v4, 15, v4
-; GCN-DL-NEXT:    s_bfe_u32 s7, s4, 0x40010
-; GCN-DL-NEXT:    v_mov_b32_e32 v6, s5
-; GCN-DL-NEXT:    s_bfe_u32 s6, s2, 0x4000c
-; GCN-DL-NEXT:    s_bfe_u32 s8, s4, 0x40014
-; GCN-DL-NEXT:    v_mov_b32_e32 v7, s7
-; GCN-DL-NEXT:    s_bfe_u32 s5, s2, 0x40010
-; GCN-DL-NEXT:    s_bfe_u32 s9, s4, 0x40018
-; GCN-DL-NEXT:    v_mov_b32_e32 v8, s8
-; GCN-DL-NEXT:    s_bfe_u32 s7, s2, 0x40014
-; GCN-DL-NEXT:    s_bfe_u32 s8, s2, 0x40018
-; GCN-DL-NEXT:    s_lshr_b32 s4, s4, 28
-; GCN-DL-NEXT:    v_mov_b32_e32 v9, s9
-; GCN-DL-NEXT:    s_lshr_b32 s2, s2, 28
-; GCN-DL-NEXT:    s_waitcnt vmcnt(0)
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s1, v5, v2
-; GCN-DL-NEXT:    v_and_b32_e32 v2, 15, v2
-; GCN-DL-NEXT:    v_add_u32_e32 v2, v2, v4
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s5, v7, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s7, v8, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s8, v9, v2
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s4
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
-; GCN-DL-NEXT:    v_and_b32_e32 v2, 15, v2
-; GCN-DL-NEXT:    global_store_byte v[0:1], v2, off
-; GCN-DL-NEXT:    s_endpgm
+; GFX9-DL-LABEL: udot8_acc4:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_and_b32 s0, s2, 15
+; GFX9-DL-NEXT:    s_and_b32 s1, s4, 15
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s4, 0x40008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s6
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x40008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s5
+; GFX9-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
+; GFX9-DL-NEXT:    v_mul_u32_u24_e32 v4, s7, v4
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x4000c
+; GFX9-DL-NEXT:    v_and_b32_e32 v4, 15, v4
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s4, 0x40010
+; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s5
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x4000c
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s4, 0x40014
+; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s7
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s2, 0x40010
+; GFX9-DL-NEXT:    s_bfe_u32 s9, s4, 0x40018
+; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s8
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x40014
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s2, 0x40018
+; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 28
+; GFX9-DL-NEXT:    v_mov_b32_e32 v9, s9
+; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 28
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s1, v5, v2
+; GFX9-DL-NEXT:    v_and_b32_e32 v2, 15, v2
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v4
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s5, v7, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s7, v8, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s8, v9, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX9-DL-NEXT:    v_and_b32_e32 v2, 15, v2
+; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                       <8 x i4> addrspace(1)* %src2,
                                       i4 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
   %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
 
   %v1e0 = extractelement <8 x i4> %vec1, i64 0
   %v2e0 = extractelement <8 x i4> %vec2, i64 0

 ; GFX9-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s4
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
 ; GFX9-NEXT:    v_and_b32_e32 v2, 15, v2
 ; GFX9-NEXT:    global_store_byte v[0:1], v2, off
 ; GFX9-NEXT:    s_endpgm
 ;
-; GCN-DL-LABEL: udot8_CommutationInsideMAD:
-; GCN-DL:       ; %bb.0: ; %entry
-; GCN-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GCN-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
-; GCN-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
-; GCN-DL-NEXT:    v_mov_b32_e32 v0, s0
-; GCN-DL-NEXT:    v_mov_b32_e32 v1, s1
-; GCN-DL-NEXT:    global_load_ubyte v2, v[0:1], off
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_and_b32 s0, s2, 15
-; GCN-DL-NEXT:    s_and_b32 s1, s4, 15
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s1
-; GCN-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
-; GCN-DL-NEXT:    s_bfe_u32 s6, s4, 0x40008
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s5
-; GCN-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
-; GCN-DL-NEXT:    s_bfe_u32 s7, s4, 0x4000c
-; GCN-DL-NEXT:    v_mov_b32_e32 v5, s6
-; GCN-DL-NEXT:    s_bfe_u32 s5, s2, 0x40008
-; GCN-DL-NEXT:    s_bfe_u32 s8, s4, 0x40010
-; GCN-DL-NEXT:    v_mov_b32_e32 v6, s7
-; GCN-DL-NEXT:    s_bfe_u32 s6, s2, 0x4000c
-; GCN-DL-NEXT:    s_bfe_u32 s9, s4, 0x40014
-; GCN-DL-NEXT:    v_mov_b32_e32 v7, s8
-; GCN-DL-NEXT:    s_bfe_u32 s7, s2, 0x40010
-; GCN-DL-NEXT:    s_bfe_u32 s10, s4, 0x40018
-; GCN-DL-NEXT:    v_mov_b32_e32 v8, s9
-; GCN-DL-NEXT:    s_bfe_u32 s8, s2, 0x40014
-; GCN-DL-NEXT:    s_bfe_u32 s9, s2, 0x40018
-; GCN-DL-NEXT:    s_lshr_b32 s4, s4, 28
-; GCN-DL-NEXT:    v_mov_b32_e32 v9, s10
-; GCN-DL-NEXT:    s_lshr_b32 s2, s2, 28
-; GCN-DL-NEXT:    s_waitcnt vmcnt(0)
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s5, v5, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s7, v7, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s4
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
-; GCN-DL-NEXT:    v_and_b32_e32 v2, 15, v2
-; GCN-DL-NEXT:    global_store_byte v[0:1], v2, off
-; GCN-DL-NEXT:    s_endpgm
+; GFX9-DL-LABEL: udot8_CommutationInsideMAD:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_and_b32 s0, s2, 15
+; GFX9-DL-NEXT:    s_and_b32 s1, s4, 15
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s4, 0x40008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s5
+; GFX9-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s4, 0x4000c
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s6
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s2, 0x40008
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s4, 0x40010
+; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s7
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x4000c
+; GFX9-DL-NEXT:    s_bfe_u32 s9, s4, 0x40014
+; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s8
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x40010
+; GFX9-DL-NEXT:    s_bfe_u32 s10, s4, 0x40018
+; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s9
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s2, 0x40014
+; GFX9-DL-NEXT:    s_bfe_u32 s9, s2, 0x40018
+; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 28
+; GFX9-DL-NEXT:    v_mov_b32_e32 v9, s10
+; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 28
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s5, v5, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s7, v7, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX9-DL-NEXT:    v_and_b32_e32 v2, 15, v2
+; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                                       <8 x i4> addrspace(1)* %src2,
                                                       i4 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
   %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
 
   %v1e0 = extractelement <8 x i4> %vec1, i64 0
   %v2e0 = extractelement <8 x i4> %vec2, i64 0

 ; GFX9-NEXT:    v_mov_b32_e32 v4, s12
 ; GFX9-NEXT:    v_mad_u32_u24 v3, s1, v4, v3
 ; GFX9-NEXT:    v_mov_b32_e32 v4, s11
 ; GFX9-NEXT:    v_mad_u32_u24 v3, s0, v4, v3
 ; GFX9-NEXT:    v_add_u32_e32 v2, v2, v3
 ; GFX9-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX9-NEXT:    s_endpgm
 ;
-; GCN-DL-LABEL: udot8_multiuses_mul1:
-; GCN-DL:       ; %bb.0: ; %entry
-; GCN-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GCN-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
-; GCN-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
-; GCN-DL-NEXT:    s_load_dword s5, s[0:1], 0x0
-; GCN-DL-NEXT:    v_mov_b32_e32 v0, s0
-; GCN-DL-NEXT:    v_mov_b32_e32 v1, s1
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_lshr_b32 s0, s2, 28
-; GCN-DL-NEXT:    s_bfe_u32 s17, s4, 0x40004
-; GCN-DL-NEXT:    s_lshr_b32 s11, s4, 28
-; GCN-DL-NEXT:    s_bfe_u32 s12, s4, 0x40018
-; GCN-DL-NEXT:    s_bfe_u32 s13, s4, 0x40014
-; GCN-DL-NEXT:    s_bfe_u32 s14, s4, 0x40010
-; GCN-DL-NEXT:    s_bfe_u32 s15, s4, 0x4000c
-; GCN-DL-NEXT:    s_bfe_u32 s16, s4, 0x40008
-; GCN-DL-NEXT:    s_and_b32 s4, s4, 15
-; GCN-DL-NEXT:    s_bfe_u32 s1, s2, 0x40018
-; GCN-DL-NEXT:    s_bfe_u32 s6, s2, 0x40014
-; GCN-DL-NEXT:    s_bfe_u32 s7, s2, 0x40010
-; GCN-DL-NEXT:    s_bfe_u32 s8, s2, 0x4000c
-; GCN-DL-NEXT:    s_bfe_u32 s9, s2, 0x40008
-; GCN-DL-NEXT:    s_bfe_u32 s10, s2, 0x40004
-; GCN-DL-NEXT:    s_and_b32 s2, s2, 15
-; GCN-DL-NEXT:    v_mov_b32_e32 v2, s4
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s5
-; GCN-DL-NEXT:    v_mad_u32_u24 v3, s2, v2, v3
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s17
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s2, v2, v3
-; GCN-DL-NEXT:    v_mad_u32_u24 v3, s10, v4, v3
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s16
-; GCN-DL-NEXT:    v_mad_u32_u24 v3, s9, v4, v3
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s15
-; GCN-DL-NEXT:    v_mad_u32_u24 v3, s8, v4, v3
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s14
-; GCN-DL-NEXT:    v_mad_u32_u24 v3, s7, v4, v3
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s13
-; GCN-DL-NEXT:    v_mad_u32_u24 v3, s6, v4, v3
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s12
-; GCN-DL-NEXT:    v_mad_u32_u24 v3, s1, v4, v3
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s11
-; GCN-DL-NEXT:    v_mad_u32_u24 v3, s0, v4, v3
-; GCN-DL-NEXT:    v_add_u32_e32 v2, v2, v3
-; GCN-DL-NEXT:    global_store_dword v[0:1], v2, off
-; GCN-DL-NEXT:    s_endpgm
+; GFX9-DL-LABEL: udot8_multiuses_mul1:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_lshr_b32 s0, s2, 28
+; GFX9-DL-NEXT:    s_bfe_u32 s17, s4, 0x40004
+; GFX9-DL-NEXT:    s_lshr_b32 s11, s4, 28
+; GFX9-DL-NEXT:    s_bfe_u32 s12, s4, 0x40018
+; GFX9-DL-NEXT:    s_bfe_u32 s13, s4, 0x40014
+; GFX9-DL-NEXT:    s_bfe_u32 s14, s4, 0x40010
+; GFX9-DL-NEXT:    s_bfe_u32 s15, s4, 0x4000c
+; GFX9-DL-NEXT:    s_bfe_u32 s16, s4, 0x40008
+; GFX9-DL-NEXT:    s_and_b32 s4, s4, 15
+; GFX9-DL-NEXT:    s_bfe_u32 s1, s2, 0x40018
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x40014
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x40010
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s2, 0x4000c
+; GFX9-DL-NEXT:    s_bfe_u32 s9, s2, 0x40008
+; GFX9-DL-NEXT:    s_bfe_u32 s10, s2, 0x40004
+; GFX9-DL-NEXT:    s_and_b32 s2, s2, 15
+; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s4
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-DL-NEXT:    v_mad_u32_u24 v3, s2, v2, v3
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s17
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v2, v3
+; GFX9-DL-NEXT:    v_mad_u32_u24 v3, s10, v4, v3
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s16
+; GFX9-DL-NEXT:    v_mad_u32_u24 v3, s9, v4, v3
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s15
+; GFX9-DL-NEXT:    v_mad_u32_u24 v3, s8, v4, v3
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s14
+; GFX9-DL-NEXT:    v_mad_u32_u24 v3, s7, v4, v3
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s13
+; GFX9-DL-NEXT:    v_mad_u32_u24 v3, s6, v4, v3
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s12
+; GFX9-DL-NEXT:    v_mad_u32_u24 v3, s1, v4, v3
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s11
+; GFX9-DL-NEXT:    v_mad_u32_u24 v3, s0, v4, v3
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v3
+; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                                 <8 x i4> addrspace(1)* %src2,
                                                 i32 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
   %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
 
   %v1e0 = extractelement <8 x i4> %vec1, i64 0
   %cv1e0 = zext i4 %v1e0 to i32

 ; GFX9-NEXT:    v_mad_u32_u24 v2, s6, v3, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s12
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s1, v3, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s11
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
 ; GFX9-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX9-NEXT:    s_endpgm
 ;
-; GCN-DL-LABEL: udot8_acc32_vecMul:
-; GCN-DL:       ; %bb.0: ; %entry
-; GCN-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GCN-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
-; GCN-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
-; GCN-DL-NEXT:    s_load_dword s5, s[0:1], 0x0
-; GCN-DL-NEXT:    v_mov_b32_e32 v0, s0
-; GCN-DL-NEXT:    v_mov_b32_e32 v1, s1
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    v_mov_b32_e32 v2, s4
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s5
-; GCN-DL-NEXT:    v_dot8_u32_u4 v2, s2, v2, v3
-; GCN-DL-NEXT:    global_store_dword v[0:1], v2, off
-; GCN-DL-NEXT:    s_endpgm
+; GFX9-DL-LABEL: udot8_acc32_vecMul:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s4
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-DL-NEXT:    v_dot8_u32_u4 v2, s2, v2, v3
+; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                               <8 x i4> addrspace(1)* %src2,
                                               i32 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
   %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
 
   %cvec1 = zext <8 x i4> %vec1 to <8 x i32>
   %cvec2 = zext <8 x i4> %vec2 to <8 x i32>

 ; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
 ; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NEXT:    v_add_u32_e32 v2, v2, v6
 ; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NEXT:    global_store_short v[0:1], v2, off
 ; GFX9-NEXT:    s_endpgm
 ;
-; GCN-DL-LABEL: udot8_acc16_vecMul:
-; GCN-DL:       ; %bb.0: ; %entry
-; GCN-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GCN-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
-; GCN-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
-; GCN-DL-NEXT:    v_mov_b32_e32 v0, s0
-; GCN-DL-NEXT:    v_mov_b32_e32 v1, s1
-; GCN-DL-NEXT:    global_load_ushort v2, v[0:1], off
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_and_b32 s0, s2, 15
-; GCN-DL-NEXT:    s_and_b32 s1, s4, 15
-; GCN-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
-; GCN-DL-NEXT:    s_pack_ll_b32_b16 s1, s1, s5
-; GCN-DL-NEXT:    s_bfe_u32 s6, s2, 0x40004
-; GCN-DL-NEXT:    s_pack_ll_b32_b16 s0, s0, s6
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s1
-; GCN-DL-NEXT:    s_bfe_u32 s5, s4, 0x40008
-; GCN-DL-NEXT:    s_bfe_u32 s7, s4, 0x4000c
-; GCN-DL-NEXT:    s_pack_ll_b32_b16 s5, s5, s7
-; GCN-DL-NEXT:    v_pk_mul_lo_u16 v3, s0, v3
-; GCN-DL-NEXT:    s_bfe_u32 s1, s2, 0x40008
-; GCN-DL-NEXT:    s_bfe_u32 s6, s2, 0x4000c
-; GCN-DL-NEXT:    s_pack_ll_b32_b16 s1, s1, s6
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s5
-; GCN-DL-NEXT:    s_bfe_u32 s0, s4, 0x40010
-; GCN-DL-NEXT:    s_bfe_u32 s7, s4, 0x40014
-; GCN-DL-NEXT:    s_pack_ll_b32_b16 s0, s0, s7
-; GCN-DL-NEXT:    v_pk_mul_lo_u16 v4, s1, v4
-; GCN-DL-NEXT:    s_bfe_u32 s5, s2, 0x40010
-; GCN-DL-NEXT:    s_bfe_u32 s6, s2, 0x40014
-; GCN-DL-NEXT:    s_bfe_u32 s1, s4, 0x40018
-; GCN-DL-NEXT:    s_lshr_b32 s4, s4, 28
-; GCN-DL-NEXT:    v_mov_b32_e32 v5, s0
-; GCN-DL-NEXT:    s_pack_ll_b32_b16 s5, s5, s6
-; GCN-DL-NEXT:    s_bfe_u32 s0, s2, 0x40018
-; GCN-DL-NEXT:    s_lshr_b32 s2, s2, 28
-; GCN-DL-NEXT:    s_pack_ll_b32_b16 s1, s1, s4
-; GCN-DL-NEXT:    v_pk_mul_lo_u16 v5, s5, v5
-; GCN-DL-NEXT:    s_pack_ll_b32_b16 s0, s0, s2
-; GCN-DL-NEXT:    v_mov_b32_e32 v6, s1
-; GCN-DL-NEXT:    v_pk_mul_lo_u16 v6, s0, v6
-; GCN-DL-NEXT:    s_waitcnt vmcnt(0)
-; GCN-DL-NEXT:    v_add_u32_e32 v2, v3, v2
-; GCN-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GCN-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:BYTE_0
-; GCN-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GCN-DL-NEXT:    v_add_u32_e32 v2, v2, v5
-; GCN-DL-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GCN-DL-NEXT:    v_add_u32_e32 v2, v2, v6
-; GCN-DL-NEXT:    v_add_u32_sdwa v2, v2, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GCN-DL-NEXT:    global_store_short v[0:1], v2, off
-; GCN-DL-NEXT:    s_endpgm
+; GFX9-DL-LABEL: udot8_acc16_vecMul:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ushort v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_and_b32 s0, s2, 15
+; GFX9-DL-NEXT:    s_and_b32 s1, s4, 15
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s1, s1, s5
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x40004
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s0, s0, s6
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40008
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s4, 0x4000c
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s5, s5, s7
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v3, s0, v3
+; GFX9-DL-NEXT:    s_bfe_u32 s1, s2, 0x40008
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x4000c
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s1, s1, s6
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s5
+; GFX9-DL-NEXT:    s_bfe_u32 s0, s4, 0x40010
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s4, 0x40014
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s0, s0, s7
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v4, s1, v4
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s2, 0x40010
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x40014
+; GFX9-DL-NEXT:    s_bfe_u32 s1, s4, 0x40018
+; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 28
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s0
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s5, s5, s6
+; GFX9-DL-NEXT:    s_bfe_u32 s0, s2, 0x40018
+; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 28
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s1, s1, s4
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v5, s5, v5
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s0, s0, s2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s1
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v6, s0, v6
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v3, v2
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:BYTE_0
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v5
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v6
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                               <8 x i4> addrspace(1)* %src2,
                                               i16 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
   %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
 
   %cvec1 = zext <8 x i4> %vec1 to <8 x i16>
   %cvec2 = zext <8 x i4> %vec2 to <8 x i16>

 ; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 8, v4
 ; GFX9-NEXT:    v_add_u32_e32 v2, v2, v3
 ; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
 ; GFX9-NEXT:    global_store_byte v[0:1], v2, off
 ; GFX9-NEXT:    s_endpgm
 ;
-; GCN-DL-LABEL: udot8_acc8_vecMul:
-; GCN-DL:       ; %bb.0: ; %entry
-; GCN-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GCN-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
-; GCN-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
-; GCN-DL-NEXT:    v_mov_b32_e32 v0, s0
-; GCN-DL-NEXT:    v_mov_b32_e32 v1, s1
-; GCN-DL-NEXT:    global_load_ubyte v2, v[0:1], off
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_and_b32 s0, s2, 15
-; GCN-DL-NEXT:    s_and_b32 s1, s4, 15
-; GCN-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
-; GCN-DL-NEXT:    s_bfe_u32 s6, s4, 0x40008
-; GCN-DL-NEXT:    s_bfe_u32 s7, s4, 0x4000c
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s1
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s5
-; GCN-DL-NEXT:    s_bfe_u32 s8, s2, 0x40004
-; GCN-DL-NEXT:    v_mov_b32_e32 v5, s6
-; GCN-DL-NEXT:    s_bfe_u32 s9, s2, 0x40008
-; GCN-DL-NEXT:    v_mov_b32_e32 v6, s7
-; GCN-DL-NEXT:    s_bfe_u32 s10, s2, 0x4000c
-; GCN-DL-NEXT:    v_mul_lo_u16_e32 v3, s0, v3
-; GCN-DL-NEXT:    v_mul_lo_u16_sdwa v4, s8, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; GCN-DL-NEXT:    v_mul_lo_u16_e32 v5, s9, v5
-; GCN-DL-NEXT:    v_mul_lo_u16_sdwa v6, s10, v6 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; GCN-DL-NEXT:    v_or_b32_e32 v3, v3, v4
-; GCN-DL-NEXT:    v_or_b32_sdwa v4, v5, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; GCN-DL-NEXT:    s_bfe_u32 s1, s4, 0x40014
-; GCN-DL-NEXT:    s_bfe_u32 s5, s4, 0x40018
-; GCN-DL-NEXT:    s_bfe_u32 s0, s4, 0x40010
-; GCN-DL-NEXT:    s_lshr_b32 s4, s4, 28
-; GCN-DL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
-; GCN-DL-NEXT:    s_bfe_u32 s6, s2, 0x40010
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-DL-NEXT:    s_bfe_u32 s7, s2, 0x40014
-; GCN-DL-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-DL-NEXT:    s_bfe_u32 s8, s2, 0x40018
-; GCN-DL-NEXT:    v_mov_b32_e32 v6, s5
-; GCN-DL-NEXT:    s_lshr_b32 s2, s2, 28
-; GCN-DL-NEXT:    v_mov_b32_e32 v7, s4
-; GCN-DL-NEXT:    v_mul_lo_u16_e32 v4, s6, v4
-; GCN-DL-NEXT:    v_mul_lo_u16_sdwa v5, s7, v5 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; GCN-DL-NEXT:    v_mul_lo_u16_e32 v6, s8, v6
-; GCN-DL-NEXT:    v_mul_lo_u16_sdwa v7, s2, v7 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; GCN-DL-NEXT:    v_or_b32_e32 v4, v4, v5
-; GCN-DL-NEXT:    v_or_b32_sdwa v5, v6, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; GCN-DL-NEXT:    v_or_b32_sdwa v4, v4, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
-; GCN-DL-NEXT:    v_lshrrev_b32_e32 v5, 8, v3
-; GCN-DL-NEXT:    s_waitcnt vmcnt(0)
-; GCN-DL-NEXT:    v_add_u32_e32 v2, v3, v2
-; GCN-DL-NEXT:    v_add_u32_e32 v2, v2, v5
-; GCN-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_2
-; GCN-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
-; GCN-DL-NEXT:    v_add_u32_e32 v2, v2, v4
-; GCN-DL-NEXT:    v_lshrrev_b32_e32 v3, 8, v4
-; GCN-DL-NEXT:    v_add_u32_e32 v2, v2, v3
-; GCN-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GCN-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
-; GCN-DL-NEXT:    global_store_byte v[0:1], v2, off
-; GCN-DL-NEXT:    s_endpgm
+; GFX9-DL-LABEL: udot8_acc8_vecMul:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_and_b32 s0, s2, 15
+; GFX9-DL-NEXT:    s_and_b32 s1, s4, 15
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s4, 0x40008
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s4, 0x4000c
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s5
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s2, 0x40004
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s6
+; GFX9-DL-NEXT:    s_bfe_u32 s9, s2, 0x40008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s7
+; GFX9-DL-NEXT:    s_bfe_u32 s10, s2, 0x4000c
+; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v3, s0, v3
+; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v4, s8, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v5, s9, v5
+; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v6, s10, v6 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-DL-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX9-DL-NEXT:    v_or_b32_sdwa v4, v5, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-DL-NEXT:    s_bfe_u32 s1, s4, 0x40014
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40018
+; GFX9-DL-NEXT:    s_bfe_u32 s0, s4, 0x40010
+; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 28
+; GFX9-DL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x40010
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s0
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x40014
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s1
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s2, 0x40018
+; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s5
+; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 28
+; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s4
+; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v4, s6, v4
+; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v5, s7, v5 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v6, s8, v6
+; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v7, s2, v7 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-DL-NEXT:    v_or_b32_e32 v4, v4, v5
+; GFX9-DL-NEXT:    v_or_b32_sdwa v5, v6, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-DL-NEXT:    v_or_b32_sdwa v4, v4, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v5, 8, v3
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v3, v2
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v5
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_2
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v4
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v3, 8, v4
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v3
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
+; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                              <8 x i4> addrspace(1)* %src2,
                                              i8 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
   %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
 
   %cvec1 = zext <8 x i4> %vec1 to <8 x i8>
   %cvec2 = zext <8 x i4> %vec2 to <8 x i8>

 ; GFX9-NEXT:    v_mad_u32_u24 v2, s7, v8, v2
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s8, v9, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s4
 ; GFX9-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
 ; GFX9-NEXT:    v_and_b32_e32 v2, 15, v2
 ; GFX9-NEXT:    global_store_byte v[0:1], v2, off
 ; GFX9-NEXT:    s_endpgm
 ;
-; GCN-DL-LABEL: udot8_acc4_vecMul:
-; GCN-DL:       ; %bb.0: ; %entry
-; GCN-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GCN-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
-; GCN-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
-; GCN-DL-NEXT:    v_mov_b32_e32 v0, s0
-; GCN-DL-NEXT:    v_mov_b32_e32 v1, s1
-; GCN-DL-NEXT:    global_load_ubyte v2, v[0:1], off
-; GCN-DL-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-DL-NEXT:    s_and_b32 s0, s2, 15
-; GCN-DL-NEXT:    s_and_b32 s1, s4, 15
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s1
-; GCN-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
-; GCN-DL-NEXT:    s_bfe_u32 s6, s4, 0x40008
-; GCN-DL-NEXT:    v_mov_b32_e32 v4, s6
-; GCN-DL-NEXT:    s_bfe_u32 s7, s2, 0x40008
-; GCN-DL-NEXT:    v_mov_b32_e32 v5, s5
-; GCN-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
-; GCN-DL-NEXT:    v_mul_u32_u24_e32 v4, s7, v4
-; GCN-DL-NEXT:    s_bfe_u32 s5, s4, 0x4000c
-; GCN-DL-NEXT:    v_and_b32_e32 v4, 15, v4
-; GCN-DL-NEXT:    s_bfe_u32 s7, s4, 0x40010
-; GCN-DL-NEXT:    v_mov_b32_e32 v6, s5
-; GCN-DL-NEXT:    s_bfe_u32 s6, s2, 0x4000c
-; GCN-DL-NEXT:    s_bfe_u32 s8, s4, 0x40014
-; GCN-DL-NEXT:    v_mov_b32_e32 v7, s7
-; GCN-DL-NEXT:    s_bfe_u32 s5, s2, 0x40010
-; GCN-DL-NEXT:    s_bfe_u32 s9, s4, 0x40018
-; GCN-DL-NEXT:    v_mov_b32_e32 v8, s8
-; GCN-DL-NEXT:    s_bfe_u32 s7, s2, 0x40014
-; GCN-DL-NEXT:    s_bfe_u32 s8, s2, 0x40018
-; GCN-DL-NEXT:    s_lshr_b32 s4, s4, 28
-; GCN-DL-NEXT:    v_mov_b32_e32 v9, s9
-; GCN-DL-NEXT:    s_lshr_b32 s2, s2, 28
-; GCN-DL-NEXT:    s_waitcnt vmcnt(0)
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s1, v5, v2
-; GCN-DL-NEXT:    v_and_b32_e32 v2, 15, v2
-; GCN-DL-NEXT:    v_add_u32_e32 v2, v2, v4
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s5, v7, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s7, v8, v2
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s8, v9, v2
-; GCN-DL-NEXT:    v_mov_b32_e32 v3, s4
-; GCN-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
-; GCN-DL-NEXT:    v_and_b32_e32 v2, 15, v2
-; GCN-DL-NEXT:    global_store_byte v[0:1], v2, off
-; GCN-DL-NEXT:    s_endpgm
+; GFX9-DL-LABEL: udot8_acc4_vecMul:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_and_b32 s0, s2, 15
+; GFX9-DL-NEXT:    s_and_b32 s1, s4, 15
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s4, 0x40008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s6
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x40008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s5
+; GFX9-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
+; GFX9-DL-NEXT:    v_mul_u32_u24_e32 v4, s7, v4
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x4000c
+; GFX9-DL-NEXT:    v_and_b32_e32 v4, 15, v4
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s4, 0x40010
+; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s5
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x4000c
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s4, 0x40014
+; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s7
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s2, 0x40010
+; GFX9-DL-NEXT:    s_bfe_u32 s9, s4, 0x40018
+; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s8
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x40014
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s2, 0x40018
+; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 28
+; GFX9-DL-NEXT:    v_mov_b32_e32 v9, s9
+; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 28
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s1, v5, v2
+; GFX9-DL-NEXT:    v_and_b32_e32 v2, 15, v2
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v4
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s5, v7, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s7, v8, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s8, v9, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
+; GFX9-DL-NEXT:    v_and_b32_e32 v2, 15, v2
+; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
                                              <8 x i4> addrspace(1)* %src2,
                                              i4 addrspace(1)* nocapture %dst) {
 entry:
   %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
   %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
 
   %mul = mul <8 x i4> %vec1, %vec2
   %mul0 = extractelement <8 x i4> %mul, i64 0
entry:
   %add5 = add i4 %add4, %mul4
   %add6 = add i4 %add5, %mul5
   %add7 = add i4 %add6, %mul6
   %add8 = add i4 %add7, %mul7
 
   store i4 %add8, i4 addrspace(1)* %dst, align 4
   ret void
 }
+
+define amdgpu_kernel void @idot8_acc32(<8 x i4> addrspace(1)* %src1,
+; GFX7-LABEL: idot8_acc32:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s0, s[8:9], 0x0
+; GFX7-NEXT:    s_load_dword s1, s[10:11], 0x0
+; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_bfe_i32 s8, s0, 0x40000
+; GFX7-NEXT:    s_bfe_i32 s9, s1, 0x40000
+; GFX7-NEXT:    s_bfe_i32 s11, s1, 0x40004
+; GFX7-NEXT:    v_mov_b32_e32 v0, s9
+; GFX7-NEXT:    v_mov_b32_e32 v1, s2
+; GFX7-NEXT:    v_mad_i32_i24 v0, s8, v0, v1
+; GFX7-NEXT:    s_bfe_i32 s10, s0, 0x40004
+; GFX7-NEXT:    v_mov_b32_e32 v1, s11
+; GFX7-NEXT:    s_bfe_i32 s13, s1, 0x40008
+; GFX7-NEXT:    v_mad_i32_i24 v0, s10, v1, v0
+; GFX7-NEXT:    s_bfe_i32 s12, s0, 0x40008
+; GFX7-NEXT:    v_mov_b32_e32 v1, s13
+; GFX7-NEXT:    s_bfe_i32 s15, s1, 0x4000c
+; GFX7-NEXT:    v_mad_i32_i24 v0, s12, v1, v0
+; GFX7-NEXT:    s_bfe_i32 s14, s0, 0x4000c
+; GFX7-NEXT:    v_mov_b32_e32 v1, s15
+; GFX7-NEXT:    s_bfe_i32 s17, s1, 0x40010
+; GFX7-NEXT:    v_mad_i32_i24 v0, s14, v1, v0
+; GFX7-NEXT:    s_bfe_i32 s16, s0, 0x40010
+; GFX7-NEXT:    v_mov_b32_e32 v1, s17
+; GFX7-NEXT:    s_bfe_i32 s19, s1, 0x40014
+; GFX7-NEXT:    s_bfe_i32 s21, s1, 0x40018
+; GFX7-NEXT:    v_mad_i32_i24 v0, s16, v1, v0
+; GFX7-NEXT:    s_bfe_i32 s18, s0, 0x40014
+; GFX7-NEXT:    v_mov_b32_e32 v1, s19
+; GFX7-NEXT:    s_bfe_i32 s20, s0, 0x40018
+; GFX7-NEXT:    v_mad_i32_i24 v0, s18, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s21
+; GFX7-NEXT:    s_ashr_i32 s1, s1, 28
+; GFX7-NEXT:    v_mad_i32_i24 v0, s20, v1, v0
+; GFX7-NEXT:    s_ashr_i32 s0, s0, 28
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    v_mad_i32_i24 v0, s0, v1, v0
+; GFX7-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: idot8_acc32:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX8-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_bfe_i32 s0, s2, 0x40000
+; GFX8-NEXT:    s_bfe_i32 s1, s4, 0x40000
+; GFX8-NEXT:    s_bfe_i32 s7, s4, 0x40004
+; GFX8-NEXT:    v_mov_b32_e32 v2, s1
+; GFX8-NEXT:    v_mov_b32_e32 v3, s5
+; GFX8-NEXT:    v_mad_i32_i24 v2, s0, v2, v3
+; GFX8-NEXT:    s_bfe_i32 s6, s2, 0x40004
+; GFX8-NEXT:    v_mov_b32_e32 v3, s7
+; GFX8-NEXT:    s_bfe_i32 s9, s4, 0x40008
+; GFX8-NEXT:    v_mad_i32_i24 v2, s6, v3, v2
+; GFX8-NEXT:    s_bfe_i32 s8, s2, 0x40008
+; GFX8-NEXT:    v_mov_b32_e32 v3, s9
+; GFX8-NEXT:    s_bfe_i32 s11, s4, 0x4000c
+; GFX8-NEXT:    v_mad_i32_i24 v2, s8, v3, v2
+; GFX8-NEXT:    s_bfe_i32 s10, s2, 0x4000c
+; GFX8-NEXT:    v_mov_b32_e32 v3, s11
+; GFX8-NEXT:    s_bfe_i32 s13, s4, 0x40010
+; GFX8-NEXT:    v_mad_i32_i24 v2, s10, v3, v2
+; GFX8-NEXT:    s_bfe_i32 s12, s2, 0x40010
+; GFX8-NEXT:    v_mov_b32_e32 v3, s13
+; GFX8-NEXT:    s_bfe_i32 s15, s4, 0x40014
+; GFX8-NEXT:    s_bfe_i32 s17, s4, 0x40018
+; GFX8-NEXT:    v_mad_i32_i24 v2, s12, v3, v2
+; GFX8-NEXT:    s_bfe_i32 s14, s2, 0x40014
+; GFX8-NEXT:    v_mov_b32_e32 v3, s15
+; GFX8-NEXT:    s_bfe_i32 s16, s2, 0x40018
+; GFX8-NEXT:    v_mad_i32_i24 v2, s14, v3, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s17
+; GFX8-NEXT:    s_ashr_i32 s4, s4, 28
+; GFX8-NEXT:    v_mad_i32_i24 v2, s16, v3, v2
+; GFX8-NEXT:    s_ashr_i32 s2, s2, 28
+; GFX8-NEXT:    v_mov_b32_e32 v3, s4
+; GFX8-NEXT:    v_mad_i32_i24 v2, s2, v3, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: idot8_acc32:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_bfe_i32 s0, s2, 0x40000
+; GFX9-NEXT:    s_bfe_i32 s1, s4, 0x40000
+; GFX9-NEXT:    s_bfe_i32 s7, s4, 0x40004
+; GFX9-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-NEXT:    v_mad_i32_i24 v2, s0, v2, v3
+; GFX9-NEXT:    s_bfe_i32 s6, s2, 0x40004
+; GFX9-NEXT:    v_mov_b32_e32 v3, s7
+; GFX9-NEXT:    s_bfe_i32 s9, s4, 0x40008
+; GFX9-NEXT:    v_mad_i32_i24 v2, s6, v3, v2
+; GFX9-NEXT:    s_bfe_i32 s8, s2, 0x40008
+; GFX9-NEXT:    v_mov_b32_e32 v3, s9
+; GFX9-NEXT:    s_bfe_i32 s11, s4, 0x4000c
+; GFX9-NEXT:    v_mad_i32_i24 v2, s8, v3, v2
+; GFX9-NEXT:    s_bfe_i32 s10, s2, 0x4000c
+; GFX9-NEXT:    v_mov_b32_e32 v3, s11
+; GFX9-NEXT:    s_bfe_i32 s13, s4, 0x40010
+; GFX9-NEXT:    v_mad_i32_i24 v2, s10, v3, v2
+; GFX9-NEXT:    s_bfe_i32 s12, s2, 0x40010
+; GFX9-NEXT:    v_mov_b32_e32 v3, s13
+; GFX9-NEXT:    s_bfe_i32 s15, s4, 0x40014
+; GFX9-NEXT:    s_bfe_i32 s17, s4, 0x40018
+; GFX9-NEXT:    v_mad_i32_i24 v2, s12, v3, v2
+; GFX9-NEXT:    s_bfe_i32 s14, s2, 0x40014
+; GFX9-NEXT:    v_mov_b32_e32 v3, s15
+; GFX9-NEXT:    s_bfe_i32 s16, s2, 0x40018
+; GFX9-NEXT:    v_mad_i32_i24 v2, s14, v3, v2
+; GFX9-NEXT:    v_mov_b32_e32 v3, s17
+; GFX9-NEXT:    s_ashr_i32 s4, s4, 28
+; GFX9-NEXT:    v_mad_i32_i24 v2, s16, v3, v2
+; GFX9-NEXT:    s_ashr_i32 s2, s2, 28
+; GFX9-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-NEXT:    v_mad_i32_i24 v2, s2, v3, v2
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: idot8_acc32:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s4
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-DL-NEXT:    v_dot8_i32_i4 v2, s2, v2, v3
+; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                       <8 x i4> addrspace(1)* %src2,
+                                       i32 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
+  %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
+
+  %v1e0 = extractelement <8 x i4> %vec1, i64 0
+  %cv1e0 = sext i4 %v1e0 to i32
+  %v2e0 = extractelement <8 x i4> %vec2, i64 0
+  %cv2e0 = sext i4 %v2e0 to i32
+  %mul0 = mul nuw nsw i32 %cv1e0, %cv2e0
+
+  %v1e1 = extractelement <8 x i4> %vec1, i64 1
+  %cv1e1 = sext i4 %v1e1 to i32
+  %v2e1 = extractelement <8 x i4> %vec2, i64 1
+  %cv2e1 = sext i4 %v2e1 to i32
+  %mul1 = mul nuw nsw i32 %cv1e1, %cv2e1
+
+  %v1e2 = extractelement <8 x i4> %vec1, i64 2
+  %cv1e2 = sext i4 %v1e2 to i32
+  %v2e2 = extractelement <8 x i4> %vec2, i64 2
+  %cv2e2 = sext i4 %v2e2 to i32
+  %mul2 = mul nuw nsw i32 %cv1e2, %cv2e2
+
+  %v1e3 = extractelement <8 x i4> %vec1, i64 3
+  %cv1e3 = sext i4 %v1e3 to i32
+  %v2e3 = extractelement <8 x i4> %vec2, i64 3
+  %cv2e3 = sext i4 %v2e3 to i32
+  %mul3 = mul nuw nsw i32 %cv1e3, %cv2e3
+
+  %v1e4 = extractelement <8 x i4> %vec1, i64 4
+  %cv1e4 = sext i4 %v1e4 to i32
+  %v2e4 = extractelement <8 x i4> %vec2, i64 4
+  %cv2e4 = sext i4 %v2e4 to i32
+  %mul4 = mul nuw nsw i32 %cv1e4, %cv2e4
+
+  %v1e5 = extractelement <8 x i4> %vec1, i64 5
+  %cv1e5 = sext i4 %v1e5 to i32
+  %v2e5 = extractelement <8 x i4> %vec2, i64 5
+  %cv2e5 = sext i4 %v2e5 to i32
+  %mul5 = mul nuw nsw i32 %cv1e5, %cv2e5
+
+  %v1e6 = extractelement <8 x i4> %vec1, i64 6
+  %cv1e6 = sext i4 %v1e6 to i32
+  %v2e6 = extractelement <8 x i4> %vec2, i64 6
+  %cv2e6 = sext i4 %v2e6 to i32
+  %mul6 = mul nuw nsw i32 %cv1e6, %cv2e6
+
+  %v1e7 = extractelement <8 x i4> %vec1, i64 7
+  %cv1e7 = sext i4 %v1e7 to i32
+  %v2e7 = extractelement <8 x i4> %vec2, i64 7
+  %cv2e7 = sext i4 %v2e7 to i32
+  %mul7 = mul nuw nsw i32 %cv1e7, %cv2e7
+
+  %acc = load i32, i32 addrspace(1)* %dst, align 4
+  %add1 = add i32 %mul0, %acc
+  %add2 = add i32 %add1, %mul1
+  %add3 = add i32 %add2, %mul2
+  %add4 = add i32 %add3, %mul3
+  %add5 = add i32 %add4, %mul4
+  %add6 = add i32 %add5, %mul5
+  %add7 = add i32 %add6, %mul6
+  %add8 = add i32 %add7, %mul7
+
+  store i32 %add8, i32 addrspace(1)* %dst, align 4
+  ret void
+}
+
+; TODO: Once the unnecessary zero extentions of the elements are removed;
+; pattern recognizer will kick in.
+define amdgpu_kernel void @idot8_acc16(<8 x i4> addrspace(1)* %src1,
+; GFX7-LABEL: idot8_acc16:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s1, s[8:9], 0x0
+; GFX7-NEXT:    buffer_load_ushort v0, off, s[4:7], 0
+; GFX7-NEXT:    s_load_dword s2, s[10:11], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_bfe_i32 s8, s1, 0x40000
+; GFX7-NEXT:    s_bfe_i32 s10, s1, 0x40004
+; GFX7-NEXT:    s_bfe_i32 s9, s2, 0x40000
+; GFX7-NEXT:    s_bfe_i32 s11, s2, 0x40004
+; GFX7-NEXT:    s_and_b32 s9, s9, s0
+; GFX7-NEXT:    s_bfe_i32 s13, s2, 0x40008
+; GFX7-NEXT:    s_and_b32 s11, s11, s0
+; GFX7-NEXT:    s_and_b32 s8, s8, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s9
+; GFX7-NEXT:    s_bfe_i32 s12, s1, 0x40008
+; GFX7-NEXT:    s_bfe_i32 s15, s2, 0x4000c
+; GFX7-NEXT:    s_and_b32 s13, s13, s0
+; GFX7-NEXT:    s_and_b32 s10, s10, s0
+; GFX7-NEXT:    v_mov_b32_e32 v2, s11
+; GFX7-NEXT:    s_bfe_i32 s14, s1, 0x4000c
+; GFX7-NEXT:    s_bfe_i32 s17, s2, 0x40010
+; GFX7-NEXT:    s_and_b32 s15, s15, s0
+; GFX7-NEXT:    s_and_b32 s12, s12, s0
+; GFX7-NEXT:    v_mov_b32_e32 v3, s13
+; GFX7-NEXT:    s_bfe_i32 s16, s1, 0x40010
+; GFX7-NEXT:    s_bfe_i32 s19, s2, 0x40014
+; GFX7-NEXT:    s_and_b32 s17, s17, s0
+; GFX7-NEXT:    s_and_b32 s14, s14, s0
+; GFX7-NEXT:    v_mov_b32_e32 v4, s15
+; GFX7-NEXT:    s_bfe_i32 s21, s2, 0x40018
+; GFX7-NEXT:    s_bfe_i32 s18, s1, 0x40014
+; GFX7-NEXT:    s_and_b32 s19, s19, s0
+; GFX7-NEXT:    s_and_b32 s16, s16, s0
+; GFX7-NEXT:    v_mov_b32_e32 v5, s17
+; GFX7-NEXT:    s_bfe_i32 s20, s1, 0x40018
+; GFX7-NEXT:    s_ashr_i32 s2, s2, 28
+; GFX7-NEXT:    s_and_b32 s21, s21, s0
+; GFX7-NEXT:    s_and_b32 s18, s18, s0
+; GFX7-NEXT:    v_mov_b32_e32 v6, s19
+; GFX7-NEXT:    s_ashr_i32 s1, s1, 28
+; GFX7-NEXT:    s_and_b32 s20, s20, s0
+; GFX7-NEXT:    s_and_b32 s2, s2, s0
+; GFX7-NEXT:    v_mov_b32_e32 v7, s21
+; GFX7-NEXT:    s_and_b32 s0, s1, s0
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_mad_u32_u24 v0, s8, v1, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s10, v2, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s12, v3, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s14, v4, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s16, v5, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s18, v6, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s20, v7, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s2
+; GFX7-NEXT:    v_mad_u32_u24 v0, s0, v1, v0
+; GFX7-NEXT:    buffer_store_short v0, off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: idot8_acc16:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ushort v2, v[0:1]
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s0, s2, 4
+; GFX8-NEXT:    s_lshr_b32 s1, s4, 4
+; GFX8-NEXT:    s_bfe_i32 s5, s4, 0x40000
+; GFX8-NEXT:    v_lshlrev_b16_e64 v3, 12, s0
+; GFX8-NEXT:    v_lshlrev_b16_e64 v4, 12, s1
+; GFX8-NEXT:    s_bfe_i32 s0, s4, 0x40008
+; GFX8-NEXT:    v_mov_b32_e32 v5, s5
+; GFX8-NEXT:    s_bfe_i32 s6, s2, 0x40000
+; GFX8-NEXT:    s_lshr_b32 s1, s2, 12
+; GFX8-NEXT:    s_lshr_b32 s5, s4, 12
+; GFX8-NEXT:    v_mov_b32_e32 v6, s0
+; GFX8-NEXT:    s_bfe_i32 s7, s2, 0x40008
+; GFX8-NEXT:    v_ashrrev_i16_e32 v3, 12, v3
+; GFX8-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
+; GFX8-NEXT:    v_lshlrev_b16_e64 v7, 12, s1
+; GFX8-NEXT:    v_lshlrev_b16_e64 v8, 12, s5
+; GFX8-NEXT:    v_mul_i32_i24_e32 v6, s7, v6
+; GFX8-NEXT:    s_lshr_b32 s0, s2, 20
+; GFX8-NEXT:    s_lshr_b32 s1, s4, 20
+; GFX8-NEXT:    s_bfe_i32 s5, s4, 0x40010
+; GFX8-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
+; GFX8-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
+; GFX8-NEXT:    v_lshlrev_b16_e64 v9, 12, s0
+; GFX8-NEXT:    v_lshlrev_b16_e64 v10, 12, s1
+; GFX8-NEXT:    s_bfe_i32 s8, s2, 0x40010
+; GFX8-NEXT:    v_mov_b32_e32 v13, s5
+; GFX8-NEXT:    s_lshr_b32 s0, s2, 28
+; GFX8-NEXT:    s_lshr_b32 s9, s4, 28
+; GFX8-NEXT:    s_bfe_i32 s4, s4, 0x40018
+; GFX8-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX8-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX8-NEXT:    v_lshlrev_b16_e64 v11, 12, s0
+; GFX8-NEXT:    v_lshlrev_b16_e64 v12, 12, s9
+; GFX8-NEXT:    s_bfe_i32 s2, s2, 0x40018
+; GFX8-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
+; GFX8-NEXT:    v_ashrrev_i16_e32 v12, 12, v12
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mad_i32_i24 v2, s6, v5, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, v3, v4, v2
+; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v6, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
+; GFX8-NEXT:    v_mad_u32_u24 v2, v7, v8, v2
+; GFX8-NEXT:    v_mad_i32_i24 v2, s8, v13, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, v9, v10, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s4
+; GFX8-NEXT:    v_mad_i32_i24 v2, s2, v3, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, v11, v12, v2
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: idot8_acc16:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    global_load_ushort v2, v[0:1], off
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b32 s0, s2, 4
+; GFX9-NEXT:    s_lshr_b32 s1, s4, 4
+; GFX9-NEXT:    s_bfe_i32 s5, s4, 0x40000
+; GFX9-NEXT:    v_lshlrev_b16_e64 v3, 12, s0
+; GFX9-NEXT:    v_lshlrev_b16_e64 v4, 12, s1
+; GFX9-NEXT:    s_bfe_i32 s0, s4, 0x40008
+; GFX9-NEXT:    v_mov_b32_e32 v5, s5
+; GFX9-NEXT:    s_bfe_i32 s6, s2, 0x40000
+; GFX9-NEXT:    s_lshr_b32 s1, s2, 12
+; GFX9-NEXT:    s_lshr_b32 s5, s4, 12
+; GFX9-NEXT:    v_mov_b32_e32 v6, s0
+; GFX9-NEXT:    s_bfe_i32 s7, s2, 0x40008
+; GFX9-NEXT:    v_ashrrev_i16_e32 v3, 12, v3
+; GFX9-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
+; GFX9-NEXT:    v_lshlrev_b16_e64 v7, 12, s1
+; GFX9-NEXT:    v_lshlrev_b16_e64 v8, 12, s5
+; GFX9-NEXT:    v_mul_i32_i24_e32 v6, s7, v6
+; GFX9-NEXT:    s_lshr_b32 s0, s2, 20
+; GFX9-NEXT:    s_lshr_b32 s1, s4, 20
+; GFX9-NEXT:    s_bfe_i32 s5, s4, 0x40010
+; GFX9-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
+; GFX9-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
+; GFX9-NEXT:    v_lshlrev_b16_e64 v9, 12, s0
+; GFX9-NEXT:    v_lshlrev_b16_e64 v10, 12, s1
+; GFX9-NEXT:    s_bfe_i32 s8, s2, 0x40010
+; GFX9-NEXT:    v_mov_b32_e32 v13, s5
+; GFX9-NEXT:    s_lshr_b32 s0, s2, 28
+; GFX9-NEXT:    s_lshr_b32 s9, s4, 28
+; GFX9-NEXT:    s_bfe_i32 s4, s4, 0x40018
+; GFX9-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX9-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX9-NEXT:    v_lshlrev_b16_e64 v11, 12, s0
+; GFX9-NEXT:    v_lshlrev_b16_e64 v12, 12, s9
+; GFX9-NEXT:    s_bfe_i32 s2, s2, 0x40018
+; GFX9-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
+; GFX9-NEXT:    v_ashrrev_i16_e32 v12, 12, v12
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_mad_i32_i24 v2, s6, v5, v2
+; GFX9-NEXT:    v_mad_u32_u24 v2, v3, v4, v2
+; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
+; GFX9-NEXT:    v_mad_u32_u24 v2, v7, v8, v2
+; GFX9-NEXT:    v_mad_i32_i24 v2, s8, v13, v2
+; GFX9-NEXT:    v_mad_u32_u24 v2, v9, v10, v2
+; GFX9-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-NEXT:    v_mad_i32_i24 v2, s2, v3, v2
+; GFX9-NEXT:    v_mad_u32_u24 v2, v11, v12, v2
+; GFX9-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: idot8_acc16:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ushort v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_lshr_b32 s0, s2, 4
+; GFX9-DL-NEXT:    s_lshr_b32 s1, s4, 4
+; GFX9-DL-NEXT:    s_bfe_i32 s5, s4, 0x40000
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v3, 12, s0
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v4, 12, s1
+; GFX9-DL-NEXT:    s_bfe_i32 s0, s4, 0x40008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s5
+; GFX9-DL-NEXT:    s_bfe_i32 s6, s2, 0x40000
+; GFX9-DL-NEXT:    s_lshr_b32 s1, s2, 12
+; GFX9-DL-NEXT:    s_lshr_b32 s5, s4, 12
+; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s0
+; GFX9-DL-NEXT:    s_bfe_i32 s7, s2, 0x40008
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v3, 12, v3
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v7, 12, s1
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v8, 12, s5
+; GFX9-DL-NEXT:    v_mul_i32_i24_e32 v6, s7, v6
+; GFX9-DL-NEXT:    s_lshr_b32 s0, s2, 20
+; GFX9-DL-NEXT:    s_lshr_b32 s1, s4, 20
+; GFX9-DL-NEXT:    s_bfe_i32 s5, s4, 0x40010
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v9, 12, s0
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v10, 12, s1
+; GFX9-DL-NEXT:    s_bfe_i32 s8, s2, 0x40010
+; GFX9-DL-NEXT:    v_mov_b32_e32 v13, s5
+; GFX9-DL-NEXT:    s_lshr_b32 s0, s2, 28
+; GFX9-DL-NEXT:    s_lshr_b32 s9, s4, 28
+; GFX9-DL-NEXT:    s_bfe_i32 s4, s4, 0x40018
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v11, 12, s0
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v12, 12, s9
+; GFX9-DL-NEXT:    s_bfe_i32 s2, s2, 0x40018
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v12, 12, v12
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s6, v5, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, v3, v4, v2
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, v7, v8, v2
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s8, v13, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, v9, v10, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s2, v3, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, v11, v12, v2
+; GFX9-DL-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                       <8 x i4> addrspace(1)* %src2,
+                                       i16 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
+  %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
+
+  %v1e0 = extractelement <8 x i4> %vec1, i64 0
+  %cv1e0 = sext i4 %v1e0 to i16
+  %v2e0 = extractelement <8 x i4> %vec2, i64 0
+  %cv2e0 = sext i4 %v2e0 to i16
+  %mul0 = mul nuw nsw i16 %cv1e0, %cv2e0
+
+  %v1e1 = extractelement <8 x i4> %vec1, i64 1
+  %cv1e1 = sext i4 %v1e1 to i16
+  %v2e1 = extractelement <8 x i4> %vec2, i64 1
+  %cv2e1 = sext i4 %v2e1 to i16
+  %mul1 = mul nuw nsw i16 %cv1e1, %cv2e1
+
+  %v1e2 = extractelement <8 x i4> %vec1, i64 2
+  %cv1e2 = sext i4 %v1e2 to i16
+  %v2e2 = extractelement <8 x i4> %vec2, i64 2
+  %cv2e2 = sext i4 %v2e2 to i16
+  %mul2 = mul nuw nsw i16 %cv1e2, %cv2e2
+
+  %v1e3 = extractelement <8 x i4> %vec1, i64 3
+  %cv1e3 = sext i4 %v1e3 to i16
+  %v2e3 = extractelement <8 x i4> %vec2, i64 3
+  %cv2e3 = sext i4 %v2e3 to i16
+  %mul3 = mul nuw nsw i16 %cv1e3, %cv2e3
+
+  %v1e4 = extractelement <8 x i4> %vec1, i64 4
+  %cv1e4 = sext i4 %v1e4 to i16
+  %v2e4 = extractelement <8 x i4> %vec2, i64 4
+  %cv2e4 = sext i4 %v2e4 to i16
+  %mul4 = mul nuw nsw i16 %cv1e4, %cv2e4
+
+  %v1e5 = extractelement <8 x i4> %vec1, i64 5
+  %cv1e5 = sext i4 %v1e5 to i16
+  %v2e5 = extractelement <8 x i4> %vec2, i64 5
+  %cv2e5 = sext i4 %v2e5 to i16
+  %mul5 = mul nuw nsw i16 %cv1e5, %cv2e5
+
+  %v1e6 = extractelement <8 x i4> %vec1, i64 6
+  %cv1e6 = sext i4 %v1e6 to i16
+  %v2e6 = extractelement <8 x i4> %vec2, i64 6
+  %cv2e6 = sext i4 %v2e6 to i16
+  %mul6 = mul nuw nsw i16 %cv1e6, %cv2e6
+
+  %v1e7 = extractelement <8 x i4> %vec1, i64 7
+  %cv1e7 = sext i4 %v1e7 to i16
+  %v2e7 = extractelement <8 x i4> %vec2, i64 7
+  %cv2e7 = sext i4 %v2e7 to i16
+  %mul7 = mul nuw nsw i16 %cv1e7, %cv2e7
+
+  %acc = load i16, i16 addrspace(1)* %dst, align 4
+  %add1 = add i16 %mul0, %acc
+  %add2 = add i16 %add1, %mul1
+  %add3 = add i16 %add2, %mul2
+  %add4 = add i16 %add3, %mul3
+  %add5 = add i16 %add4, %mul4
+  %add6 = add i16 %add5, %mul5
+  %add7 = add i16 %add6, %mul6
+  %add8 = add i16 %add7, %mul7
+
+  store i16 %add8, i16 addrspace(1)* %dst, align 4
+  ret void
+}
+
+; TODO: Support this pattern.
+define amdgpu_kernel void @idot8_acc8(<8 x i4> addrspace(1)* %src1,
+; GFX7-LABEL: idot8_acc8:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_movk_i32 s0, 0xff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s1, s[8:9], 0x0
+; GFX7-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0
+; GFX7-NEXT:    s_load_dword s2, s[10:11], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_bfe_i32 s8, s1, 0x40000
+; GFX7-NEXT:    s_bfe_i32 s10, s1, 0x40004
+; GFX7-NEXT:    s_bfe_i32 s9, s2, 0x40000
+; GFX7-NEXT:    s_bfe_i32 s11, s2, 0x40004
+; GFX7-NEXT:    s_and_b32 s9, s9, s0
+; GFX7-NEXT:    s_bfe_i32 s13, s2, 0x40008
+; GFX7-NEXT:    s_and_b32 s11, s11, s0
+; GFX7-NEXT:    s_and_b32 s8, s8, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s9
+; GFX7-NEXT:    s_bfe_i32 s12, s1, 0x40008
+; GFX7-NEXT:    s_bfe_i32 s15, s2, 0x4000c
+; GFX7-NEXT:    s_and_b32 s13, s13, s0
+; GFX7-NEXT:    s_and_b32 s10, s10, s0
+; GFX7-NEXT:    v_mov_b32_e32 v2, s11
+; GFX7-NEXT:    s_bfe_i32 s14, s1, 0x4000c
+; GFX7-NEXT:    s_bfe_i32 s17, s2, 0x40010
+; GFX7-NEXT:    s_and_b32 s15, s15, s0
+; GFX7-NEXT:    s_and_b32 s12, s12, s0
+; GFX7-NEXT:    v_mov_b32_e32 v3, s13
+; GFX7-NEXT:    s_bfe_i32 s16, s1, 0x40010
+; GFX7-NEXT:    s_bfe_i32 s19, s2, 0x40014
+; GFX7-NEXT:    s_and_b32 s17, s17, s0
+; GFX7-NEXT:    s_and_b32 s14, s14, s0
+; GFX7-NEXT:    v_mov_b32_e32 v4, s15
+; GFX7-NEXT:    s_bfe_i32 s21, s2, 0x40018
+; GFX7-NEXT:    s_bfe_i32 s18, s1, 0x40014
+; GFX7-NEXT:    s_and_b32 s19, s19, s0
+; GFX7-NEXT:    s_and_b32 s16, s16, s0
+; GFX7-NEXT:    v_mov_b32_e32 v5, s17
+; GFX7-NEXT:    s_bfe_i32 s20, s1, 0x40018
+; GFX7-NEXT:    s_ashr_i32 s2, s2, 28
+; GFX7-NEXT:    s_and_b32 s21, s21, s0
+; GFX7-NEXT:    s_and_b32 s18, s18, s0
+; GFX7-NEXT:    v_mov_b32_e32 v6, s19
+; GFX7-NEXT:    s_ashr_i32 s1, s1, 28
+; GFX7-NEXT:    s_and_b32 s20, s20, s0
+; GFX7-NEXT:    s_and_b32 s2, s2, s0
+; GFX7-NEXT:    v_mov_b32_e32 v7, s21
+; GFX7-NEXT:    s_and_b32 s0, s1, s0
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_mad_u32_u24 v0, s8, v1, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s10, v2, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s12, v3, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s14, v4, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s16, v5, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s18, v6, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s20, v7, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s2
+; GFX7-NEXT:    v_mad_u32_u24 v0, s0, v1, v0
+; GFX7-NEXT:    buffer_store_byte v0, off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: idot8_acc8:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_movk_i32 s2, 0xff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ubyte v2, v[0:1]
+; GFX8-NEXT:    s_load_dword s0, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s7, s0, 4
+; GFX8-NEXT:    s_lshr_b32 s11, s1, 4
+; GFX8-NEXT:    v_lshlrev_b16_e64 v3, 12, s7
+; GFX8-NEXT:    v_lshlrev_b16_e64 v4, 12, s11
+; GFX8-NEXT:    s_bfe_i32 s13, s1, 0x40000
+; GFX8-NEXT:    v_ashrrev_i16_e32 v3, 12, v3
+; GFX8-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
+; GFX8-NEXT:    s_lshr_b32 s6, s0, 12
+; GFX8-NEXT:    s_lshr_b32 s10, s1, 12
+; GFX8-NEXT:    s_bfe_i32 s15, s1, 0x40008
+; GFX8-NEXT:    s_bfe_i32 s12, s0, 0x40000
+; GFX8-NEXT:    v_mov_b32_e32 v12, s13
+; GFX8-NEXT:    v_lshlrev_b16_e64 v6, 12, s6
+; GFX8-NEXT:    v_lshlrev_b16_e64 v7, 12, s10
+; GFX8-NEXT:    s_bfe_i32 s14, s0, 0x40008
+; GFX8-NEXT:    v_mov_b32_e32 v5, s15
+; GFX8-NEXT:    v_and_b32_e32 v3, s2, v3
+; GFX8-NEXT:    v_and_b32_e32 v4, s2, v4
+; GFX8-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
+; GFX8-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
+; GFX8-NEXT:    s_lshr_b32 s5, s0, 20
+; GFX8-NEXT:    s_lshr_b32 s9, s1, 20
+; GFX8-NEXT:    v_mul_i32_i24_e32 v5, s14, v5
+; GFX8-NEXT:    v_lshlrev_b16_e64 v8, 12, s5
+; GFX8-NEXT:    v_lshlrev_b16_e64 v9, 12, s9
+; GFX8-NEXT:    s_bfe_i32 s17, s1, 0x40010
+; GFX8-NEXT:    v_and_b32_e32 v6, s2, v6
+; GFX8-NEXT:    v_and_b32_e32 v7, s2, v7
+; GFX8-NEXT:    s_lshr_b32 s8, s1, 28
+; GFX8-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
+; GFX8-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX8-NEXT:    s_lshr_b32 s4, s0, 28
+; GFX8-NEXT:    s_bfe_i32 s16, s0, 0x40010
+; GFX8-NEXT:    v_mov_b32_e32 v13, s17
+; GFX8-NEXT:    v_lshlrev_b16_e64 v10, 12, s4
+; GFX8-NEXT:    v_lshlrev_b16_e64 v11, 12, s8
+; GFX8-NEXT:    s_bfe_i32 s1, s1, 0x40018
+; GFX8-NEXT:    v_and_b32_e32 v8, s2, v8
+; GFX8-NEXT:    v_and_b32_e32 v9, s2, v9
+; GFX8-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX8-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
+; GFX8-NEXT:    s_bfe_i32 s0, s0, 0x40018
+; GFX8-NEXT:    v_and_b32_e32 v10, s2, v10
+; GFX8-NEXT:    v_and_b32_e32 v11, s2, v11
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mad_i32_i24 v2, s12, v12, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, v3, v4, v2
+; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v5, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
+; GFX8-NEXT:    v_mad_u32_u24 v2, v6, v7, v2
+; GFX8-NEXT:    v_mad_i32_i24 v2, s16, v13, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, v8, v9, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_mad_i32_i24 v2, s0, v3, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, v10, v11, v2
+; GFX8-NEXT:    flat_store_byte v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: idot8_acc8:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NEXT:    s_movk_i32 s2, 0xff
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-NEXT:    s_load_dword s0, s[4:5], 0x0
+; GFX9-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b32 s7, s0, 4
+; GFX9-NEXT:    s_lshr_b32 s11, s1, 4
+; GFX9-NEXT:    v_lshlrev_b16_e64 v3, 12, s7
+; GFX9-NEXT:    v_lshlrev_b16_e64 v4, 12, s11
+; GFX9-NEXT:    s_bfe_i32 s13, s1, 0x40000
+; GFX9-NEXT:    v_ashrrev_i16_e32 v3, 12, v3
+; GFX9-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
+; GFX9-NEXT:    s_lshr_b32 s6, s0, 12
+; GFX9-NEXT:    s_lshr_b32 s10, s1, 12
+; GFX9-NEXT:    s_bfe_i32 s15, s1, 0x40008
+; GFX9-NEXT:    s_bfe_i32 s12, s0, 0x40000
+; GFX9-NEXT:    v_mov_b32_e32 v12, s13
+; GFX9-NEXT:    v_lshlrev_b16_e64 v6, 12, s6
+; GFX9-NEXT:    v_lshlrev_b16_e64 v7, 12, s10
+; GFX9-NEXT:    s_bfe_i32 s14, s0, 0x40008
+; GFX9-NEXT:    v_mov_b32_e32 v5, s15
+; GFX9-NEXT:    v_and_b32_e32 v3, s2, v3
+; GFX9-NEXT:    v_and_b32_e32 v4, s2, v4
+; GFX9-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
+; GFX9-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
+; GFX9-NEXT:    s_lshr_b32 s5, s0, 20
+; GFX9-NEXT:    s_lshr_b32 s9, s1, 20
+; GFX9-NEXT:    v_mul_i32_i24_e32 v5, s14, v5
+; GFX9-NEXT:    v_lshlrev_b16_e64 v8, 12, s5
+; GFX9-NEXT:    v_lshlrev_b16_e64 v9, 12, s9
+; GFX9-NEXT:    s_bfe_i32 s17, s1, 0x40010
+; GFX9-NEXT:    v_and_b32_e32 v6, s2, v6
+; GFX9-NEXT:    v_and_b32_e32 v7, s2, v7
+; GFX9-NEXT:    s_lshr_b32 s8, s1, 28
+; GFX9-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
+; GFX9-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX9-NEXT:    s_lshr_b32 s4, s0, 28
+; GFX9-NEXT:    s_bfe_i32 s16, s0, 0x40010
+; GFX9-NEXT:    v_mov_b32_e32 v13, s17
+; GFX9-NEXT:    v_lshlrev_b16_e64 v10, 12, s4
+; GFX9-NEXT:    v_lshlrev_b16_e64 v11, 12, s8
+; GFX9-NEXT:    s_bfe_i32 s1, s1, 0x40018
+; GFX9-NEXT:    v_and_b32_e32 v8, s2, v8
+; GFX9-NEXT:    v_and_b32_e32 v9, s2, v9
+; GFX9-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX9-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
+; GFX9-NEXT:    s_bfe_i32 s0, s0, 0x40018
+; GFX9-NEXT:    v_and_b32_e32 v10, s2, v10
+; GFX9-NEXT:    v_and_b32_e32 v11, s2, v11
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_mad_i32_i24 v2, s12, v12, v2
+; GFX9-NEXT:    v_mad_u32_u24 v2, v3, v4, v2
+; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
+; GFX9-NEXT:    v_mad_u32_u24 v2, v6, v7, v2
+; GFX9-NEXT:    v_mad_i32_i24 v2, s16, v13, v2
+; GFX9-NEXT:    v_mad_u32_u24 v2, v8, v9, v2
+; GFX9-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-NEXT:    v_mad_i32_i24 v2, s0, v3, v2
+; GFX9-NEXT:    v_mad_u32_u24 v2, v10, v11, v2
+; GFX9-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: idot8_acc8:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_movk_i32 s2, 0xff
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-DL-NEXT:    s_load_dword s0, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_lshr_b32 s7, s0, 4
+; GFX9-DL-NEXT:    s_lshr_b32 s11, s1, 4
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v3, 12, s7
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v4, 12, s11
+; GFX9-DL-NEXT:    s_bfe_i32 s13, s1, 0x40000
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v3, 12, v3
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
+; GFX9-DL-NEXT:    s_lshr_b32 s6, s0, 12
+; GFX9-DL-NEXT:    s_lshr_b32 s10, s1, 12
+; GFX9-DL-NEXT:    s_bfe_i32 s15, s1, 0x40008
+; GFX9-DL-NEXT:    s_bfe_i32 s12, s0, 0x40000
+; GFX9-DL-NEXT:    v_mov_b32_e32 v12, s13
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v6, 12, s6
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v7, 12, s10
+; GFX9-DL-NEXT:    s_bfe_i32 s14, s0, 0x40008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s15
+; GFX9-DL-NEXT:    v_and_b32_e32 v3, s2, v3
+; GFX9-DL-NEXT:    v_and_b32_e32 v4, s2, v4
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
+; GFX9-DL-NEXT:    s_lshr_b32 s5, s0, 20
+; GFX9-DL-NEXT:    s_lshr_b32 s9, s1, 20
+; GFX9-DL-NEXT:    v_mul_i32_i24_e32 v5, s14, v5
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v8, 12, s5
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v9, 12, s9
+; GFX9-DL-NEXT:    s_bfe_i32 s17, s1, 0x40010
+; GFX9-DL-NEXT:    v_and_b32_e32 v6, s2, v6
+; GFX9-DL-NEXT:    v_and_b32_e32 v7, s2, v7
+; GFX9-DL-NEXT:    s_lshr_b32 s8, s1, 28
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX9-DL-NEXT:    s_lshr_b32 s4, s0, 28
+; GFX9-DL-NEXT:    s_bfe_i32 s16, s0, 0x40010
+; GFX9-DL-NEXT:    v_mov_b32_e32 v13, s17
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v10, 12, s4
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v11, 12, s8
+; GFX9-DL-NEXT:    s_bfe_i32 s1, s1, 0x40018
+; GFX9-DL-NEXT:    v_and_b32_e32 v8, s2, v8
+; GFX9-DL-NEXT:    v_and_b32_e32 v9, s2, v9
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
+; GFX9-DL-NEXT:    s_bfe_i32 s0, s0, 0x40018
+; GFX9-DL-NEXT:    v_and_b32_e32 v10, s2, v10
+; GFX9-DL-NEXT:    v_and_b32_e32 v11, s2, v11
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s12, v12, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, v3, v4, v2
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, v6, v7, v2
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s16, v13, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, v8, v9, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s0, v3, v2
+; GFX9-DL-NEXT:    v_mad_u32_u24 v2, v10, v11, v2
+; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                       <8 x i4> addrspace(1)* %src2,
+                                       i8 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
+  %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
+
+  %v1e0 = extractelement <8 x i4> %vec1, i64 0
+  %cv1e0 = sext i4 %v1e0 to i8
+  %v2e0 = extractelement <8 x i4> %vec2, i64 0
+  %cv2e0 = sext i4 %v2e0 to i8
+  %mul0 = mul nuw nsw i8 %cv1e0, %cv2e0
+
+  %v1e1 = extractelement <8 x i4> %vec1, i64 1
+  %cv1e1 = sext i4 %v1e1 to i8
+  %v2e1 = extractelement <8 x i4> %vec2, i64 1
+  %cv2e1 = sext i4 %v2e1 to i8
+  %mul1 = mul nuw nsw i8 %cv1e1, %cv2e1
+
+  %v1e2 = extractelement <8 x i4> %vec1, i64 2
+  %cv1e2 = sext i4 %v1e2 to i8
+  %v2e2 = extractelement <8 x i4> %vec2, i64 2
+  %cv2e2 = sext i4 %v2e2 to i8
+  %mul2 = mul nuw nsw i8 %cv1e2, %cv2e2
+
+  %v1e3 = extractelement <8 x i4> %vec1, i64 3
+  %cv1e3 = sext i4 %v1e3 to i8
+  %v2e3 = extractelement <8 x i4> %vec2, i64 3
+  %cv2e3 = sext i4 %v2e3 to i8
+  %mul3 = mul nuw nsw i8 %cv1e3, %cv2e3
+
+  %v1e4 = extractelement <8 x i4> %vec1, i64 4
+  %cv1e4 = sext i4 %v1e4 to i8
+  %v2e4 = extractelement <8 x i4> %vec2, i64 4
+  %cv2e4 = sext i4 %v2e4 to i8
+  %mul4 = mul nuw nsw i8 %cv1e4, %cv2e4
+
+  %v1e5 = extractelement <8 x i4> %vec1, i64 5
+  %cv1e5 = sext i4 %v1e5 to i8
+  %v2e5 = extractelement <8 x i4> %vec2, i64 5
+  %cv2e5 = sext i4 %v2e5 to i8
+  %mul5 = mul nuw nsw i8 %cv1e5, %cv2e5
+
+  %v1e6 = extractelement <8 x i4> %vec1, i64 6
+  %cv1e6 = sext i4 %v1e6 to i8
+  %v2e6 = extractelement <8 x i4> %vec2, i64 6
+  %cv2e6 = sext i4 %v2e6 to i8
+  %mul6 = mul nuw nsw i8 %cv1e6, %cv2e6
+
+  %v1e7 = extractelement <8 x i4> %vec1, i64 7
+  %cv1e7 = sext i4 %v1e7 to i8
+  %v2e7 = extractelement <8 x i4> %vec2, i64 7
+  %cv2e7 = sext i4 %v2e7 to i8
+  %mul7 = mul nuw nsw i8 %cv1e7, %cv2e7
+
+  %acc = load i8, i8 addrspace(1)* %dst, align 4
+  %add1 = add i8 %mul0, %acc
+  %add2 = add i8 %add1, %mul1
+  %add3 = add i8 %add2, %mul2
+  %add4 = add i8 %add3, %mul3
+  %add5 = add i8 %add4, %mul4
+  %add6 = add i8 %add5, %mul5
+  %add7 = add i8 %add6, %mul6
+  %add8 = add i8 %add7, %mul7
+
+  store i8 %add8, i8 addrspace(1)* %dst, align 4
+  ret void
+}
+
+; Make sure the pattern is not recognized if there are multiple uses of the
+; intermediate multiplications.
+define amdgpu_kernel void @idot8_multiuses_mul1(<8 x i4> addrspace(1)* %src1,
+; GFX7-LABEL: idot8_multiuses_mul1:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s0, s[8:9], 0x0
+; GFX7-NEXT:    s_load_dword s1, s[10:11], 0x0
+; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_bfe_i32 s8, s0, 0x40000
+; GFX7-NEXT:    s_bfe_i32 s9, s1, 0x40000
+; GFX7-NEXT:    v_mov_b32_e32 v0, s9
+; GFX7-NEXT:    v_mov_b32_e32 v1, s2
+; GFX7-NEXT:    v_mad_i32_i24 v1, s8, v0, v1
+; GFX7-NEXT:    s_bfe_i32 s11, s1, 0x40004
+; GFX7-NEXT:    s_bfe_i32 s10, s0, 0x40004
+; GFX7-NEXT:    s_bfe_i32 s13, s1, 0x40008
+; GFX7-NEXT:    v_mad_i32_i24 v0, s8, v0, v1
+; GFX7-NEXT:    v_mov_b32_e32 v2, s11
+; GFX7-NEXT:    v_mad_i32_i24 v0, s10, v2, v0
+; GFX7-NEXT:    s_bfe_i32 s12, s0, 0x40008
+; GFX7-NEXT:    v_mov_b32_e32 v2, s13
+; GFX7-NEXT:    s_bfe_i32 s15, s1, 0x4000c
+; GFX7-NEXT:    v_mad_i32_i24 v0, s12, v2, v0
+; GFX7-NEXT:    s_bfe_i32 s14, s0, 0x4000c
+; GFX7-NEXT:    v_mov_b32_e32 v2, s15
+; GFX7-NEXT:    s_bfe_i32 s17, s1, 0x40010
+; GFX7-NEXT:    v_mad_i32_i24 v0, s14, v2, v0
+; GFX7-NEXT:    s_bfe_i32 s16, s0, 0x40010
+; GFX7-NEXT:    v_mov_b32_e32 v2, s17
+; GFX7-NEXT:    s_bfe_i32 s19, s1, 0x40014
+; GFX7-NEXT:    s_bfe_i32 s21, s1, 0x40018
+; GFX7-NEXT:    v_mad_i32_i24 v0, s16, v2, v0
+; GFX7-NEXT:    s_bfe_i32 s18, s0, 0x40014
+; GFX7-NEXT:    v_mov_b32_e32 v2, s19
+; GFX7-NEXT:    s_bfe_i32 s20, s0, 0x40018
+; GFX7-NEXT:    v_mad_i32_i24 v0, s18, v2, v0
+; GFX7-NEXT:    v_mov_b32_e32 v2, s21
+; GFX7-NEXT:    s_ashr_i32 s1, s1, 28
+; GFX7-NEXT:    v_mad_i32_i24 v0, s20, v2, v0
+; GFX7-NEXT:    s_ashr_i32 s0, s0, 28
+; GFX7-NEXT:    v_mov_b32_e32 v2, s1
+; GFX7-NEXT:    v_mad_i32_i24 v0, s0, v2, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX7-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: idot8_multiuses_mul1:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX8-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_bfe_i32 s0, s2, 0x40000
+; GFX8-NEXT:    s_bfe_i32 s1, s4, 0x40000
+; GFX8-NEXT:    v_mov_b32_e32 v2, s1
+; GFX8-NEXT:    v_mov_b32_e32 v3, s5
+; GFX8-NEXT:    v_mad_i32_i24 v3, s0, v2, v3
+; GFX8-NEXT:    s_bfe_i32 s7, s4, 0x40004
+; GFX8-NEXT:    s_bfe_i32 s6, s2, 0x40004
+; GFX8-NEXT:    s_bfe_i32 s9, s4, 0x40008
+; GFX8-NEXT:    v_mad_i32_i24 v2, s0, v2, v3
+; GFX8-NEXT:    v_mov_b32_e32 v4, s7
+; GFX8-NEXT:    v_mad_i32_i24 v2, s6, v4, v2
+; GFX8-NEXT:    s_bfe_i32 s8, s2, 0x40008
+; GFX8-NEXT:    v_mov_b32_e32 v4, s9
+; GFX8-NEXT:    s_bfe_i32 s11, s4, 0x4000c
+; GFX8-NEXT:    v_mad_i32_i24 v2, s8, v4, v2
+; GFX8-NEXT:    s_bfe_i32 s10, s2, 0x4000c
+; GFX8-NEXT:    v_mov_b32_e32 v4, s11
+; GFX8-NEXT:    s_bfe_i32 s13, s4, 0x40010
+; GFX8-NEXT:    v_mad_i32_i24 v2, s10, v4, v2
+; GFX8-NEXT:    s_bfe_i32 s12, s2, 0x40010
+; GFX8-NEXT:    v_mov_b32_e32 v4, s13
+; GFX8-NEXT:    s_bfe_i32 s15, s4, 0x40014
+; GFX8-NEXT:    s_bfe_i32 s17, s4, 0x40018
+; GFX8-NEXT:    v_mad_i32_i24 v2, s12, v4, v2
+; GFX8-NEXT:    s_bfe_i32 s14, s2, 0x40014
+; GFX8-NEXT:    v_mov_b32_e32 v4, s15
+; GFX8-NEXT:    s_bfe_i32 s16, s2, 0x40018
+; GFX8-NEXT:    v_mad_i32_i24 v2, s14, v4, v2
+; GFX8-NEXT:    v_mov_b32_e32 v4, s17
+; GFX8-NEXT:    s_ashr_i32 s4, s4, 28
+; GFX8-NEXT:    v_mad_i32_i24 v2, s16, v4, v2
+; GFX8-NEXT:    s_ashr_i32 s2, s2, 28
+; GFX8-NEXT:    v_mov_b32_e32 v4, s4
+; GFX8-NEXT:    v_mad_i32_i24 v2, s2, v4, v2
+; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v3
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: idot8_multiuses_mul1:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_bfe_i32 s0, s2, 0x40000
+; GFX9-NEXT:    s_bfe_i32 s1, s4, 0x40000
+; GFX9-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-NEXT:    v_mad_i32_i24 v3, s0, v2, v3
+; GFX9-NEXT:    s_bfe_i32 s7, s4, 0x40004
+; GFX9-NEXT:    s_bfe_i32 s6, s2, 0x40004
+; GFX9-NEXT:    s_bfe_i32 s9, s4, 0x40008
+; GFX9-NEXT:    v_mad_i32_i24 v2, s0, v2, v3
+; GFX9-NEXT:    v_mov_b32_e32 v4, s7
+; GFX9-NEXT:    v_mad_i32_i24 v2, s6, v4, v2
+; GFX9-NEXT:    s_bfe_i32 s8, s2, 0x40008
+; GFX9-NEXT:    v_mov_b32_e32 v4, s9
+; GFX9-NEXT:    s_bfe_i32 s11, s4, 0x4000c
+; GFX9-NEXT:    v_mad_i32_i24 v2, s8, v4, v2
+; GFX9-NEXT:    s_bfe_i32 s10, s2, 0x4000c
+; GFX9-NEXT:    v_mov_b32_e32 v4, s11
+; GFX9-NEXT:    s_bfe_i32 s13, s4, 0x40010
+; GFX9-NEXT:    v_mad_i32_i24 v2, s10, v4, v2
+; GFX9-NEXT:    s_bfe_i32 s12, s2, 0x40010
+; GFX9-NEXT:    v_mov_b32_e32 v4, s13
+; GFX9-NEXT:    s_bfe_i32 s15, s4, 0x40014
+; GFX9-NEXT:    s_bfe_i32 s17, s4, 0x40018
+; GFX9-NEXT:    v_mad_i32_i24 v2, s12, v4, v2
+; GFX9-NEXT:    s_bfe_i32 s14, s2, 0x40014
+; GFX9-NEXT:    v_mov_b32_e32 v4, s15
+; GFX9-NEXT:    s_bfe_i32 s16, s2, 0x40018
+; GFX9-NEXT:    v_mad_i32_i24 v2, s14, v4, v2
+; GFX9-NEXT:    v_mov_b32_e32 v4, s17
+; GFX9-NEXT:    s_ashr_i32 s4, s4, 28
+; GFX9-NEXT:    v_mad_i32_i24 v2, s16, v4, v2
+; GFX9-NEXT:    s_ashr_i32 s2, s2, 28
+; GFX9-NEXT:    v_mov_b32_e32 v4, s4
+; GFX9-NEXT:    v_mad_i32_i24 v2, s2, v4, v2
+; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: idot8_multiuses_mul1:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_load_dword s5, s[0:1], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_bfe_i32 s0, s2, 0x40000
+; GFX9-DL-NEXT:    s_bfe_i32 s1, s4, 0x40000
+; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-DL-NEXT:    v_mad_i32_i24 v3, s0, v2, v3
+; GFX9-DL-NEXT:    s_bfe_i32 s7, s4, 0x40004
+; GFX9-DL-NEXT:    s_bfe_i32 s6, s2, 0x40004
+; GFX9-DL-NEXT:    s_bfe_i32 s9, s4, 0x40008
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s0, v2, v3
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s7
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s6, v4, v2
+; GFX9-DL-NEXT:    s_bfe_i32 s8, s2, 0x40008
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s9
+; GFX9-DL-NEXT:    s_bfe_i32 s11, s4, 0x4000c
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s8, v4, v2
+; GFX9-DL-NEXT:    s_bfe_i32 s10, s2, 0x4000c
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s11
+; GFX9-DL-NEXT:    s_bfe_i32 s13, s4, 0x40010
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s10, v4, v2
+; GFX9-DL-NEXT:    s_bfe_i32 s12, s2, 0x40010
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s13
+; GFX9-DL-NEXT:    s_bfe_i32 s15, s4, 0x40014
+; GFX9-DL-NEXT:    s_bfe_i32 s17, s4, 0x40018
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s12, v4, v2
+; GFX9-DL-NEXT:    s_bfe_i32 s14, s2, 0x40014
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s15
+; GFX9-DL-NEXT:    s_bfe_i32 s16, s2, 0x40018
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s14, v4, v2
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s17
+; GFX9-DL-NEXT:    s_ashr_i32 s4, s4, 28
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s16, v4, v2
+; GFX9-DL-NEXT:    s_ashr_i32 s2, s2, 28
+; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s4
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s2, v4, v2
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v3, v2
+; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                                <8 x i4> addrspace(1)* %src2,
+                                                i32 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
+  %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
+
+  %v1e0 = extractelement <8 x i4> %vec1, i64 0
+  %cv1e0 = sext i4 %v1e0 to i32
+  %v2e0 = extractelement <8 x i4> %vec2, i64 0
+  %cv2e0 = sext i4 %v2e0 to i32
+  %mul0 = mul nuw nsw i32 %cv1e0, %cv2e0
+
+  %v1e1 = extractelement <8 x i4> %vec1, i64 1
+  %cv1e1 = sext i4 %v1e1 to i32
+  %v2e1 = extractelement <8 x i4> %vec2, i64 1
+  %cv2e1 = sext i4 %v2e1 to i32
+  %mul1 = mul nuw nsw i32 %cv1e1, %cv2e1
+
+  %v1e2 = extractelement <8 x i4> %vec1, i64 2
+  %cv1e2 = sext i4 %v1e2 to i32
+  %v2e2 = extractelement <8 x i4> %vec2, i64 2
+  %cv2e2 = sext i4 %v2e2 to i32
+  %mul2 = mul nuw nsw i32 %cv1e2, %cv2e2
+
+  %v1e3 = extractelement <8 x i4> %vec1, i64 3
+  %cv1e3 = sext i4 %v1e3 to i32
+  %v2e3 = extractelement <8 x i4> %vec2, i64 3
+  %cv2e3 = sext i4 %v2e3 to i32
+  %mul3 = mul nuw nsw i32 %cv1e3, %cv2e3
+
+  %v1e4 = extractelement <8 x i4> %vec1, i64 4
+  %cv1e4 = sext i4 %v1e4 to i32
+  %v2e4 = extractelement <8 x i4> %vec2, i64 4
+  %cv2e4 = sext i4 %v2e4 to i32
+  %mul4 = mul nuw nsw i32 %cv1e4, %cv2e4
+
+  %v1e5 = extractelement <8 x i4> %vec1, i64 5
+  %cv1e5 = sext i4 %v1e5 to i32
+  %v2e5 = extractelement <8 x i4> %vec2, i64 5
+  %cv2e5 = sext i4 %v2e5 to i32
+  %mul5 = mul nuw nsw i32 %cv1e5, %cv2e5
+
+  %v1e6 = extractelement <8 x i4> %vec1, i64 6
+  %cv1e6 = sext i4 %v1e6 to i32
+  %v2e6 = extractelement <8 x i4> %vec2, i64 6
+  %cv2e6 = sext i4 %v2e6 to i32
+  %mul6 = mul nuw nsw i32 %cv1e6, %cv2e6
+
+  %v1e7 = extractelement <8 x i4> %vec1, i64 7
+  %cv1e7 = sext i4 %v1e7 to i32
+  %v2e7 = extractelement <8 x i4> %vec2, i64 7
+  %cv2e7 = sext i4 %v2e7 to i32
+  %mul7 = mul nuw nsw i32 %cv1e7, %cv2e7
+
+  %acc = load i32, i32 addrspace(1)* %dst, align 4
+  %add =  add i32  %mul0, %acc
+  %add1 = add i32 %mul0, %add
+  %add2 = add i32 %add1, %mul1
+  %add3 = add i32 %add2, %mul2
+  %add4 = add i32 %add3, %mul3
+  %add5 = add i32 %add4, %mul4
+  %add6 = add i32 %add5, %mul5
+  %add7 = add i32 %add6, %mul6
+  %add8 = add i32 %add7, %mul7
+
+  %res = add i32 %add, %add8
+  store i32 %res, i32 addrspace(1)* %dst, align 4
+  ret void
+}
+
+; TODO: Support this pattern.
+define amdgpu_kernel void @idot8_acc32_vecMul(<8 x i4> addrspace(1)* %src1,
+; GFX7-LABEL: idot8_acc32_vecMul:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s0, 0
+; GFX7-NEXT:    s_mov_b32 s12, s0
+; GFX7-NEXT:    s_mov_b32 s14, s0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s1, s[8:9], 0x0
+; GFX7-NEXT:    s_load_dword s2, s[10:11], 0x0
+; GFX7-NEXT:    s_load_dword s38, s[4:5], 0x0
+; GFX7-NEXT:    s_mov_b32 s16, s0
+; GFX7-NEXT:    s_mov_b32 s18, s0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_lshl_b32 s13, s1, 4
+; GFX7-NEXT:    s_lshl_b32 s15, s1, 12
+; GFX7-NEXT:    s_lshl_b32 s17, s1, 16
+; GFX7-NEXT:    s_lshl_b32 s19, s1, 20
+; GFX7-NEXT:    s_ashr_i64 s[10:11], s[12:13], 60
+; GFX7-NEXT:    s_lshl_b32 s13, s1, 8
+; GFX7-NEXT:    s_ashr_i64 s[8:9], s[0:1], 60
+; GFX7-NEXT:    s_lshl_b32 s21, s1, 24
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 28
+; GFX7-NEXT:    s_ashr_i64 s[22:23], s[0:1], 60
+; GFX7-NEXT:    s_mov_b32 s1, s2
+; GFX7-NEXT:    s_ashr_i64 s[24:25], s[0:1], 60
+; GFX7-NEXT:    s_lshl_b32 s1, s2, 4
+; GFX7-NEXT:    s_ashr_i64 s[26:27], s[0:1], 60
+; GFX7-NEXT:    s_lshl_b32 s1, s2, 8
+; GFX7-NEXT:    s_ashr_i64 s[28:29], s[0:1], 60
+; GFX7-NEXT:    s_lshl_b32 s1, s2, 12
+; GFX7-NEXT:    s_ashr_i64 s[30:31], s[0:1], 60
+; GFX7-NEXT:    s_lshl_b32 s1, s2, 16
+; GFX7-NEXT:    s_ashr_i64 s[32:33], s[0:1], 60
+; GFX7-NEXT:    s_lshl_b32 s1, s2, 20
+; GFX7-NEXT:    s_ashr_i64 s[34:35], s[0:1], 60
+; GFX7-NEXT:    s_lshl_b32 s1, s2, 24
+; GFX7-NEXT:    s_ashr_i64 s[36:37], s[0:1], 60
+; GFX7-NEXT:    s_lshl_b32 s1, s2, 28
+; GFX7-NEXT:    s_mov_b32 s20, s0
+; GFX7-NEXT:    s_ashr_i64 s[0:1], s[0:1], 60
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s38
+; GFX7-NEXT:    v_mad_i32_i24 v0, s22, v0, v1
+; GFX7-NEXT:    s_ashr_i64 s[20:21], s[20:21], 60
+; GFX7-NEXT:    v_mov_b32_e32 v1, s36
+; GFX7-NEXT:    v_mad_i32_i24 v0, s20, v1, v0
+; GFX7-NEXT:    s_ashr_i64 s[18:19], s[18:19], 60
+; GFX7-NEXT:    v_mov_b32_e32 v1, s34
+; GFX7-NEXT:    v_mad_i32_i24 v0, s18, v1, v0
+; GFX7-NEXT:    s_ashr_i64 s[16:17], s[16:17], 60
+; GFX7-NEXT:    v_mov_b32_e32 v1, s32
+; GFX7-NEXT:    v_mad_i32_i24 v0, s16, v1, v0
+; GFX7-NEXT:    s_ashr_i64 s[14:15], s[14:15], 60
+; GFX7-NEXT:    v_mov_b32_e32 v1, s30
+; GFX7-NEXT:    v_mad_i32_i24 v0, s14, v1, v0
+; GFX7-NEXT:    s_ashr_i64 s[12:13], s[12:13], 60
+; GFX7-NEXT:    v_mov_b32_e32 v1, s28
+; GFX7-NEXT:    v_mad_i32_i24 v0, s12, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s26
+; GFX7-NEXT:    v_mad_i32_i24 v0, s10, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s24
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    v_mad_i32_i24 v0, s8, v1, v0
+; GFX7-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: idot8_acc32_vecMul:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_mov_b32 s8, 0
+; GFX8-NEXT:    s_mov_b32 s10, s8
+; GFX8-NEXT:    s_mov_b32 s12, s8
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s9, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX8-NEXT:    s_load_dword s36, s[0:1], 0x0
+; GFX8-NEXT:    s_mov_b32 s14, s8
+; GFX8-NEXT:    s_mov_b32 s16, s8
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshl_b32 s11, s9, 4
+; GFX8-NEXT:    s_lshl_b32 s13, s9, 8
+; GFX8-NEXT:    s_lshl_b32 s15, s9, 16
+; GFX8-NEXT:    s_lshl_b32 s17, s9, 20
+; GFX8-NEXT:    s_ashr_i64 s[6:7], s[10:11], 60
+; GFX8-NEXT:    s_ashr_i64 s[10:11], s[12:13], 60
+; GFX8-NEXT:    s_lshl_b32 s13, s9, 12
+; GFX8-NEXT:    s_ashr_i64 s[4:5], s[8:9], 60
+; GFX8-NEXT:    s_lshl_b32 s19, s9, 24
+; GFX8-NEXT:    s_lshl_b32 s9, s9, 28
+; GFX8-NEXT:    s_ashr_i64 s[20:21], s[8:9], 60
+; GFX8-NEXT:    s_mov_b32 s9, s2
+; GFX8-NEXT:    s_ashr_i64 s[22:23], s[8:9], 60
+; GFX8-NEXT:    s_lshl_b32 s9, s2, 4
+; GFX8-NEXT:    s_ashr_i64 s[24:25], s[8:9], 60
+; GFX8-NEXT:    s_lshl_b32 s9, s2, 8
+; GFX8-NEXT:    s_ashr_i64 s[26:27], s[8:9], 60
+; GFX8-NEXT:    s_lshl_b32 s9, s2, 12
+; GFX8-NEXT:    s_ashr_i64 s[28:29], s[8:9], 60
+; GFX8-NEXT:    s_lshl_b32 s9, s2, 16
+; GFX8-NEXT:    s_ashr_i64 s[30:31], s[8:9], 60
+; GFX8-NEXT:    s_lshl_b32 s9, s2, 20
+; GFX8-NEXT:    s_ashr_i64 s[32:33], s[8:9], 60
+; GFX8-NEXT:    s_lshl_b32 s9, s2, 24
+; GFX8-NEXT:    s_ashr_i64 s[34:35], s[8:9], 60
+; GFX8-NEXT:    s_lshl_b32 s9, s2, 28
+; GFX8-NEXT:    s_mov_b32 s18, s8
+; GFX8-NEXT:    s_ashr_i64 s[8:9], s[8:9], 60
+; GFX8-NEXT:    v_mov_b32_e32 v0, s8
+; GFX8-NEXT:    v_mov_b32_e32 v1, s36
+; GFX8-NEXT:    v_mad_i32_i24 v0, s20, v0, v1
+; GFX8-NEXT:    s_ashr_i64 s[18:19], s[18:19], 60
+; GFX8-NEXT:    v_mov_b32_e32 v1, s34
+; GFX8-NEXT:    v_mad_i32_i24 v0, s18, v1, v0
+; GFX8-NEXT:    s_ashr_i64 s[16:17], s[16:17], 60
+; GFX8-NEXT:    v_mov_b32_e32 v1, s32
+; GFX8-NEXT:    v_mad_i32_i24 v0, s16, v1, v0
+; GFX8-NEXT:    s_ashr_i64 s[14:15], s[14:15], 60
+; GFX8-NEXT:    v_mov_b32_e32 v1, s30
+; GFX8-NEXT:    v_mad_i32_i24 v0, s14, v1, v0
+; GFX8-NEXT:    s_ashr_i64 s[12:13], s[12:13], 60
+; GFX8-NEXT:    v_mov_b32_e32 v1, s28
+; GFX8-NEXT:    v_mad_i32_i24 v0, s12, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s26
+; GFX8-NEXT:    v_mad_i32_i24 v0, s10, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s24
+; GFX8-NEXT:    v_mad_i32_i24 v0, s6, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s22
+; GFX8-NEXT:    v_mad_i32_i24 v2, s4, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: idot8_acc32_vecMul:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NEXT:    s_mov_b32 s8, 0
+; GFX9-NEXT:    s_mov_b32 s10, s8
+; GFX9-NEXT:    s_mov_b32 s12, s8
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_load_dword s9, s[4:5], 0x0
+; GFX9-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX9-NEXT:    s_load_dword s36, s[0:1], 0x0
+; GFX9-NEXT:    s_mov_b32 s14, s8
+; GFX9-NEXT:    s_mov_b32 s16, s8
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshl_b32 s11, s9, 4
+; GFX9-NEXT:    s_lshl_b32 s13, s9, 8
+; GFX9-NEXT:    s_lshl_b32 s15, s9, 16
+; GFX9-NEXT:    s_lshl_b32 s17, s9, 20
+; GFX9-NEXT:    s_ashr_i64 s[6:7], s[10:11], 60
+; GFX9-NEXT:    s_ashr_i64 s[10:11], s[12:13], 60
+; GFX9-NEXT:    s_lshl_b32 s13, s9, 12
+; GFX9-NEXT:    s_ashr_i64 s[4:5], s[8:9], 60
+; GFX9-NEXT:    s_lshl_b32 s19, s9, 24
+; GFX9-NEXT:    s_lshl_b32 s9, s9, 28
+; GFX9-NEXT:    s_ashr_i64 s[20:21], s[8:9], 60
+; GFX9-NEXT:    s_mov_b32 s9, s2
+; GFX9-NEXT:    s_ashr_i64 s[22:23], s[8:9], 60
+; GFX9-NEXT:    s_lshl_b32 s9, s2, 4
+; GFX9-NEXT:    s_ashr_i64 s[24:25], s[8:9], 60
+; GFX9-NEXT:    s_lshl_b32 s9, s2, 8
+; GFX9-NEXT:    s_ashr_i64 s[26:27], s[8:9], 60
+; GFX9-NEXT:    s_lshl_b32 s9, s2, 12
+; GFX9-NEXT:    s_ashr_i64 s[28:29], s[8:9], 60
+; GFX9-NEXT:    s_lshl_b32 s9, s2, 16
+; GFX9-NEXT:    s_ashr_i64 s[30:31], s[8:9], 60
+; GFX9-NEXT:    s_lshl_b32 s9, s2, 20
+; GFX9-NEXT:    s_ashr_i64 s[32:33], s[8:9], 60
+; GFX9-NEXT:    s_lshl_b32 s9, s2, 24
+; GFX9-NEXT:    s_ashr_i64 s[34:35], s[8:9], 60
+; GFX9-NEXT:    s_lshl_b32 s9, s2, 28
+; GFX9-NEXT:    s_mov_b32 s18, s8
+; GFX9-NEXT:    s_ashr_i64 s[8:9], s[8:9], 60
+; GFX9-NEXT:    v_mov_b32_e32 v0, s8
+; GFX9-NEXT:    v_mov_b32_e32 v1, s36
+; GFX9-NEXT:    v_mad_i32_i24 v0, s20, v0, v1
+; GFX9-NEXT:    s_ashr_i64 s[18:19], s[18:19], 60
+; GFX9-NEXT:    v_mov_b32_e32 v1, s34
+; GFX9-NEXT:    v_mad_i32_i24 v0, s18, v1, v0
+; GFX9-NEXT:    s_ashr_i64 s[16:17], s[16:17], 60
+; GFX9-NEXT:    v_mov_b32_e32 v1, s32
+; GFX9-NEXT:    v_mad_i32_i24 v0, s16, v1, v0
+; GFX9-NEXT:    s_ashr_i64 s[14:15], s[14:15], 60
+; GFX9-NEXT:    v_mov_b32_e32 v1, s30
+; GFX9-NEXT:    v_mad_i32_i24 v0, s14, v1, v0
+; GFX9-NEXT:    s_ashr_i64 s[12:13], s[12:13], 60
+; GFX9-NEXT:    v_mov_b32_e32 v1, s28
+; GFX9-NEXT:    v_mad_i32_i24 v0, s12, v1, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s26
+; GFX9-NEXT:    v_mad_i32_i24 v0, s10, v1, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s24
+; GFX9-NEXT:    v_mad_i32_i24 v0, s6, v1, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s22
+; GFX9-NEXT:    v_mad_i32_i24 v2, s4, v1, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: idot8_acc32_vecMul:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_mov_b32 s8, 0
+; GFX9-DL-NEXT:    s_mov_b32 s10, s8
+; GFX9-DL-NEXT:    s_mov_b32 s12, s8
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s9, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s2, s[6:7], 0x0
+; GFX9-DL-NEXT:    s_load_dword s36, s[0:1], 0x0
+; GFX9-DL-NEXT:    s_mov_b32 s14, s8
+; GFX9-DL-NEXT:    s_mov_b32 s16, s8
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_lshl_b32 s11, s9, 4
+; GFX9-DL-NEXT:    s_lshl_b32 s13, s9, 8
+; GFX9-DL-NEXT:    s_lshl_b32 s15, s9, 16
+; GFX9-DL-NEXT:    s_lshl_b32 s17, s9, 20
+; GFX9-DL-NEXT:    s_ashr_i64 s[6:7], s[10:11], 60
+; GFX9-DL-NEXT:    s_ashr_i64 s[10:11], s[12:13], 60
+; GFX9-DL-NEXT:    s_lshl_b32 s13, s9, 12
+; GFX9-DL-NEXT:    s_ashr_i64 s[4:5], s[8:9], 60
+; GFX9-DL-NEXT:    s_lshl_b32 s19, s9, 24
+; GFX9-DL-NEXT:    s_lshl_b32 s9, s9, 28
+; GFX9-DL-NEXT:    s_ashr_i64 s[20:21], s[8:9], 60
+; GFX9-DL-NEXT:    s_mov_b32 s9, s2
+; GFX9-DL-NEXT:    s_ashr_i64 s[22:23], s[8:9], 60
+; GFX9-DL-NEXT:    s_lshl_b32 s9, s2, 4
+; GFX9-DL-NEXT:    s_ashr_i64 s[24:25], s[8:9], 60
+; GFX9-DL-NEXT:    s_lshl_b32 s9, s2, 8
+; GFX9-DL-NEXT:    s_ashr_i64 s[26:27], s[8:9], 60
+; GFX9-DL-NEXT:    s_lshl_b32 s9, s2, 12
+; GFX9-DL-NEXT:    s_ashr_i64 s[28:29], s[8:9], 60
+; GFX9-DL-NEXT:    s_lshl_b32 s9, s2, 16
+; GFX9-DL-NEXT:    s_ashr_i64 s[30:31], s[8:9], 60
+; GFX9-DL-NEXT:    s_lshl_b32 s9, s2, 20
+; GFX9-DL-NEXT:    s_ashr_i64 s[32:33], s[8:9], 60
+; GFX9-DL-NEXT:    s_lshl_b32 s9, s2, 24
+; GFX9-DL-NEXT:    s_ashr_i64 s[34:35], s[8:9], 60
+; GFX9-DL-NEXT:    s_lshl_b32 s9, s2, 28
+; GFX9-DL-NEXT:    s_mov_b32 s18, s8
+; GFX9-DL-NEXT:    s_ashr_i64 s[8:9], s[8:9], 60
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s8
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s36
+; GFX9-DL-NEXT:    v_mad_i32_i24 v0, s20, v0, v1
+; GFX9-DL-NEXT:    s_ashr_i64 s[18:19], s[18:19], 60
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s34
+; GFX9-DL-NEXT:    v_mad_i32_i24 v0, s18, v1, v0
+; GFX9-DL-NEXT:    s_ashr_i64 s[16:17], s[16:17], 60
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s32
+; GFX9-DL-NEXT:    v_mad_i32_i24 v0, s16, v1, v0
+; GFX9-DL-NEXT:    s_ashr_i64 s[14:15], s[14:15], 60
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s30
+; GFX9-DL-NEXT:    v_mad_i32_i24 v0, s14, v1, v0
+; GFX9-DL-NEXT:    s_ashr_i64 s[12:13], s[12:13], 60
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s28
+; GFX9-DL-NEXT:    v_mad_i32_i24 v0, s12, v1, v0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s26
+; GFX9-DL-NEXT:    v_mad_i32_i24 v0, s10, v1, v0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s24
+; GFX9-DL-NEXT:    v_mad_i32_i24 v0, s6, v1, v0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s22
+; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s4, v1, v0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                              <8 x i4> addrspace(1)* %src2,
+                                              i32 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
+  %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
+
+  %cvec1 = sext <8 x i4> %vec1 to <8 x i32>
+  %cvec2 = sext <8 x i4> %vec2 to <8 x i32>
+
+  %mul = mul <8 x i32> %cvec1, %cvec2
+  %mul0 = extractelement <8 x i32> %mul, i64 0
+  %mul1 = extractelement <8 x i32> %mul, i64 1
+  %mul2 = extractelement <8 x i32> %mul, i64 2
+  %mul3 = extractelement <8 x i32> %mul, i64 3
+  %mul4 = extractelement <8 x i32> %mul, i64 4
+  %mul5 = extractelement <8 x i32> %mul, i64 5
+  %mul6 = extractelement <8 x i32> %mul, i64 6
+  %mul7 = extractelement <8 x i32> %mul, i64 7
+
+  %acc = load i32, i32 addrspace(1)* %dst, align 4
+  %add1 = add i32 %mul0, %acc
+  %add2 = add i32 %add1, %mul1
+  %add3 = add i32 %add2, %mul2
+  %add4 = add i32 %add3, %mul3
+  %add5 = add i32 %add4, %mul4
+  %add6 = add i32 %add5, %mul5
+  %add7 = add i32 %add6, %mul6
+  %add8 = add i32 %add7, %mul7
+
+  store i32 %add8, i32 addrspace(1)* %dst, align 4
+  ret void
+}
+
+; TODO: Support this pattern.
+define amdgpu_kernel void @idot8_acc16_vecMul(<8 x i4> addrspace(1)* %src1,
+; GFX7-LABEL: idot8_acc16_vecMul:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s1, s[8:9], 0x0
+; GFX7-NEXT:    buffer_load_ushort v0, off, s[4:7], 0
+; GFX7-NEXT:    s_load_dword s2, s[10:11], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_bfe_i32 s8, s1, 0x40010
+; GFX7-NEXT:    s_bfe_i32 s9, s1, 0x40014
+; GFX7-NEXT:    s_bfe_i32 s15, s2, 0x40010
+; GFX7-NEXT:    s_bfe_i32 s16, s2, 0x40014
+; GFX7-NEXT:    s_bfe_i32 s17, s2, 0x40018
+; GFX7-NEXT:    s_ashr_i32 s18, s2, 28
+; GFX7-NEXT:    s_bfe_i32 s19, s2, 0x40000
+; GFX7-NEXT:    s_bfe_i32 s20, s2, 0x40004
+; GFX7-NEXT:    s_bfe_i32 s21, s2, 0x40008
+; GFX7-NEXT:    s_bfe_i32 s2, s2, 0x4000c
+; GFX7-NEXT:    s_bfe_i32 s10, s1, 0x40018
+; GFX7-NEXT:    s_ashr_i32 s11, s1, 28
+; GFX7-NEXT:    s_bfe_i32 s12, s1, 0x40000
+; GFX7-NEXT:    v_mov_b32_e32 v4, s19
+; GFX7-NEXT:    s_bfe_i32 s13, s1, 0x40004
+; GFX7-NEXT:    v_mov_b32_e32 v3, s20
+; GFX7-NEXT:    s_bfe_i32 s14, s1, 0x40008
+; GFX7-NEXT:    v_mov_b32_e32 v2, s21
+; GFX7-NEXT:    s_bfe_i32 s1, s1, 0x4000c
+; GFX7-NEXT:    v_mov_b32_e32 v1, s2
+; GFX7-NEXT:    v_mov_b32_e32 v5, s18
+; GFX7-NEXT:    v_mov_b32_e32 v6, s17
+; GFX7-NEXT:    v_mul_i32_i24_e32 v1, s1, v1
+; GFX7-NEXT:    v_mul_i32_i24_e32 v2, s14, v2
+; GFX7-NEXT:    v_mul_i32_i24_e32 v3, s13, v3
+; GFX7-NEXT:    v_mul_i32_i24_e32 v4, s12, v4
+; GFX7-NEXT:    v_mul_i32_i24_e32 v5, s11, v5
+; GFX7-NEXT:    v_mul_i32_i24_e32 v6, s10, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX7-NEXT:    v_and_b32_e32 v4, s0, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX7-NEXT:    v_or_b32_e32 v2, v4, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 16, v5
+; GFX7-NEXT:    v_and_b32_e32 v6, s0, v6
+; GFX7-NEXT:    v_mov_b32_e32 v7, s16
+; GFX7-NEXT:    v_mov_b32_e32 v8, s15
+; GFX7-NEXT:    v_or_b32_e32 v3, v6, v5
+; GFX7-NEXT:    v_alignbit_b32 v5, v1, v2, 16
+; GFX7-NEXT:    v_mul_i32_i24_e32 v7, s9, v7
+; GFX7-NEXT:    v_mul_i32_i24_e32 v8, s8, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
+; GFX7-NEXT:    v_and_b32_e32 v8, s0, v8
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 16, v1
+; GFX7-NEXT:    v_or_b32_e32 v4, v8, v7
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 16, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 16, v3
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v5, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v6, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v7, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v3, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v8, v0
+; GFX7-NEXT:    buffer_store_short v0, off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: idot8_acc16_vecMul:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ushort v2, v[0:1]
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_lshlrev_b16_e64 v3, 12, s2
+; GFX8-NEXT:    v_lshlrev_b16_e64 v4, 12, s4
+; GFX8-NEXT:    s_lshr_b32 s0, s2, 4
+; GFX8-NEXT:    s_lshr_b32 s1, s2, 8
+; GFX8-NEXT:    s_lshr_b32 s5, s4, 4
+; GFX8-NEXT:    s_lshr_b32 s6, s4, 8
+; GFX8-NEXT:    v_lshlrev_b16_e64 v5, 12, s1
+; GFX8-NEXT:    v_lshlrev_b16_e64 v6, 12, s0
+; GFX8-NEXT:    v_lshlrev_b16_e64 v7, 12, s6
+; GFX8-NEXT:    v_lshlrev_b16_e64 v8, 12, s5
+; GFX8-NEXT:    v_ashrrev_i16_e32 v3, 12, v3
+; GFX8-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
+; GFX8-NEXT:    s_lshr_b32 s0, s2, 12
+; GFX8-NEXT:    s_lshr_b32 s1, s4, 12
+; GFX8-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
+; GFX8-NEXT:    v_ashrrev_i16_e32 v5, 12, v5
+; GFX8-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
+; GFX8-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
+; GFX8-NEXT:    v_lshlrev_b16_e64 v9, 12, s0
+; GFX8-NEXT:    v_lshlrev_b16_e64 v10, 12, s1
+; GFX8-NEXT:    s_lshr_b32 s5, s2, 16
+; GFX8-NEXT:    s_lshr_b32 s6, s4, 16
+; GFX8-NEXT:    v_mul_u32_u24_e32 v5, v5, v7
+; GFX8-NEXT:    v_lshlrev_b16_e64 v11, 12, s5
+; GFX8-NEXT:    v_lshlrev_b16_e64 v12, 12, s6
+; GFX8-NEXT:    s_lshr_b32 s0, s2, 20
+; GFX8-NEXT:    s_lshr_b32 s1, s4, 20
+; GFX8-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX8-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX8-NEXT:    v_lshlrev_b16_e64 v13, 12, s0
+; GFX8-NEXT:    v_lshlrev_b16_e64 v14, 12, s1
+; GFX8-NEXT:    s_lshr_b32 s5, s2, 24
+; GFX8-NEXT:    s_lshr_b32 s6, s4, 24
+; GFX8-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
+; GFX8-NEXT:    v_ashrrev_i16_e32 v12, 12, v12
+; GFX8-NEXT:    v_lshlrev_b16_e64 v15, 12, s5
+; GFX8-NEXT:    v_lshlrev_b16_e64 v17, 12, s6
+; GFX8-NEXT:    s_lshr_b32 s0, s2, 28
+; GFX8-NEXT:    s_lshr_b32 s1, s4, 28
+; GFX8-NEXT:    v_ashrrev_i16_e32 v13, 12, v13
+; GFX8-NEXT:    v_ashrrev_i16_e32 v14, 12, v14
+; GFX8-NEXT:    v_lshlrev_b16_e64 v16, 12, s0
+; GFX8-NEXT:    v_lshlrev_b16_e64 v18, 12, s1
+; GFX8-NEXT:    v_ashrrev_i16_e32 v15, 12, v15
+; GFX8-NEXT:    v_ashrrev_i16_e32 v17, 12, v17
+; GFX8-NEXT:    v_ashrrev_i16_e32 v16, 12, v16
+; GFX8-NEXT:    v_ashrrev_i16_e32 v18, 12, v18
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mad_u32_u24 v2, v3, v4, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, v6, v8, v2
+; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v5, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
+; GFX8-NEXT:    v_mad_u32_u24 v2, v9, v10, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, v11, v12, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, v13, v14, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, v15, v17, v2
+; GFX8-NEXT:    v_mad_u32_u24 v2, v16, v18, v2
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: idot8_acc16_vecMul:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    global_load_ushort v2, v[0:1], off
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_and_b32 s0, s2, 15
+; GFX9-NEXT:    s_bfe_u32 s1, s2, 0x40004
+; GFX9-NEXT:    s_and_b32 s5, s4, 15
+; GFX9-NEXT:    s_bfe_u32 s6, s4, 0x40004
+; GFX9-NEXT:    s_pack_ll_b32_b16 s0, s0, s1
+; GFX9-NEXT:    s_pack_ll_b32_b16 s1, s5, s6
+; GFX9-NEXT:    s_bfe_u32 s5, s2, 0x40008
+; GFX9-NEXT:    s_bfe_u32 s6, s2, 0x4000c
+; GFX9-NEXT:    v_pk_lshlrev_b16 v3, 12, s0 op_sel_hi:[0,1]
+; GFX9-NEXT:    s_pack_ll_b32_b16 s0, s5, s6
+; GFX9-NEXT:    s_bfe_u32 s9, s2, 0x40010
+; GFX9-NEXT:    s_bfe_u32 s10, s2, 0x40014
+; GFX9-NEXT:    v_pk_lshlrev_b16 v4, 12, s0 op_sel_hi:[0,1]
+; GFX9-NEXT:    s_bfe_u32 s13, s2, 0x40018
+; GFX9-NEXT:    s_pack_ll_b32_b16 s0, s9, s10
+; GFX9-NEXT:    s_lshr_b32 s2, s2, 28
+; GFX9-NEXT:    v_pk_lshlrev_b16 v5, 12, s0 op_sel_hi:[0,1]
+; GFX9-NEXT:    v_pk_lshlrev_b16 v7, 12, s1 op_sel_hi:[0,1]
+; GFX9-NEXT:    s_pack_ll_b32_b16 s0, s13, s2
+; GFX9-NEXT:    s_bfe_u32 s7, s4, 0x40008
+; GFX9-NEXT:    s_bfe_u32 s8, s4, 0x4000c
+; GFX9-NEXT:    v_pk_lshlrev_b16 v6, 12, s0 op_sel_hi:[0,1]
+; GFX9-NEXT:    s_pack_ll_b32_b16 s0, s7, s8
+; GFX9-NEXT:    v_pk_ashrrev_i16 v3, 12, v3 op_sel_hi:[0,1]
+; GFX9-NEXT:    v_pk_ashrrev_i16 v7, 12, v7 op_sel_hi:[0,1]
+; GFX9-NEXT:    v_pk_lshlrev_b16 v8, 12, s0 op_sel_hi:[0,1]
+; GFX9-NEXT:    v_pk_mul_lo_u16 v3, v3, v7
+; GFX9-NEXT:    s_bfe_u32 s11, s4, 0x40010
+; GFX9-NEXT:    s_bfe_u32 s12, s4, 0x40014
+; GFX9-NEXT:    s_pack_ll_b32_b16 s0, s11, s12
+; GFX9-NEXT:    v_pk_ashrrev_i16 v4, 12, v4 op_sel_hi:[0,1]
+; GFX9-NEXT:    v_pk_ashrrev_i16 v8, 12, v8 op_sel_hi:[0,1]
+; GFX9-NEXT:    v_pk_lshlrev_b16 v9, 12, s0 op_sel_hi:[0,1]
+; GFX9-NEXT:    s_bfe_u32 s14, s4, 0x40018
+; GFX9-NEXT:    s_lshr_b32 s4, s4, 28
+; GFX9-NEXT:    v_pk_mul_lo_u16 v4, v4, v8
+; GFX9-NEXT:    s_pack_ll_b32_b16 s0, s14, s4
+; GFX9-NEXT:    v_pk_ashrrev_i16 v5, 12, v5 op_sel_hi:[0,1]
+; GFX9-NEXT:    v_pk_ashrrev_i16 v9, 12, v9 op_sel_hi:[0,1]
+; GFX9-NEXT:    v_pk_lshlrev_b16 v10, 12, s0 op_sel_hi:[0,1]
+; GFX9-NEXT:    v_pk_mul_lo_u16 v5, v5, v9
+; GFX9-NEXT:    v_pk_ashrrev_i16 v6, 12, v6 op_sel_hi:[0,1]
+; GFX9-NEXT:    v_pk_ashrrev_i16 v10, 12, v10 op_sel_hi:[0,1]
+; GFX9-NEXT:    v_pk_mul_lo_u16 v6, v6, v10
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
+; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
+; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
+; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_add_u32_e32 v2, v2, v6
+; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: idot8_acc16_vecMul:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ushort v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_and_b32 s0, s2, 15
+; GFX9-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
+; GFX9-DL-NEXT:    s_and_b32 s5, s4, 15
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s4, 0x40004
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s0, s0, s1
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s1, s5, s6
+; GFX9-DL-NEXT:    s_bfe_u32 s5, s2, 0x40008
+; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x4000c
+; GFX9-DL-NEXT:    v_pk_lshlrev_b16 v3, 12, s0 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s0, s5, s6
+; GFX9-DL-NEXT:    s_bfe_u32 s9, s2, 0x40010
+; GFX9-DL-NEXT:    s_bfe_u32 s10, s2, 0x40014
+; GFX9-DL-NEXT:    v_pk_lshlrev_b16 v4, 12, s0 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    s_bfe_u32 s13, s2, 0x40018
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s0, s9, s10
+; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 28
+; GFX9-DL-NEXT:    v_pk_lshlrev_b16 v5, 12, s0 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    v_pk_lshlrev_b16 v7, 12, s1 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s0, s13, s2
+; GFX9-DL-NEXT:    s_bfe_u32 s7, s4, 0x40008
+; GFX9-DL-NEXT:    s_bfe_u32 s8, s4, 0x4000c
+; GFX9-DL-NEXT:    v_pk_lshlrev_b16 v6, 12, s0 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s0, s7, s8
+; GFX9-DL-NEXT:    v_pk_ashrrev_i16 v3, 12, v3 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    v_pk_ashrrev_i16 v7, 12, v7 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    v_pk_lshlrev_b16 v8, 12, s0 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v3, v3, v7
+; GFX9-DL-NEXT:    s_bfe_u32 s11, s4, 0x40010
+; GFX9-DL-NEXT:    s_bfe_u32 s12, s4, 0x40014
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s0, s11, s12
+; GFX9-DL-NEXT:    v_pk_ashrrev_i16 v4, 12, v4 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    v_pk_ashrrev_i16 v8, 12, v8 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    v_pk_lshlrev_b16 v9, 12, s0 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    s_bfe_u32 s14, s4, 0x40018
+; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 28
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v4, v4, v8
+; GFX9-DL-NEXT:    s_pack_ll_b32_b16 s0, s14, s4
+; GFX9-DL-NEXT:    v_pk_ashrrev_i16 v5, 12, v5 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    v_pk_ashrrev_i16 v9, 12, v9 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    v_pk_lshlrev_b16 v10, 12, s0 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v5, v5, v9
+; GFX9-DL-NEXT:    v_pk_ashrrev_i16 v6, 12, v6 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    v_pk_ashrrev_i16 v10, 12, v10 op_sel_hi:[0,1]
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v6, v6, v10
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v3, v2
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v5
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v6
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                              <8 x i4> addrspace(1)* %src2,
+                                              i16 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
+  %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
+
+  %cvec1 = sext <8 x i4> %vec1 to <8 x i16>
+  %cvec2 = sext <8 x i4> %vec2 to <8 x i16>
+
+  %mul = mul <8 x i16> %cvec1, %cvec2
+  %mul0 = extractelement <8 x i16> %mul, i64 0
+  %mul1 = extractelement <8 x i16> %mul, i64 1
+  %mul2 = extractelement <8 x i16> %mul, i64 2
+  %mul3 = extractelement <8 x i16> %mul, i64 3
+  %mul4 = extractelement <8 x i16> %mul, i64 4
+  %mul5 = extractelement <8 x i16> %mul, i64 5
+  %mul6 = extractelement <8 x i16> %mul, i64 6
+  %mul7 = extractelement <8 x i16> %mul, i64 7
+
+  %acc = load i16, i16 addrspace(1)* %dst, align 4
+  %add1 = add i16 %mul0, %acc
+  %add2 = add i16 %add1, %mul1
+  %add3 = add i16 %add2, %mul2
+  %add4 = add i16 %add3, %mul3
+  %add5 = add i16 %add4, %mul4
+  %add6 = add i16 %add5, %mul5
+  %add7 = add i16 %add6, %mul6
+  %add8 = add i16 %add7, %mul7
+
+  store i16 %add8, i16 addrspace(1)* %dst, align 4
+  ret void
+}
+
+; TODO: Support this pattern.
+define amdgpu_kernel void @idot8_acc8_vecMul(<8 x i4> addrspace(1)* %src1,
+; GFX7-LABEL: idot8_acc8_vecMul:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_movk_i32 s0, 0xff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s2, s[8:9], 0x0
+; GFX7-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0
+; GFX7-NEXT:    s_load_dword s8, s[10:11], 0x0
+; GFX7-NEXT:    s_mov_b32 s1, 0xffff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_bfe_i32 s9, s2, 0x40000
+; GFX7-NEXT:    s_bfe_i32 s10, s2, 0x40004
+; GFX7-NEXT:    s_bfe_i32 s16, s8, 0x40000
+; GFX7-NEXT:    s_bfe_i32 s17, s8, 0x40004
+; GFX7-NEXT:    s_bfe_i32 s18, s8, 0x40008
+; GFX7-NEXT:    s_bfe_i32 s19, s8, 0x4000c
+; GFX7-NEXT:    s_bfe_i32 s20, s8, 0x40010
+; GFX7-NEXT:    s_bfe_i32 s21, s8, 0x40014
+; GFX7-NEXT:    s_bfe_i32 s22, s8, 0x40018
+; GFX7-NEXT:    s_ashr_i32 s8, s8, 28
+; GFX7-NEXT:    v_mov_b32_e32 v7, s17
+; GFX7-NEXT:    v_mov_b32_e32 v8, s16
+; GFX7-NEXT:    s_bfe_i32 s11, s2, 0x40008
+; GFX7-NEXT:    v_mov_b32_e32 v6, s18
+; GFX7-NEXT:    s_bfe_i32 s12, s2, 0x4000c
+; GFX7-NEXT:    v_mov_b32_e32 v5, s19
+; GFX7-NEXT:    s_bfe_i32 s13, s2, 0x40010
+; GFX7-NEXT:    v_mov_b32_e32 v4, s20
+; GFX7-NEXT:    s_bfe_i32 s14, s2, 0x40014
+; GFX7-NEXT:    v_mov_b32_e32 v3, s21
+; GFX7-NEXT:    s_bfe_i32 s15, s2, 0x40018
+; GFX7-NEXT:    v_mov_b32_e32 v2, s22
+; GFX7-NEXT:    s_ashr_i32 s2, s2, 28
+; GFX7-NEXT:    v_mov_b32_e32 v1, s8
+; GFX7-NEXT:    v_mul_i32_i24_e32 v1, s2, v1
+; GFX7-NEXT:    v_mul_i32_i24_e32 v2, s15, v2
+; GFX7-NEXT:    v_mul_i32_i24_e32 v3, s14, v3
+; GFX7-NEXT:    v_mul_i32_i24_e32 v4, s13, v4
+; GFX7-NEXT:    v_mul_i32_i24_e32 v5, s12, v5
+; GFX7-NEXT:    v_mul_i32_i24_e32 v6, s11, v6
+; GFX7-NEXT:    v_mul_i32_i24_e32 v7, s10, v7
+; GFX7-NEXT:    v_mul_i32_i24_e32 v8, s9, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
+; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 8, v3
+; GFX7-NEXT:    v_and_b32_e32 v4, s0, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
+; GFX7-NEXT:    v_and_b32_e32 v6, s0, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 8, v7
+; GFX7-NEXT:    v_and_b32_e32 v8, s0, v8
+; GFX7-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX7-NEXT:    v_or_b32_e32 v2, v4, v3
+; GFX7-NEXT:    v_or_b32_e32 v3, v6, v5
+; GFX7-NEXT:    v_or_b32_e32 v4, v8, v7
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX7-NEXT:    v_and_b32_e32 v2, s1, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX7-NEXT:    v_and_b32_e32 v4, s1, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX7-NEXT:    v_or_b32_e32 v2, v4, v3
+; GFX7-NEXT:    v_alignbit_b32 v3, v1, v2, 8
+; GFX7-NEXT:    v_alignbit_b32 v4, v1, v2, 16
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 24, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 8, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 16, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 24, v1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v3, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v5, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v6, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v7, v0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v8, v0
+; GFX7-NEXT:    buffer_store_byte v0, off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX8-LABEL: idot8_acc8_vecMul:
+; GFX8:       ; %bb.0: ; %entry
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX8-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_load_ubyte v2, v[0:1]
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s0, s2, 4
+; GFX8-NEXT:    s_lshr_b32 s1, s2, 12
+; GFX8-NEXT:    s_lshr_b32 s5, s2, 8
+; GFX8-NEXT:    s_lshr_b32 s6, s4, 4
+; GFX8-NEXT:    s_lshr_b32 s7, s4, 12
+; GFX8-NEXT:    s_lshr_b32 s8, s4, 8
+; GFX8-NEXT:    v_lshlrev_b16_e64 v3, 12, s5
+; GFX8-NEXT:    v_lshlrev_b16_e64 v4, 12, s1
+; GFX8-NEXT:    v_lshlrev_b16_e64 v5, 12, s0
+; GFX8-NEXT:    v_lshlrev_b16_e64 v7, 12, s8
+; GFX8-NEXT:    v_lshlrev_b16_e64 v8, 12, s7
+; GFX8-NEXT:    v_lshlrev_b16_e64 v9, 12, s6
+; GFX8-NEXT:    v_lshlrev_b16_e64 v6, 12, s2
+; GFX8-NEXT:    v_lshlrev_b16_e64 v10, 12, s4
+; GFX8-NEXT:    v_ashrrev_i16_e32 v3, 12, v3
+; GFX8-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
+; GFX8-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
+; GFX8-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
+; GFX8-NEXT:    v_ashrrev_i16_e32 v5, 12, v5
+; GFX8-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX8-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
+; GFX8-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX8-NEXT:    v_mul_u32_u24_sdwa v3, v3, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
+; GFX8-NEXT:    v_mul_u32_u24_sdwa v4, v4, v8 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
+; GFX8-NEXT:    v_mul_u32_u24_sdwa v5, v5, v9 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
+; GFX8-NEXT:    v_mul_u32_u24_sdwa v6, v6, v10 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
+; GFX8-NEXT:    s_lshr_b32 s0, s2, 20
+; GFX8-NEXT:    s_lshr_b32 s1, s2, 16
+; GFX8-NEXT:    s_lshr_b32 s5, s2, 28
+; GFX8-NEXT:    s_lshr_b32 s2, s2, 24
+; GFX8-NEXT:    s_lshr_b32 s6, s4, 20
+; GFX8-NEXT:    s_lshr_b32 s7, s4, 16
+; GFX8-NEXT:    s_lshr_b32 s8, s4, 28
+; GFX8-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX8-NEXT:    v_or_b32_sdwa v5, v6, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v3, v5, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b16_e64 v7, 12, s2
+; GFX8-NEXT:    v_lshlrev_b16_e64 v8, 12, s5
+; GFX8-NEXT:    v_lshlrev_b16_e64 v9, 12, s1
+; GFX8-NEXT:    v_lshlrev_b16_e64 v10, 12, s0
+; GFX8-NEXT:    v_lshlrev_b16_e64 v11, 12, s4
+; GFX8-NEXT:    v_lshlrev_b16_e64 v12, 12, s8
+; GFX8-NEXT:    v_lshlrev_b16_e64 v13, 12, s7
+; GFX8-NEXT:    v_lshlrev_b16_e64 v14, 12, s6
+; GFX8-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
+; GFX8-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
+; GFX8-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
+; GFX8-NEXT:    v_ashrrev_i16_e32 v12, 12, v12
+; GFX8-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX8-NEXT:    v_ashrrev_i16_e32 v13, 12, v13
+; GFX8-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX8-NEXT:    v_ashrrev_i16_e32 v14, 12, v14
+; GFX8-NEXT:    v_lshrrev_b32_e32 v5, 8, v3
+; GFX8-NEXT:    v_mul_u32_u24_sdwa v7, v7, v11 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
+; GFX8-NEXT:    v_mul_u32_u24_sdwa v8, v8, v12 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
+; GFX8-NEXT:    v_mul_u32_u24_sdwa v9, v9, v13 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
+; GFX8-NEXT:    v_mul_u32_u24_sdwa v10, v10, v14 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
+; GFX8-NEXT:    v_or_b32_sdwa v9, v9, v10 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v7, v7, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v4, v9, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 8, v4
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v3
+; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v5, v2
+; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_2
+; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
+; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v6
+; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
+; GFX8-NEXT:    flat_store_byte v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: idot8_acc8_vecMul:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b32 s0, s2, 4
+; GFX9-NEXT:    s_lshr_b32 s1, s2, 12
+; GFX9-NEXT:    s_lshr_b32 s5, s2, 8
+; GFX9-NEXT:    s_lshr_b32 s6, s4, 4
+; GFX9-NEXT:    s_lshr_b32 s7, s4, 12
+; GFX9-NEXT:    s_lshr_b32 s8, s4, 8
+; GFX9-NEXT:    v_lshlrev_b16_e64 v3, 12, s5
+; GFX9-NEXT:    v_lshlrev_b16_e64 v4, 12, s1
+; GFX9-NEXT:    v_lshlrev_b16_e64 v5, 12, s0
+; GFX9-NEXT:    v_lshlrev_b16_e64 v7, 12, s8
+; GFX9-NEXT:    v_lshlrev_b16_e64 v8, 12, s7
+; GFX9-NEXT:    v_lshlrev_b16_e64 v9, 12, s6
+; GFX9-NEXT:    v_lshlrev_b16_e64 v6, 12, s2
+; GFX9-NEXT:    v_lshlrev_b16_e64 v10, 12, s4
+; GFX9-NEXT:    v_ashrrev_i16_e32 v3, 12, v3
+; GFX9-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
+; GFX9-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
+; GFX9-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
+; GFX9-NEXT:    v_ashrrev_i16_e32 v5, 12, v5
+; GFX9-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX9-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
+; GFX9-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX9-NEXT:    v_mul_lo_u16_e32 v6, v6, v10
+; GFX9-NEXT:    v_mul_lo_u16_sdwa v5, v5, v9 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT:    v_mul_lo_u16_sdwa v4, v4, v8 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT:    v_mul_lo_u16_e32 v3, v3, v7
+; GFX9-NEXT:    s_lshr_b32 s0, s2, 20
+; GFX9-NEXT:    s_lshr_b32 s1, s2, 16
+; GFX9-NEXT:    s_lshr_b32 s5, s2, 28
+; GFX9-NEXT:    s_lshr_b32 s2, s2, 24
+; GFX9-NEXT:    s_lshr_b32 s6, s4, 20
+; GFX9-NEXT:    s_lshr_b32 s7, s4, 16
+; GFX9-NEXT:    s_lshr_b32 s8, s4, 28
+; GFX9-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX9-NEXT:    v_or_b32_sdwa v5, v6, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT:    v_or_b32_sdwa v3, v5, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX9-NEXT:    v_lshlrev_b16_e64 v9, 12, s2
+; GFX9-NEXT:    v_lshlrev_b16_e64 v10, 12, s5
+; GFX9-NEXT:    v_lshlrev_b16_e64 v11, 12, s1
+; GFX9-NEXT:    v_lshlrev_b16_e64 v12, 12, s0
+; GFX9-NEXT:    v_lshlrev_b16_e64 v13, 12, s4
+; GFX9-NEXT:    v_lshlrev_b16_e64 v14, 12, s8
+; GFX9-NEXT:    v_lshlrev_b16_e64 v15, 12, s7
+; GFX9-NEXT:    v_lshlrev_b16_e64 v16, 12, s6
+; GFX9-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX9-NEXT:    v_ashrrev_i16_e32 v13, 12, v13
+; GFX9-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX9-NEXT:    v_ashrrev_i16_e32 v14, 12, v14
+; GFX9-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
+; GFX9-NEXT:    v_ashrrev_i16_e32 v15, 12, v15
+; GFX9-NEXT:    v_ashrrev_i16_e32 v12, 12, v12
+; GFX9-NEXT:    v_ashrrev_i16_e32 v16, 12, v16
+; GFX9-NEXT:    v_lshrrev_b32_e32 v5, 8, v3
+; GFX9-NEXT:    v_mul_lo_u16_sdwa v12, v12, v16 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT:    v_mul_lo_u16_e32 v11, v11, v15
+; GFX9-NEXT:    v_mul_lo_u16_sdwa v10, v10, v14 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT:    v_mul_lo_u16_e32 v9, v9, v13
+; GFX9-NEXT:    v_or_b32_sdwa v7, v11, v12 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT:    v_or_b32_sdwa v8, v9, v10 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT:    v_or_b32_sdwa v4, v7, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
+; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
+; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_2
+; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
+; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
+; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 8, v4
+; GFX9-NEXT:    v_add_u32_e32 v2, v2, v3
+; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
+; GFX9-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX9-DL-LABEL: idot8_acc8_vecMul:
+; GFX9-DL:       ; %bb.0: ; %entry
+; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
+; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-DL-NEXT:    s_lshr_b32 s0, s2, 4
+; GFX9-DL-NEXT:    s_lshr_b32 s1, s2, 12
+; GFX9-DL-NEXT:    s_lshr_b32 s5, s2, 8
+; GFX9-DL-NEXT:    s_lshr_b32 s6, s4, 4
+; GFX9-DL-NEXT:    s_lshr_b32 s7, s4, 12
+; GFX9-DL-NEXT:    s_lshr_b32 s8, s4, 8
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v3, 12, s5
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v4, 12, s1
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v5, 12, s0
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v7, 12, s8
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v8, 12, s7
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v9, 12, s6
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v6, 12, s2
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v10, 12, s4
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v3, 12, v3
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v5, 12, v5
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v6, v6, v10
+; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v5, v5, v9 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v4, v4, v8 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v3, v3, v7
+; GFX9-DL-NEXT:    s_lshr_b32 s0, s2, 20
+; GFX9-DL-NEXT:    s_lshr_b32 s1, s2, 16
+; GFX9-DL-NEXT:    s_lshr_b32 s5, s2, 28
+; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 24
+; GFX9-DL-NEXT:    s_lshr_b32 s6, s4, 20
+; GFX9-DL-NEXT:    s_lshr_b32 s7, s4, 16
+; GFX9-DL-NEXT:    s_lshr_b32 s8, s4, 28
+; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 24
+; GFX9-DL-NEXT:    v_or_b32_sdwa v5, v6, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-DL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-DL-NEXT:    v_or_b32_sdwa v3, v5, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v9, 12, s2
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v10, 12, s5
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v11, 12, s1
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v12, 12, s0
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v13, 12, s4
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v14, 12, s8
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v15, 12, s7
+; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v16, 12, s6
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v13, 12, v13
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v14, 12, v14
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v15, 12, v15
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v12, 12, v12
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v16, 12, v16
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v5, 8, v3
+; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v12, v12, v16 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v11, v11, v15
+; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v10, v10, v14 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v9, v9, v13
+; GFX9-DL-NEXT:    v_or_b32_sdwa v7, v11, v12 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-DL-NEXT:    v_or_b32_sdwa v8, v9, v10 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-DL-NEXT:    v_or_b32_sdwa v4, v7, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v3, v2
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v5
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_2
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v4
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v3, 8, v4
+; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v3
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
+; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
+; GFX9-DL-NEXT:    s_endpgm
+                                             <8 x i4> addrspace(1)* %src2,
+                                             i8 addrspace(1)* nocapture %dst) {
+entry:
+  %vec1 = load <8 x i4>, <8 x i4> addrspace(1)* %src1
+  %vec2 = load <8 x i4>, <8 x i4> addrspace(1)* %src2
+
+  %cvec1 = sext <8 x i4> %vec1 to <8 x i8>
+  %cvec2 = sext <8 x i4> %vec2 to <8 x i8>
+
+  %mul = mul <8 x i8> %cvec1, %cvec2
+  %mul0 = extractelement <8 x i8> %mul, i64 0
+  %mul1 = extractelement <8 x i8> %mul, i64 1
+  %mul2 = extractelement <8 x i8> %mul, i64 2
+  %mul3 = extractelement <8 x i8> %mul, i64 3
+  %mul4 = extractelement <8 x i8> %mul, i64 4
+  %mul5 = extractelement <8 x i8> %mul, i64 5
+  %mul6 = extractelement <8 x i8> %mul, i64 6
+  %mul7 = extractelement <8 x i8> %mul, i64 7
+
+  %acc = load i8, i8 addrspace(1)* %dst, align 4
+  %add1 = add i8 %mul0, %acc
+  %add2 = add i8 %add1, %mul1
+  %add3 = add i8 %add2, %mul2
+  %add4 = add i8 %add3, %mul3
+  %add5 = add i8 %add4, %mul4
+  %add6 = add i8 %add5, %mul5
+  %add7 = add i8 %add6, %mul6
+  %add8 = add i8 %add7, %mul7
+
+  store i8 %add8, i8 addrspace(1)* %dst, align 4
+  ret void
+}