[AMDGPU]: Fix type signatures for wmma intrinsics, NFC (llvm#80087)

Make the wmma intrinsic type signatures to be canonical. We need
a type signature as long as the type is not fixed. However, when an
argument's type matches a previous argument's type, we do not need the
signature for this argument.

 This patch fixes three general cases:
  1. add missing signatures
  2. remove signatures for matching arguments
3. reorer the signatures -- return type signature should always appear
first

Author: changpeng

llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll

@@ -112,85 +112,85 @@ bb:
 ; CHECK: DIVERGENT: %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.f16.v8f32.v8f16.v16f16.i16(<8 x half> %A, <16 x half> %B, <8 x float> %C, i16 %Index)
 define amdgpu_kernel void @swmmac_f32_16x16x32_f16(<8 x half> %A, <16 x half> %B, <8 x float> %C, i16 %Index, ptr addrspace(1) %out) {
 bb:
-  %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.f16(<8 x half> %A, <16 x half> %B, <8 x float> %C, i16 %Index)
+  %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.f16.v8f32.v8f16.v16f16.i16(<8 x half> %A, <16 x half> %B, <8 x float> %C, i16 %Index)
     store <8 x float> %tmp0, ptr addrspace(1) %out, align 32
     ret void
 }
 
 ; CHECK: DIVERGENT: %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.bf16.v8f32.v8i16.v16i16.i16(<8 x i16> %A, <16 x i16> %B, <8 x float> %C, i16 %Index)
 define amdgpu_kernel void @swmmac_f32_16x16x32_bf16(<8 x i16> %A, <16 x i16> %B, <8 x float> %C, i16 %Index, ptr addrspace(1) %out) {
 bb:
-  %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.bf16(<8 x i16> %A, <16 x i16> %B, <8 x float> %C, i16 %Index)
+  %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.bf16.v8f32.v8i16.v16i16.i16(<8 x i16> %A, <16 x i16> %B, <8 x float> %C, i16 %Index)
     store <8 x float> %tmp0, ptr addrspace(1) %out, align 32
     ret void
 }
 
 ; CHECK: DIVERGENT: %tmp0 = call <8 x half> @llvm.amdgcn.swmmac.f16.16x16x32.f16.v8f16.v8f16.v16f16.i16(<8 x half> %A, <16 x half> %B, <8 x half> %C, i16 %Index)
 define amdgpu_kernel void @swmmac_f16_16x16x32_f16(<8 x half> %A, <16 x half> %B, <8 x half> %C, i16 %Index, ptr addrspace(1) %out) {
 bb:
-  %tmp0 = call <8 x half> @llvm.amdgcn.swmmac.f16.16x16x32.f16(<8 x half> %A, <16 x half> %B, <8 x half> %C, i16 %Index)
+  %tmp0 = call <8 x half> @llvm.amdgcn.swmmac.f16.16x16x32.f16.v8f16.v8f16.v16f16.i16(<8 x half> %A, <16 x half> %B, <8 x half> %C, i16 %Index)
     store <8 x half> %tmp0, ptr addrspace(1) %out, align 32
     ret void
 }
 
 ; CHECK: DIVERGENT: %tmp0 = call <8 x i16> @llvm.amdgcn.swmmac.bf16.16x16x32.bf16.v8i16.v8i16.v16i16.i16(<8 x i16> %A, <16 x i16> %B, <8 x i16> %C, i16 %Index)
 define amdgpu_kernel void @swmmac_bf16_16x16x32_bf16(<8 x i16> %A, <16 x i16> %B, <8 x i16> %C, i16 %Index, ptr addrspace(1) %out) {
 bb:
-  %tmp0 = call <8 x i16> @llvm.amdgcn.swmmac.bf16.16x16x32.bf16(<8 x i16> %A, <16 x i16> %B, <8 x i16> %C, i16 %Index)
+  %tmp0 = call <8 x i16> @llvm.amdgcn.swmmac.bf16.16x16x32.bf16.v8i16.v8i16.v16i16.i16(<8 x i16> %A, <16 x i16> %B, <8 x i16> %C, i16 %Index)
   store <8 x i16> %tmp0, ptr addrspace(1) %out, align 32
   ret void
 }
 
 ; CHECK: DIVERGENT: %tmp0 = call <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x32.iu8.v8i32.v2i32.v4i32.i16(i1 false, <2 x i32> %A, i1 false, <4 x i32> %B, <8 x i32> %C, i16 %Index, i1 false)
 define amdgpu_kernel void @swmmac_i32_16x16x32_iu8(<2 x i32> %A, <4 x i32> %B, <8 x i32> %C, i16 %Index, ptr addrspace(1) %out) {
 bb:
-  %tmp0 = call <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x32.iu8(i1 false, <2 x i32> %A, i1 false, <4 x i32> %B, <8 x i32> %C, i16 %Index, i1 false)
+  %tmp0 = call <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x32.iu8.v8i32.v2i32.v4i32.i16(i1 false, <2 x i32> %A, i1 false, <4 x i32> %B, <8 x i32> %C, i16 %Index, i1 false)
   store <8 x i32> %tmp0, ptr addrspace(1) %out, align 32
   ret void
 }
 
 ; CHECK: DIVERGENT: %tmp0 = call <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x32.iu4.v8i32.i32.v2i32.i16(i1 false, i32 %A, i1 false, <2 x i32> %B, <8 x i32> %C, i16 %Index, i1 false)
 define amdgpu_kernel void @swmmac_i32_16x16x32_iu4(i32 %A, <2 x i32> %B, <8 x i32> %C, i16 %Index, ptr addrspace(1) %out) {
 bb:
-  %tmp0 = call <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x32.iu4(i1 false, i32 %A, i1 false, <2 x i32> %B, <8 x i32> %C, i16 %Index, i1 false)
+  %tmp0 = call <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x32.iu4.v8i32.i32.v2i32.i16(i1 false, i32 %A, i1 false, <2 x i32> %B, <8 x i32> %C, i16 %Index, i1 false)
   store <8 x i32> %tmp0, ptr addrspace(1) %out, align 32
   ret void
 }
 
 ; CHECK: DIVERGENT: %tmp0 = call <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x64.iu4.v8i32.v2i32.v4i32.i16(i1 false, <2 x i32> %A, i1 false, <4 x i32> %B, <8 x i32> %C, i16 %Index, i1 false)
 define amdgpu_kernel void @swmmac_i32_16x16x64_iu4(<2 x i32> %A, <4 x i32> %B, <8 x i32> %C, i16 %Index, ptr addrspace(1) %out) {
 bb:
-  %tmp0 = call <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x64.iu4(i1 false, <2 x i32> %A, i1 false, <4 x i32> %B, <8 x i32> %C, i16 %Index, i1 false)
+  %tmp0 = call <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x64.iu4.v8i32.v2i32.v4i32.i16(i1 false, <2 x i32> %A, i1 false, <4 x i32> %B, <8 x i32> %C, i16 %Index, i1 false)
   store <8 x i32> %tmp0, ptr addrspace(1) %out, align 32
   ret void
 }
 
 ; CHECK: DIVERGENT: %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.fp8.fp8.v8f32.v2i32.v4i32.i16(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index)
 define amdgpu_kernel void @swmmac_f32_16x16x32_fp8.fp8(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index, ptr addrspace(1) %out) {
 bb:
-  %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.fp8.fp8(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index)
+  %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.fp8.fp8.v8f32.v2i32.v4i32.i16(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index)
   store <8 x float> %tmp0, ptr addrspace(1) %out, align 32
   ret void
 }
 
 ; CHECK: DIVERGENT: %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.fp8.bf8.v8f32.v2i32.v4i32.i16(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index)
 define amdgpu_kernel void @swmmac_f32_16x16x32_fp8.bf8(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index, ptr addrspace(1) %out) {
 bb:
-  %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.fp8.bf8(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index)
+  %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.fp8.bf8.v8f32.v2i32.v4i32.i16(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index)
   store <8 x float> %tmp0, ptr addrspace(1) %out, align 32
   ret void
 }
 
 ; CHECK: DIVERGENT: %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.bf8.fp8.v8f32.v2i32.v4i32.i16(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index)
-define amdgpu_kernel void @swmmac_f32_16x16x32_bf8.fp8(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index, ptr addrspace(1) %out) {
+define amdgpu_kernel void @swmmac_f32_16x16x32_bf8.fp8.v8f32.v2i32.v4i32.i16(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index, ptr addrspace(1) %out) {
 bb:
   %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.bf8.fp8(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index)
   store <8 x float> %tmp0, ptr addrspace(1) %out, align 32
   ret void
 }
 
 ; CHECK: DIVERGENT: %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.bf8.bf8.v8f32.v2i32.v4i32.i16(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index)
-define amdgpu_kernel void @swmmac_f32_16x16x32_bf8.bf8(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index, ptr addrspace(1) %out) {
+define amdgpu_kernel void @swmmac_f32_16x16x32_bf8.bf8.v8f32.v2i32.v4i32.i16(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index, ptr addrspace(1) %out) {
 bb:
   %tmp0 = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.bf8.bf8(<2 x i32> %A, <4 x i32> %B, <8 x float> %C, i16 %Index)
   store <8 x float> %tmp0, ptr addrspace(1) %out, align 32
@@ -284,13 +284,13 @@ declare <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16.v16f16.v16f16(<16 x half>
 declare <16 x i16> @llvm.amdgcn.wmma.bf16.16x16x16.bf16.v16i16.v16i16(<16 x i16>, <16 x i16> , <16 x i16>, i1 immarg) #1
 declare <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8.v8i32.v4i32(i1 immarg, <4 x i32>, i1 immarg, <4 x i32> , <8 x i32>, i1 immarg) #1
 declare <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4.v8i32.v2i32(i1 immarg, <2 x i32>, i1 immarg, <2 x i32> , <8 x i32>, i1 immarg) #1
-declare <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.f16(<8 x half>, <16 x half>, <8 x float>, i16)
-declare <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.bf16(<8 x i16>, <16 x i16>, <8 x float>, i16)
-declare <8 x half> @llvm.amdgcn.swmmac.f16.16x16x32.f16(<8 x half>, <16 x half>, <8 x half>, i16)
-declare <8 x i16> @llvm.amdgcn.swmmac.bf16.16x16x32.bf16(<8 x i16>, <16 x i16>, <8 x i16>, i16)
-declare <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x32.iu8(i1 immarg, <2 x i32>, i1 immarg, <4 x i32>, <8 x i32>, i16, i1)
-declare <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x32.iu4(i1 immarg, i32, i1 immarg, <2 x i32>, <8 x i32>, i16, i1)
-declare <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x64.iu4(i1 immarg, <2 x i32>, i1 immarg, <4 x i32>, <8 x i32>, i16, i1)
+declare <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.f16.v8f32.v8f16.v16f16.i16(<8 x half>, <16 x half>, <8 x float>, i16)
+declare <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.bf16.v8f32.v8i16.v16i16.i16(<8 x i16>, <16 x i16>, <8 x float>, i16)
+declare <8 x half> @llvm.amdgcn.swmmac.f16.16x16x32.f16.v8f16.v8f16.v16f16.i16(<8 x half>, <16 x half>, <8 x half>, i16)
+declare <8 x i16> @llvm.amdgcn.swmmac.bf16.16x16x32.bf16.v8i16.v8i16.v16i16.i16(<8 x i16>, <16 x i16>, <8 x i16>, i16)
+declare <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x32.iu8.v8i32.v2i32.v4i32.i16(i1 immarg, <2 x i32>, i1 immarg, <4 x i32>, <8 x i32>, i16, i1)
+declare <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x32.iu4.v8i32.i32.v2i32.i16(i1 immarg, i32, i1 immarg, <2 x i32>, <8 x i32>, i16, i1)
+declare <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x64.iu4.v8i32.v2i32.v4i32.i16(i1 immarg, <2 x i32>, i1 immarg, <4 x i32>, <8 x i32>, i16, i1)
 declare <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.fp8.fp8(<2 x i32>, <4 x i32>, <8 x float>, i16)
 declare <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.fp8.bf8(<2 x i32>, <4 x i32>, <8 x float>, i16)
 declare <8 x float> @llvm.amdgcn.swmmac.f32.16x16x32.bf8.fp8(<2 x i32>, <4 x i32>, <8 x float>, i16)