AMDGPU/GlobalISel: Select llvm.amdgcn.fdot2

[arsenm]

Summary

I'm slighly worried about the generated checks, since they won't catch
incorrect modifiers being added at the end of the line.

[nhaehnle]

LGTM

I'm slighly worried about the generated checks, since they won't catch incorrect modifiers being added at the end of the line.

Maybe the script could be changed to add {{$}} everywhere? Though that would end up looking really ugly :/

[arsenm]

4c1c9422a3adab64433b8fde31f1ac346459b491

llvm/lib/Target/AMDGPU/AMDGPUInstrInfo.td

 >;
 
 def AMDGPUumed3 : SDNode<"AMDGPUISD::UMED3", AMDGPUDTIntTernaryOp,
   []
 >;
 
 def AMDGPUfmed3_impl : SDNode<"AMDGPUISD::FMED3", SDTFPTernaryOp, []>;
 
-def AMDGPUfdot2 : SDNode<"AMDGPUISD::FDOT2",
+def AMDGPUfdot2_impl : SDNode<"AMDGPUISD::FDOT2",
                   SDTypeProfile<1, 4, [SDTCisSameAs<0, 3>, SDTCisSameAs<1, 2>,
                                        SDTCisFP<0>, SDTCisVec<1>,
                                        SDTCisInt<4>]>,
                   []>;
 
 def AMDGPUperm : SDNode<"AMDGPUISD::PERM", AMDGPUDTIntTernaryOp, []>;
 
 // SI+ export

 
 def AMDGPUbfe_i32 : PatFrags<(ops node:$src0, node:$src1, node:$src2),
   [(int_amdgcn_sbfe node:$src0, node:$src1, node:$src2),
    (AMDGPUbfe_i32_impl node:$src0, node:$src1, node:$src2)]>;
 
 def AMDGPUbfe_u32 : PatFrags<(ops node:$src0, node:$src1, node:$src2),
   [(int_amdgcn_ubfe node:$src0, node:$src1, node:$src2),
    (AMDGPUbfe_u32_impl node:$src0, node:$src1, node:$src2)]>;
+
+def AMDGPUfdot2 : PatFrags<(ops node:$src0, node:$src1, node:$src2, node:$src3),
+  [(int_amdgcn_fdot2 node:$src0, node:$src1, node:$src2, node:$src3),
+   (AMDGPUfdot2_impl node:$src0, node:$src1, node:$src2, node:$src3)]>;

llvm/lib/Target/AMDGPU/VOP3PInstructions.td

 } // End let IsDOT = 1
 
 multiclass DotPats<SDPatternOperator dot_op,
                    VOP3PInst dot_inst> {
   let SubtargetPredicate = dot_inst.SubtargetPredicate in
   def : GCNPat <
     (dot_op (dot_inst.Pfl.Src0VT (VOP3PMods0 dot_inst.Pfl.Src0VT:$src0, i32:$src0_modifiers)),
             (dot_inst.Pfl.Src1VT (VOP3PMods dot_inst.Pfl.Src1VT:$src1, i32:$src1_modifiers)),
-            (dot_inst.Pfl.Src2VT (VOP3PMods dot_inst.Pfl.Src2VT:$src2, i32:$src2_modifiers)), i1:$clamp),
-    (dot_inst $src0_modifiers, $src0, $src1_modifiers, $src1, $src2_modifiers, $src2, (as_i1timm $clamp))>;
+            (dot_inst.Pfl.Src2VT (VOP3PMods dot_inst.Pfl.Src2VT:$src2, i32:$src2_modifiers)), timm:$clamp),
+    (dot_inst $src0_modifiers, VSrc_v2f16:$src0,
+              $src1_modifiers, VSrc_v2f16:$src1,
+              $src2_modifiers, VSrc_f32:$src2, timm:$clamp)>;
 }
 
 defm : DotPats<AMDGPUfdot2, V_DOT2_F32_F16>;
 defm : DotPats<int_amdgcn_sdot2, V_DOT2_I32_I16>;
 defm : DotPats<int_amdgcn_udot2, V_DOT2_U32_U16>;
 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>;

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.fdot2.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx906 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX906 %s
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx1011 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10 %s
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx1012 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10 %s
+
+define float @v_fdot2(<2 x half> %a, <2 x half> %b, float %c) {
+; GFX906-LABEL: v_fdot2:
+; GFX906:       ; %bb.0:
+; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT:    v_dot2_f32_f16 v0, v0, v1, v2
+; GFX906-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_fdot2:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_dot2_f32_f16 v0, v0, v1, v2
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+  %r = call float @llvm.amdgcn.fdot2(<2 x half> %a, <2 x half> %b, float %c, i1 false)
+  ret float %r
+}
+
+define float @v_fdot2_clamp(<2 x half> %a, <2 x half> %b, float %c) {
+; GFX906-LABEL: v_fdot2_clamp:
+; GFX906:       ; %bb.0:
+; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT:    v_dot2_f32_f16 v0, v0, v1, v2 clamp
+; GFX906-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_fdot2_clamp:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_dot2_f32_f16 v0, v0, v1, v2 clamp
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+  %r = call float @llvm.amdgcn.fdot2(<2 x half> %a, <2 x half> %b, float %c, i1 true)
+  ret float %r
+}
+
+define float @v_fdot2_neg_a(<2 x half> %a, <2 x half> %b, float %c) {
+; GFX906-LABEL: v_fdot2_neg_a:
+; GFX906:       ; %bb.0:
+; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT:    v_dot2_f32_f16 v0, v0, v1, v2 neg_lo:[1,0,0] neg_hi:[1,0,0]
+; GFX906-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_fdot2_neg_a:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_dot2_f32_f16 v0, v0, v1, v2 neg_lo:[1,0,0] neg_hi:[1,0,0]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+  %neg.a = fneg <2 x half> %a
+  %r = call float @llvm.amdgcn.fdot2(<2 x half> %neg.a, <2 x half> %b, float %c, i1 false)
+  ret float %r
+}
+
+define float @v_fdot2_neg_b(<2 x half> %a, <2 x half> %b, float %c) {
+; GFX906-LABEL: v_fdot2_neg_b:
+; GFX906:       ; %bb.0:
+; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT:    v_dot2_f32_f16 v0, v0, v1, v2 neg_lo:[0,1,0] neg_hi:[0,1,0]
+; GFX906-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_fdot2_neg_b:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_dot2_f32_f16 v0, v0, v1, v2 neg_lo:[0,1,0] neg_hi:[0,1,0]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+  %neg.b = fneg <2 x half> %b
+  %r = call float @llvm.amdgcn.fdot2(<2 x half> %a, <2 x half> %neg.b, float %c, i1 false)
+  ret float %r
+}
+
+define float @v_fdot2_neg_a_neg_b(<2 x half> %a, <2 x half> %b, float %c) {
+; GFX906-LABEL: v_fdot2_neg_a_neg_b:
+; GFX906:       ; %bb.0:
+; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT:    v_dot2_f32_f16 v0, v1, v1, v2 neg_lo:[1,1,0] neg_hi:[1,1,0]
+; GFX906-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_fdot2_neg_a_neg_b:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_dot2_f32_f16 v0, v1, v1, v2 neg_lo:[1,1,0] neg_hi:[1,1,0]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+  %neg.a = fneg <2 x half> %b
+  %neg.b = fneg <2 x half> %b
+  %r = call float @llvm.amdgcn.fdot2(<2 x half> %neg.a, <2 x half> %neg.b, float %c, i1 false)
+  ret float %r
+}
+
+define float @v_fdot2_neg_c(<2 x half> %a, <2 x half> %b, float %c) {
+; GFX906-LABEL: v_fdot2_neg_c:
+; GFX906:       ; %bb.0:
+; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT:    v_dot2_f32_f16 v0, v0, v1, v2 neg_lo:[0,0,1] neg_hi:[0,0,1]
+; GFX906-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_fdot2_neg_c:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_dot2_f32_f16 v0, v0, v1, v2 neg_lo:[0,0,1] neg_hi:[0,0,1]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+  %neg.c = fneg float %c
+  %r = call float @llvm.amdgcn.fdot2(<2 x half> %a, <2 x half> %b, float %neg.c, i1 false)
+  ret float %r
+}
+
+define float @v_fdot2_inline_literal_a(<2 x half> %b, float %c) {
+; GFX906-LABEL: v_fdot2_inline_literal_a:
+; GFX906:       ; %bb.0:
+; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT:    s_movk_i32 s4, 0x4000
+; GFX906-NEXT:    s_pack_ll_b32_b16 s4, s4, s4
+; GFX906-NEXT:    v_dot2_f32_f16 v0, s4, v0, v1
+; GFX906-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_fdot2_inline_literal_a:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_movk_i32 s4, 0x4000
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_pack_ll_b32_b16 s4, s4, s4
+; GFX10-NEXT:    v_dot2_f32_f16 v0, s4, v0, v1
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+  %ret = tail call float @llvm.amdgcn.fdot2(<2 x half> <half 2.0, half 2.0>, <2 x half> %b, float %c, i1 false)
+  ret float %ret
+}
+
+define float @v_fdot2_inline_literal_b(<2 x half> %a, float %c) {
+; GFX906-LABEL: v_fdot2_inline_literal_b:
+; GFX906:       ; %bb.0:
+; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT:    s_movk_i32 s4, 0x4000
+; GFX906-NEXT:    s_pack_ll_b32_b16 s4, s4, s4
+; GFX906-NEXT:    v_dot2_f32_f16 v0, v0, s4, v1
+; GFX906-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_fdot2_inline_literal_b:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_movk_i32 s4, 0x4000
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_pack_ll_b32_b16 s4, s4, s4
+; GFX10-NEXT:    v_dot2_f32_f16 v0, v0, s4, v1
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+  %ret = tail call float @llvm.amdgcn.fdot2(<2 x half> %a, <2 x half> <half 2.0, half 2.0>, float %c, i1 false)
+  ret float %ret
+}
+
+define float @v_fdot2_inline_literal_c(<2 x half> %a, <2 x half> %b) {
+; GFX906-LABEL: v_fdot2_inline_literal_c:
+; GFX906:       ; %bb.0:
+; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT:    v_dot2_f32_f16 v0, v0, v1, 1.0
+; GFX906-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_fdot2_inline_literal_c:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_dot2_f32_f16 v0, v0, v1, 1.0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+  %ret = tail call float @llvm.amdgcn.fdot2(<2 x half> %a, <2 x half> %b, float 1.0, i1 false)
+  ret float %ret
+}
+
+declare float @llvm.amdgcn.fdot2(<2 x half>, <2 x half>, float, i1 immarg) #0
+
+attributes #0 = { nounwind readnone speculatable }