[llvm-branch-commits] [clang] [llvm] AMDGPU: Add v_mfma_i32_16x16x64_i8 for gfx950 (PR #116728)

via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Mon Nov 18 17:46:16 PST 2024


llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-amdgpu

Author: Matt Arsenault (arsenm)

<details>
<summary>Changes</summary>



---

Patch is 33.44 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/116728.diff


13 Files Affected:

- (modified) clang/include/clang/Basic/BuiltinsAMDGPU.def (+1-1) 
- (modified) clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl (+6) 
- (modified) clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl (+7) 
- (modified) clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950.cl (+4-1) 
- (modified) llvm/include/llvm/IR/IntrinsicsAMDGPU.td (+1) 
- (modified) llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp (+2-1) 
- (modified) llvm/lib/Target/AMDGPU/SIInstrInfo.td (+1) 
- (modified) llvm/lib/Target/AMDGPU/VOP3PInstructions.td (+7-1) 
- (modified) llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll (+8) 
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll (+285) 
- (modified) llvm/test/MC/AMDGPU/mai-gfx950.s (+55) 
- (modified) llvm/test/MC/Disassembler/AMDGPU/gfx950_mai.txt (+34) 
- (modified) llvm/test/tools/llvm-mca/AMDGPU/gfx950.s (+9-3) 


``````````diff
diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 7ce8f2c1669d67..be5e7928b48fcb 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -436,7 +436,7 @@ TARGET_BUILTIN(__builtin_amdgcn_cvt_sr_fp8_f32, "ifiiIi", "nc", "fp8-conversion-
 //===----------------------------------------------------------------------===//
 TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x32_f16, "V4fV8hV8hV4fIiIiIi", "nc", "gfx950-insts")
 TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_f16, "V16fV8hV8hV16fIiIiIi", "nc", "gfx950-insts")
-
+TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_16x16x64_i8, "V4iV4iV4iV4iIiIiIi", "nc", "gfx950-insts")
 TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_bf16, "V16fV8yV8yV16fIiIiIi", "nc", "gfx950-insts")
 
 //===----------------------------------------------------------------------===//
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl b/clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl
index 841d8fcad0fee0..77d0e464ae5d2e 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl
@@ -431,4 +431,10 @@ v16f test_mfma_f32_32x32x16_bf16(v8bf16 a, v8bf16 b, v16f c) {
   return __builtin_amdgcn_mfma_f32_32x32x16_bf16(a, b, c, 1, 2, 3);
 }
 
+// CHECK-GFX950-LABEL: @test_mfma_i32_16x16x64_i8(
+// CHECK-GFX950: tail call <4 x i32> @llvm.amdgcn.mfma.i32.16x16x64.i8(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c, i32 1, i32 2, i32 3)
+v4i test_mfma_i32_16x16x64_i8(v4i a, v4i b, v4i c) {
+  return __builtin_amdgcn_mfma_i32_16x16x64_i8(a, b, c, 1, 2, 3);
+}
+
 #endif
diff --git a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl
index 4af67763c40dd2..bb805b56e10e3a 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl
@@ -5,6 +5,7 @@ typedef float float4 __attribute__((ext_vector_type(4)));
 typedef float float16 __attribute__((ext_vector_type(16)));
 typedef half half8 __attribute__((ext_vector_type(8)));
 typedef __bf16 bfloat8 __attribute__((ext_vector_type(8)));
+typedef int int4 __attribute__((ext_vector_type(4)));
 
 
 void test_mfma_f32_16x16x32_f16(__global float4* out, half8 a, half8 b, float4 c, int X) {
@@ -26,3 +27,9 @@ void test_mfma_f32_32x32x16_bf16(__global float16* out, bfloat8 a, bfloat8 b, fl
   *out = __builtin_amdgcn_mfma_f32_32x32x16_bf16(a, b, c, 0, X, 0);  // expected-error{{argument to '__builtin_amdgcn_mfma_f32_32x32x16_bf16' must be a constant integer}}
   *out = __builtin_amdgcn_mfma_f32_32x32x16_bf16(a, b, c, 0, 0, X);  // expected-error{{argument to '__builtin_amdgcn_mfma_f32_32x32x16_bf16' must be a constant integer}}
 }
+
+void test_mfma_i32_16x16x64_i8(__global int4* out, int4 a, int4 b, int4 c, int X) {
+  *out = __builtin_amdgcn_mfma_i32_16x16x64_i8(a, b, c, X, 0, 0); // expected-error{{argument to '__builtin_amdgcn_mfma_i32_16x16x64_i8' must be a constant integer}}
+  *out = __builtin_amdgcn_mfma_i32_16x16x64_i8(a, b, c, 0, X, 0);  // expected-error{{argument to '__builtin_amdgcn_mfma_i32_16x16x64_i8' must be a constant integer}}
+  *out = __builtin_amdgcn_mfma_i32_16x16x64_i8(a, b, c, 0, 0, X);  // expected-error{{argument to '__builtin_amdgcn_mfma_i32_16x16x64_i8' must be a constant integer}}
+}
diff --git a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950.cl b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950.cl
index e0fd2aa5c58a02..3553fc3e4a49c6 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950.cl
@@ -5,11 +5,14 @@ typedef float float4 __attribute__((ext_vector_type(4)));
 typedef float float16 __attribute__((ext_vector_type(16)));
 typedef half half8 __attribute__((ext_vector_type(8)));
 typedef __bf16 bfloat8 __attribute__((ext_vector_type(8)));
+typedef int int4 __attribute__((ext_vector_type(4)));
 
 void test(__global float4* out0, half8 a0, half8 b0, float4 c0,
           __global float16* out1, half8 a1, half8 b1, float16 c1,
-          __global float16* out2, bfloat8 a2, bfloat8 b2, float16 c2) {
+          __global float16* out2, bfloat8 a2, bfloat8 b2, float16 c2,
+          __global int4* out3, int4 a3, int4 b3, int4 c3) {
   *out0 = __builtin_amdgcn_mfma_f32_16x16x32_f16(a0, b0, c0, 0, 0, 0); // expected-error{{'__builtin_amdgcn_mfma_f32_16x16x32_f16' needs target feature gfx950-insts}}
   *out1 = __builtin_amdgcn_mfma_f32_32x32x16_f16(a1, b1, c1, 0, 0, 0); // expected-error{{'__builtin_amdgcn_mfma_f32_32x32x16_f16' needs target feature gfx950-insts}}
   *out2 = __builtin_amdgcn_mfma_f32_32x32x16_bf16(a2, b2, c2, 0, 0, 0); // expected-error{{'__builtin_amdgcn_mfma_f32_32x32x16_bf16' needs target feature gfx950-insts}}
+  *out3 = __builtin_amdgcn_mfma_i32_16x16x64_i8(a3, b3, c3, 0, 0, 0); // expected-error{{'__builtin_amdgcn_mfma_i32_16x16x64_i8' needs target feature gfx950-insts}}
 }
diff --git a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
index 596dc3c9656244..98781ed55dc47c 100644
--- a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
+++ b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
@@ -3138,6 +3138,7 @@ def int_amdgcn_cvt_sr_fp8_f32 : ClangBuiltin<"__builtin_amdgcn_cvt_sr_fp8_f32">,
 defset list<Intrinsic> AMDGPUMFMAIntrinsics950 = {
 def int_amdgcn_mfma_f32_16x16x32_f16 : AMDGPUMfmaIntrinsic<llvm_v4f32_ty, llvm_v8f16_ty>;
 def int_amdgcn_mfma_f32_32x32x16_f16 : AMDGPUMfmaIntrinsic<llvm_v16f32_ty, llvm_v8f16_ty>;
+def int_amdgcn_mfma_i32_16x16x64_i8 : AMDGPUMfmaIntrinsic<llvm_v4i32_ty, llvm_v4i32_ty>;
 
 def int_amdgcn_mfma_f32_32x32x16_bf16 : AMDGPUMfmaIntrinsic<llvm_v16f32_ty, llvm_v8bf16_ty>;
 def int_amdgcn_mfma_scale_f32_16x16x128_f8f6f4 : AMDGPUMfmaScaleIntrinsic<llvm_v4f32_ty, llvm_v8i32_ty>;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
index 0467d26bd3093d..9bb4e5be6299b5 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -4749,7 +4749,8 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
     case Intrinsic::amdgcn_mfma_f32_32x32x16_fp8_bf8:
     case Intrinsic::amdgcn_mfma_f32_32x32x16_fp8_fp8:
     case Intrinsic::amdgcn_mfma_f32_16x16x32_f16:
-    case Intrinsic::amdgcn_mfma_f32_32x32x16_f16: {
+    case Intrinsic::amdgcn_mfma_f32_32x32x16_f16:
+    case Intrinsic::amdgcn_mfma_i32_16x16x64_i8: {
       // Default for MAI intrinsics.
       // srcC can also be an immediate which can be folded later.
       // FIXME: Should we eventually add an alternative mapping with AGPR src
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.td b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
index cedbfc97d2c6be..ef5b14f90c7983 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.td
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
@@ -2880,6 +2880,7 @@ def VOP_V16F32_V8F16_V8F16_V16F32 : VOPProfile <[v16f32, v8f16, v8f16, v16f32]>;
 def VOP_V16F32_V8BF16_V8BF16_V16F32 : VOPProfile <[v16f32, v8bf16, v8bf16, v16f32]>;
 def VOP_V4F32_V8I32_V8I32_V4F32   : VOPProfile <[v4f32,  v8i32, v8i32, v4f32]>;
 def VOP_V16F32_V8I32_V8I32_V16F32   : VOPProfile <[v16f32, v8i32, v8i32, v16f32]>;
+def VOP_V4I32_V4I32_V4I32_V4I32     : VOPProfile <[v4i32,  v4i32, v4i32, v4i32]>;
 
 
 class Commutable_REV <string revOp, bit isOrig> {
diff --git a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
index f38d3f35cc8429..d1427a45f0e83d 100644
--- a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -647,6 +647,11 @@ def VOPProfileMAI_F32_V8I32_X128_VCD : VOPProfileMAI<VOP_V4F32_V8I32_V8I32_V4F32
 def VOPProfileMAI_F32_V8I32_X512     : VOPProfileMAI<VOP_V16F32_V8I32_V8I32_V16F32,    AISrc_512_f32,  ADst_512, AVSrc_256>;
 def VOPProfileMAI_F32_V8I32_X512_VCD : VOPProfileMAI<VOP_V16F32_V8I32_V8I32_V16F32,    VISrc_512_f32,  VDst_512, AVSrc_256>;
 
+// For i32_16x16x64_i8
+def VOPProfileMAI_I32_V4I32_X128     : VOPProfileMAI<VOP_V4I32_V4I32_V4I32_V4I32,    AISrc_128_f32,  ADst_128, AVSrc_128>;
+def VOPProfileMAI_I32_V4I32_X128_VCD : VOPProfileMAI<VOP_V4I32_V4I32_V4I32_V4I32,    VISrc_128_f32,  VDst_128, AVSrc_128>;
+
+
 class MFMATable <bit is_mac, string Name> {
   bit IsMac = is_mac;
   string FMAOp = Name;
@@ -815,6 +820,7 @@ defm V_MFMA_F32_32X32X4BF16 : MAIInst<"v_mfma_f32_32x32x4bf16", "F32_V2I16_X16",
 let SubtargetPredicate = HasGFX950Insts, is_gfx940_xdl = 1 in {
 defm V_MFMA_F32_16X16X32_F16   : MAIInst<"v_mfma_f32_16x16x32f16",    "F32_V8F16_X32", int_amdgcn_mfma_f32_16x16x32_f16>;
 defm V_MFMA_F32_32X32X16_F16   : MAIInst<"v_mfma_f32_32x32x16f16",    "F32_V8F16_X16", int_amdgcn_mfma_f32_32x32x16_f16>;
+defm V_MFMA_I32_16X16X64_I8    : MAIInst<"v_mfma_i32_16x16x64i8",     "I32_V4I32_X128", int_amdgcn_mfma_i32_16x16x64_i8>;
 defm V_MFMA_F32_32X32X16_BF16  : MAIInst<"v_mfma_f32_32x32x16bf16",   "F32_V8BF16_X16", int_amdgcn_mfma_f32_32x32x16_bf16>;
 
 defm V_MFMA_F32_16X16X128_F8F6F4 : MAIInst<"v_mfma_f32_16x16x128f8f6f4",
@@ -946,7 +952,6 @@ def VOP_V4F32_V16F16_V16F16_V4F32 : VOPProfile <[v4f32, v16f16, v16f16, v4f32]>;
 def VOP_V4F32_V16I16_V16I16_V4F32 : VOPProfile <[v4f32, v16i16, v16i16, v4f32]>;
 def VOP_V8F16_V16F16_V16F16_V8F16 : VOPProfile <[v8f16, v16f16, v16f16, v8f16]>;
 def VOP_V8I16_V16I16_V16I16_V8I16 : VOPProfile <[v8i16, v16i16, v16i16, v8i16]>;
-def VOP_V4I32_V4I32_V4I32_V4I32 : VOPProfile <[v4i32, v4i32, v4i32, v4i32]>;
 def VOP_V4I32_V2I32_V2I32_V4I32 : VOPProfile <[v4i32, v2i32, v2i32, v4i32]>;
 
 
@@ -1888,6 +1893,7 @@ defm V_MFMA_F64_4X4X4F64        : VOP3P_Real_MFMA_gfx90a <0x6f>;
 
 defm V_MFMA_F32_16X16X32_F16     : VOP3P_Real_MFMA_gfx950 <0x54, "v_mfma_f32_16x16x32_f16">;
 defm V_MFMA_F32_32X32X16_F16     : VOP3P_Real_MFMA_gfx950 <0x55, "v_mfma_f32_32x32x16_f16">;
+defm V_MFMA_I32_16X16X64_I8      : VOP3P_Real_MFMA_gfx950 <0x36, "v_mfma_i32_16x16x64_i8">;
 defm V_MFMA_F32_32X32X16_BF16    : VOP3P_Real_MFMA_gfx950 <0x37, "v_mfma_f32_32x32x16_bf16">;
 
 defm V_MFMA_F32_16X16X128_F8F6F4 : VOP3P_Real_MFMA_gfx950 <0x2d, "v_mfma_f32_16x16x128_f8f6f4">;
diff --git a/llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll b/llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll
index a8684a55ab5c2d..9a6eb32e7e8780 100644
--- a/llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll
+++ b/llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll
@@ -305,6 +305,14 @@ define amdgpu_kernel void @mfma_f32_scale_32x32x64_f8f6f4(<8 x i32> %arg0, <8 x
   ret void
 }
 
+declare <4 x i32> @llvm.amdgcn.mfma.i32.16x16x64.i8(<4 x i32>, <4 x i32>, <4 x i32>, i32 immarg, i32 immarg, i32 immarg)
+
+; CHECK: DIVERGENT: %result = call <4 x i32> @llvm.amdgcn.mfma.i32.16x16x64.i8(<4 x i32> %arg0, <4 x i32> %arg1, <4 x i32> %arg2, i32 immarg 0, i32 immarg 0, i32 immarg 0)
+define amdgpu_kernel void @mfma_i32_16x16x64_i8(<4 x i32> %arg0, <4 x i32> %arg1, <4 x i32> %arg2, ptr addrspace(1) %out) {
+  %result = call <4 x i32> @llvm.amdgcn.mfma.i32.16x16x64.i8(<4 x i32> %arg0, <4 x i32> %arg1, <4 x i32> %arg2, i32 immarg 0, i32 immarg 0, i32 immarg 0)
+  store <4 x i32> %result, ptr addrspace(1) %out
+  ret void
+}
 
 declare i32 @llvm.amdgcn.ds.swizzle(i32, i32) #1
 declare i32 @llvm.amdgcn.permlane16.i32(i32, i32, i32, i32, i1, i1) #1
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll
index 88d04e9fb428a2..555756d3fdf3f8 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll
@@ -272,3 +272,288 @@ define <16 x float> @test_mfma_f32_32x32x16_f16__flags__mac(<16 x float> %arg2,
   %result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.f16(<8 x half> %arg0, <8 x half> %arg1, <16 x float> %arg2, i32 1, i32 1, i32 1)
   ret <16 x float> %result
 }
+
+define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd_mac(<8 x half> %arg0, <8 x half> %arg1, <16 x float> %arg2, ptr addrspace(1) %out) #0 {
+; SDAG-LABEL: test_mfma_f32_32x32x16_f16__vgprcd_mac:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_load_dwordx8 s[20:27], s[0:1], 0x24
+; SDAG-NEXT:    s_load_dwordx16 s[4:19], s[0:1], 0x64
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[20:21]
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[22:23]
+; SDAG-NEXT:    v_mov_b64_e32 v[20:21], s[24:25]
+; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[4:5]
+; SDAG-NEXT:    v_mov_b64_e32 v[22:23], s[26:27]
+; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[6:7]
+; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[8:9]
+; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[10:11]
+; SDAG-NEXT:    v_mov_b64_e32 v[8:9], s[12:13]
+; SDAG-NEXT:    v_mov_b64_e32 v[10:11], s[14:15]
+; SDAG-NEXT:    v_mov_b64_e32 v[12:13], s[16:17]
+; SDAG-NEXT:    v_mov_b64_e32 v[14:15], s[18:19]
+; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xa4
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[16:19], v[20:23], v[0:15]
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    s_nop 7
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; SDAG-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; SDAG-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
+; SDAG-NEXT:    s_endpgm
+;
+; GISEL-LABEL: test_mfma_f32_32x32x16_f16__vgprcd_mac:
+; GISEL:       ; %bb.0:
+; GISEL-NEXT:    s_load_dwordx8 s[20:27], s[0:1], 0x24
+; GISEL-NEXT:    s_load_dwordx16 s[4:19], s[0:1], 0x64
+; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[4:5]
+; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[26:27]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[6:7]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[18:19]
+; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xa4
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[16:19], v[20:23], v[0:15]
+; GISEL-NEXT:    v_mov_b32_e32 v16, 0
+; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-NEXT:    s_nop 7
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
+; GISEL-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; GISEL-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; GISEL-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; GISEL-NEXT:    s_endpgm
+  %result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.f16(<8 x half> %arg0, <8 x half> %arg1, <16 x float> %arg2, i32 0, i32 0, i32 0)
+  store <16 x float> %result, ptr addrspace(1) %out
+  ret void
+}
+
+define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd_mac_flags(<8 x half> %arg0, <8 x half> %arg1, <16 x float> %arg2, ptr addrspace(1) %out) #0 {
+; SDAG-LABEL: test_mfma_f32_32x32x16_f16__vgprcd_mac_flags:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_load_dwordx8 s[20:27], s[0:1], 0x24
+; SDAG-NEXT:    s_load_dwordx16 s[4:19], s[0:1], 0x64
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[20:21]
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[22:23]
+; SDAG-NEXT:    v_mov_b64_e32 v[20:21], s[24:25]
+; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[4:5]
+; SDAG-NEXT:    v_mov_b64_e32 v[22:23], s[26:27]
+; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[6:7]
+; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[8:9]
+; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[10:11]
+; SDAG-NEXT:    v_mov_b64_e32 v[8:9], s[12:13]
+; SDAG-NEXT:    v_mov_b64_e32 v[10:11], s[14:15]
+; SDAG-NEXT:    v_mov_b64_e32 v[12:13], s[16:17]
+; SDAG-NEXT:    v_mov_b64_e32 v[14:15], s[18:19]
+; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xa4
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[16:19], v[20:23], v[0:15] cbsz:3 abid:2 blgp:1
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    s_nop 7
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; SDAG-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; SDAG-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
+; SDAG-NEXT:    s_endpgm
+;
+; GISEL-LABEL: test_mfma_f32_32x32x16_f16__vgprcd_mac_flags:
+; GISEL:       ; %bb.0:
+; GISEL-NEXT:    s_load_dwordx8 s[20:27], s[0:1], 0x24
+; GISEL-NEXT:    s_load_dwordx16 s[4:19], s[0:1], 0x64
+; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[4:5]
+; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[26:27]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[6:7]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[18:19]
+; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xa4
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[16:19], v[20:23], v[0:15] cbsz:3 abid:2 blgp:1
+; GISEL-NEXT:    v_mov_b32_e32 v16, 0
+; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-NEXT:    s_nop 7
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
+; GISEL-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; GISEL-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; GISEL-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; GISEL-NEXT:    s_endpgm
+  %result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.f16(<8 x half> %arg0, <8 x half> %arg1, <16 x float> %arg2, i32 3, i32 2, i32 1)
+  store <16 x float> %result, ptr addrspace(1) %out
+  ret void
+}
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.mfma.i32.16x16x64.i8
+; --------------------------------------------------------------------
+
+declare <4 x i32> @llvm.amdgcn.mfma.i32.16x16x64.i8(<4 x i32>, <4 x i32>, <4 x i32>, i32 immarg, i32 immarg, i32 immarg)
+
+define <4 x i32> @test_mfma_i32_16x16x64_i8(<4 x i32> %arg0, <4 x i32> %arg1, <4 x i32> %arg2) {
+; GCN-LABEL: test_mfma_i32_16x16x64_i8:
+; GCN:       ; %bb.0:
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_accvgpr_write_b32 a0, v8
+; GCN-NEXT:    v_accvgpr_write_b32 a1, v9
+; GCN-NEXT:    v_accvgpr_write_b32 a2, v10
+; GCN-NEXT:    v_accvgpr_write_b32 a3, v11
+; GCN-NEXT:    s_nop 1
+; GCN-NEXT:    v_mfma_i32_16x16x64_i8 a[0:3], v[0:3], v[4:7], a[0:3]
+; GCN-NEXT:    s_nop 3
+; GCN-NEXT:    v_accvgpr_read_b32 v0, a0
+; GCN-NEXT:    v_accvgpr_read_b32 v1, a1
+; GCN-NEXT:    v_accvgpr_read_b32 v2, a2
+; GCN-NEXT:    v_accvgpr_read_b32 v3, a3
+; GCN-NEXT:    s_setpc_b64 s[30:31]
+  %result = call <4 x i32> @llvm.amdgcn.mfma.i32.16x16x64.i8(<4 x i32> %arg0, <4 x i32> %arg1, <4 x i32> %arg2, i32 0, i32 0, i32 0)
+  ret <4 x i32> %result
+}
+
+define <4 x i32> @test_mfma_i32_16x16x64_i8__flags(<4 x i32> %arg0, <4 x i32> %arg1, <4 x i32> %arg2) {
+; GCN-LABEL: test_mfma_i32_16x16x64_i8__flags:
+; GCN:       ; %bb.0:
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_accvgpr_write_b32 a0, v8
+; GCN-NEXT:    v_accvgpr_write_b32 a1, v9
+; GCN-NEXT:    v_accvgpr_write_b32 a2, v10
+; GCN-NEXT:    v_accvgpr_write_b32 a3, v11
+; GCN-NEXT:    s_nop 1
+; GCN-NEXT:    v_mfma_i32_16x16x64_i8 a[0:3], v[0:3], v[4:7], a[0:3] cbsz:1 abid:1 blgp:1
+; GCN-NEXT:    s_nop 3
+; GCN-NEXT:    v_accvgpr_read_b32 v0, a0
+; GCN-NEXT:    v_accvgpr_read_b32 v1, a1
+; GCN-NEXT:    v_accvgpr_read_b32 v2, a2
+; GCN-NEXT:    v_accvgpr_read_b32 v3, a3
+; GCN-NEXT:    s_setpc_b64 s[30:31]
+  %result = call <4 x i32> @llvm.amdgcn.mfma.i32.16x16x64.i8(<4 x i32> %arg0, <4 x i32> %arg1, <4 x i32> %arg2, i32 1, i32 1, i32 1)
+  ret <4 x i32...
[truncated]

``````````

</details>


https://github.com/llvm/llvm-project/pull/116728


More information about the llvm-branch-commits mailing list