R600: Add intrinsics for mad24

[arsenm]

Forgot to attach tests

[arsenm]

r209456. r209457 expands it for older GPUs, and r209458 also expands mul24

lib/Target/R600/AMDGPUISelLowering.h

   BFM, // Insert a range of bits into a 32-bit word.
   MUL_U24,
   MUL_I24,
+  MAD_U24,
+  MAD_I24,
   TEXTURE_FETCH,
   EXPORT,
   CONST_ADDRESS,

lib/Target/R600/AMDGPUISelLowering.cpp

       return DAG.getNode(AMDGPUISD::MUL_I24, DL, VT,
                          Op.getOperand(1), Op.getOperand(2));
 
+    case AMDGPUIntrinsic::AMDGPU_umad24:
+      return DAG.getNode(AMDGPUISD::MAD_U24, DL, VT,
+                         Op.getOperand(1), Op.getOperand(2), Op.getOperand(3));
+
+    case AMDGPUIntrinsic::AMDGPU_imad24:
+      return DAG.getNode(AMDGPUISD::MAD_I24, DL, VT,
+                         Op.getOperand(1), Op.getOperand(2), Op.getOperand(3));
+
     case AMDGPUIntrinsic::AMDGPU_bfe_i32:
       return DAG.getNode(AMDGPUISD::BFE_I32, DL, VT,
                          Op.getOperand(1),

   NODE_NAME_CASE(BFM)
   NODE_NAME_CASE(MUL_U24)
   NODE_NAME_CASE(MUL_I24)
+  NODE_NAME_CASE(MAD_U24)
+  NODE_NAME_CASE(MAD_I24)
   NODE_NAME_CASE(URECIP)
   NODE_NAME_CASE(DOT4)
   NODE_NAME_CASE(EXPORT)

lib/Target/R600/AMDGPUInstrInfo.td

 def AMDGPUmul_i24 : SDNode<"AMDGPUISD::MUL_I24", SDTIntBinOp,
   [SDNPCommutative]
 >;
+
+def AMDGPUmad_u24 : SDNode<"AMDGPUISD::MAD_U24", AMDGPUDTIntTernaryOp,
+  []
+>;
+def AMDGPUmad_i24 : SDNode<"AMDGPUISD::MAD_I24", AMDGPUDTIntTernaryOp,
+  []
+>;

lib/Target/R600/AMDGPUInstructions.td

 >;
 */
 
+class IMad24Pat<Instruction Inst> : Pat <
+  (add (AMDGPUmul_i24 i32:$src0, i32:$src1), i32:$src2),
+  (Inst $src0, $src1, $src2)
+>;
+
+class UMad24Pat<Instruction Inst> : Pat <
+  (add (AMDGPUmul_u24 i32:$src0, i32:$src1), i32:$src2),
+  (Inst $src0, $src1, $src2)
+>;
+
+
 include "R600Instructions.td"
 include "R700Instructions.td"
 include "EvergreenInstructions.td"

lib/Target/R600/AMDGPUIntrinsics.td

   def int_AMDGPU_umin : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_umul24 : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_imul24 : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_imad24 : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_umad24 : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_cube : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty], [IntrNoMem]>;
   def int_AMDGPU_bfi : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_bfe_i32 : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;

lib/Target/R600/CaymanInstructions.td

 let Predicates = [isCayman] in {
 
 def MULADD_INT24_cm : R600_3OP <0x08, "MULADD_INT24",
-  [(set i32:$dst, (add (AMDGPUmul_i24 i32:$src0, i32:$src1), i32:$src2))], VecALU
+  [(set i32:$dst, (AMDGPUmad_i24 i32:$src0, i32:$src1, i32:$src2))], VecALU
 >;
 def MUL_INT24_cm : R600_2OP <0x5B, "MUL_INT24",
   [(set i32:$dst, (AMDGPUmul_i24 i32:$src0, i32:$src1))], VecALU
 >;
 
+def : IMad24Pat<MULADD_INT24_cm>;
+
 let isVector = 1 in {
 
 def RECIP_IEEE_cm : RECIP_IEEE_Common<0x86>;

lib/Target/R600/SIInstructions.td

 def V_MAD_LEGACY_F32 : VOP3_32 <0x00000140, "V_MAD_LEGACY_F32", []>;
 def V_MAD_F32 : VOP3_32 <0x00000141, "V_MAD_F32", []>;
 def V_MAD_I32_I24 : VOP3_32 <0x00000142, "V_MAD_I32_I24",
-  [(set i32:$dst, (add (AMDGPUmul_i24 i32:$src0, i32:$src1), i32:$src2))]
+    [(set i32:$dst, (AMDGPUmad_i24 i32:$src0, i32:$src1, i32:$src2))]
 >;
 def V_MAD_U32_U24 : VOP3_32 <0x00000143, "V_MAD_U32_U24",
-  [(set i32:$dst, (add (AMDGPUmul_u24 i32:$src0, i32:$src1), i32:$src2))]
+  [(set i32:$dst, (AMDGPUmad_u24 i32:$src0, i32:$src1, i32:$src2))]
 >;
 
 } // End neverHasSideEffects
+
+def : IMad24Pat<V_MAD_I32_I24>;
+def : UMad24Pat<V_MAD_U32_U24>;
+
 def V_CUBEID_F32 : VOP3_32 <0x00000144, "V_CUBEID_F32", []>;
 def V_CUBESC_F32 : VOP3_32 <0x00000145, "V_CUBESC_F32", []>;
 def V_CUBETC_F32 : VOP3_32 <0x00000146, "V_CUBETC_F32", []>;

test/CodeGen/R600/llvm.AMDGPU.imad24.ll

+; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
+; RUN: llc -march=r600 -mcpu=cayman -verify-machineinstrs < %s | FileCheck -check-prefix=CM -check-prefix=FUNC %s
+
+declare i32 @llvm.AMDGPU.imad24(i32, i32, i32) nounwind readnone
+
+; FUNC-LABEL: @test_imad24
+; SI: V_MAD_I32_I24
+; CM: MULADD_INT24
+define void @test_imad24(i32 addrspace(1)* %out, i32 %src0, i32 %src1, i32 %src2) nounwind {
+  %mad = call i32 @llvm.AMDGPU.imad24(i32 %src0, i32 %src1, i32 %src2) nounwind readnone
+  store i32 %mad, i32 addrspace(1)* %out, align 4
+  ret void
+}
+

test/CodeGen/R600/llvm.AMDGPU.umad24.ll

+; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
+
+declare i32 @llvm.AMDGPU.umad24(i32, i32, i32) nounwind readnone
+
+; FUNC-LABEL: @test_umad24
+; SI: V_MAD_U32_U24
+define void @test_umad24(i32 addrspace(1)* %out, i32 %src0, i32 %src1, i32 %src2) nounwind {
+  %mad = call i32 @llvm.AMDGPU.umad24(i32 %src0, i32 %src1, i32 %src2) nounwind readnone
+  store i32 %mad, i32 addrspace(1)* %out, align 4
+  ret void
+}
+