[llvm-branch-commits] [clang] [llvm] AMDGPU: Add v_mfma_i32_32x32x32_i8 for gfx950 (PR #117052)

Matt Arsenault via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Wed Nov 20 13:22:42 PST 2024


https://github.com/arsenm created https://github.com/llvm/llvm-project/pull/117052

None

>From 522f52695d561499f935a527c230ceec079eba58 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Mon, 29 Jan 2024 17:44:14 +0530
Subject: [PATCH] AMDGPU: Add v_mfma_i32_32x32x32_i8 for gfx950

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def  |   1 +
 .../CodeGenOpenCL/builtins-amdgcn-mfma.cl     |   6 +
 .../builtins-amdgcn-error-gfx950-param.cl     |   7 +
 .../builtins-amdgcn-error-gfx950.cl           |   1 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td      |   1 +
 .../Target/AMDGPU/AMDGPURegisterBankInfo.cpp  |   3 +-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td   |   6 +
 .../UniformityAnalysis/AMDGPU/intrinsics.ll   |   9 +
 .../CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll | 760 ++++++++++++++++++
 llvm/test/MC/AMDGPU/mai-gfx950.s              |  48 ++
 .../MC/Disassembler/AMDGPU/gfx950_mai.txt     |  28 +
 llvm/test/tools/llvm-mca/AMDGPU/gfx950.s      |  10 +-
 12 files changed, 876 insertions(+), 4 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index b4c7c12fec6475..6c51e52a2ef665 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -441,6 +441,7 @@ TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x32_f16, "V4fV8hV8hV4fIiIiIi", "nc
 TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_f16, "V16fV8hV8hV16fIiIiIi", "nc", "gfx950-insts")
 TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_bf16, "V16fV8yV8yV16fIiIiIi", "nc", "gfx950-insts")
 TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_16x16x64_i8, "V4iV4iV4iV4iIiIiIi", "nc", "gfx950-insts")
+TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_32x32x32_i8, "V16iV4iV4iV16iIiIiIi", "nc", "gfx950-insts")
 
 //===----------------------------------------------------------------------===//
 // GFX12+ only builtins.
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl b/clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl
index b69db6410a7905..b21394b6982631 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl
@@ -452,4 +452,10 @@ 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);
 }
 
+// CHECK-GFX950-LABEL: @test_mfma_i32_32x32x32_i8(
+// CHECK-GFX950: tail call <16 x i32> @llvm.amdgcn.mfma.i32.32x32x32.i8(<4 x i32> %a, <4 x i32> %b, <16 x i32> %c, i32 1, i32 2, i32 3)
+v16i test_mfma_i32_32x32x32_i8(v4i a, v4i b, v16i c) {
+  return __builtin_amdgcn_mfma_i32_32x32x32_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 3597042fadcf88..9c14c0541ff3b8 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl
@@ -7,6 +7,7 @@ typedef half half8 __attribute__((ext_vector_type(8)));
 typedef __bf16 bfloat8 __attribute__((ext_vector_type(8)));
 typedef int int4 __attribute__((ext_vector_type(4)));
 typedef int int8 __attribute__((ext_vector_type(8)));
+typedef int int16 __attribute__((ext_vector_type(16)));
 
 
 void test_mfma_f32_16x16x32_f16(__global float4* out, half8 a, half8 b, float4 c, int X) {
@@ -48,3 +49,9 @@ 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, 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}}
 }
+
+void test_mfma_i32_32x32x32_i8(__global int16* out, int4 a, int4 b, int16 c, int X) {
+  *out = __builtin_amdgcn_mfma_i32_32x32x32_i8(a, b, c, X, 0, 0); // expected-error{{argument to '__builtin_amdgcn_mfma_i32_32x32x32_i8' must be a constant integer}}
+  *out = __builtin_amdgcn_mfma_i32_32x32x32_i8(a, b, c, 0, X, 0);  // expected-error{{argument to '__builtin_amdgcn_mfma_i32_32x32x32_i8' must be a constant integer}}
+  *out = __builtin_amdgcn_mfma_i32_32x32x32_i8(a, b, c, 0, 0, X);  // expected-error{{argument to '__builtin_amdgcn_mfma_i32_32x32x32_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 3a27fbf2439353..71a110066342cb 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950.cl
@@ -32,6 +32,7 @@ void test(__global float4* out0, half8 a0, half8 b0, float4 c0,
   *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}}
+  *out4 = __builtin_amdgcn_mfma_i32_32x32x32_i8(a4, b4, c4, 0, 0, 0); // expected-error{{'__builtin_amdgcn_mfma_i32_32x32x32_i8' needs target feature gfx950-insts}}
   *out14 = __builtin_amdgcn_mfma_scale_f32_16x16x128_f8f6f4(a14, b14, c14, 0, 0, 0, d14, 0, e14); // expected-error{{'__builtin_amdgcn_mfma_scale_f32_16x16x128_f8f6f4' needs target feature gfx950-insts}}
   *out15 = __builtin_amdgcn_mfma_scale_f32_32x32x64_f8f6f4(a15, b15, c15, 0, 0, 0, d15, 0, e15); // expected-error{{'__builtin_amdgcn_mfma_scale_f32_32x32x64_f8f6f4' needs target feature gfx950-insts}}
 }
diff --git a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
index 35f45095b1ad04..b5d5eae0c7cd7e 100644
--- a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
+++ b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
@@ -3147,6 +3147,7 @@ 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_i32_32x32x32_i8 : AMDGPUMfmaIntrinsic<llvm_v16i32_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>;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
index f8ef1879496bb2..ea9c688c710418 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -4750,7 +4750,8 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
     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_i32_16x16x64_i8: {
+    case Intrinsic::amdgcn_mfma_i32_16x16x64_i8:
+    case Intrinsic::amdgcn_mfma_i32_32x32x32_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/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
index 7a83e76ba4c919..98d5e3f199cddd 100644
--- a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -725,6 +725,10 @@ def VOPProfileMAI_F32_V4I32_V4I32_X512_VCD : VOPProfileMAI<VOP_V16F32_V4I32_V4I3
 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>;
 
+// For i32_32x32x32_i8
+def VOPProfileMAI_I32_V4I32_X16     : VOPProfileMAI<VOP_V16I32_V4I32_V4I32_V16I32,    AISrc_512_b32,  ADst_512, AVSrc_128>;
+def VOPProfileMAI_I32_V4I32_X16_VCD : VOPProfileMAI<VOP_V16I32_V4I32_V4I32_V16I32,    VISrc_512_b32,  VDst_512, AVSrc_128>;
+
 
 class MFMATable <bit is_mac, string Name> {
   bit IsMac = is_mac;
@@ -950,6 +954,7 @@ defm V_MFMA_F32_16X16X32_F16   : MAIInst<"v_mfma_f32_16x16x32f16",    "F32_V8F16
 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_I32_32X32X32_I8    : MAIInst<"v_mfma_i32_32x32x32i8",     "I32_V4I32_X16", int_amdgcn_mfma_i32_32x32x32_i8>;
 
 defm V_MFMA_F32_16X16X128_F8F6F4 : MAIInst_SrcFormats_mc<"v_mfma_f32_16x16x128f8f6f4",
   "_X128", mfma_f32_16x16x128_f8f6f4>;
@@ -2075,6 +2080,7 @@ defm V_MFMA_F32_16X16X32_F16     : VOP3P_Real_MFMA_gfx950 <0x54, "v_mfma_f32_16x
 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_I32_32X32X32_I8      : VOP3P_Real_MFMA_gfx950 <0x38, "v_mfma_i32_32x32x32_i8">;
 
 defm V_MFMA_LD_SCALE_B32 : VOP3P_Real_vi <0x2c>;
 defm V_MFMA_F32_16X16X128_F8F6F4 : VOP3P_Real_MFMA_F8F6F4_gfx950_mc <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 48c6c42b66b9a5..26021a56790aae 100644
--- a/llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll
+++ b/llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll
@@ -314,6 +314,15 @@ define amdgpu_kernel void @mfma_i32_16x16x64_i8(<4 x i32> %arg0, <4 x i32> %arg1
   ret void
 }
 
+declare <16 x i32> @llvm.amdgcn.mfma.i32.32x32x32.i8(<4 x i32>, <4 x i32>, <16 x i32>, i32 immarg, i32 immarg, i32 immarg)
+
+; CHECK: DIVERGENT: %result = call <16 x i32> @llvm.amdgcn.mfma.i32.32x32x32.i8(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, i32 immarg 0, i32 immarg 0, i32 immarg 0)
+define amdgpu_kernel void @mfma_i32_32x32x32_i8(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, ptr addrspace(1) %out) {
+  %result = call <16 x i32> @llvm.amdgcn.mfma.i32.32x32x32.i8(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, i32 immarg 0, i32 immarg 0, i32 immarg 0)
+  store <16 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
 declare i32 @llvm.amdgcn.permlanex16.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 555756d3fdf3f8..29668b2c5bfdbf 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll
@@ -555,5 +555,765 @@ define amdgpu_kernel void @test_mfma_i32_16x16x64_i8_no_agpr__vgprcd__flags(ptr
   ret void
 }
 
+; --------------------------------------------------------------------
+; llvm.amdgcn.mfma.i32.32x32x32.i8
+; --------------------------------------------------------------------
+
+declare <16 x i32> @llvm.amdgcn.mfma.i32.32x32x32.i8(<4 x i32>, <4 x i32>, <16 x i32>, i32 immarg, i32 immarg, i32 immarg)
+
+define amdgpu_kernel void @test_mfma_i32_32x32x32_i8(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2) #1 {
+; SDAG-LABEL: test_mfma_i32_32x32x32_i8:
+; 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:    v_mov_b64_e32 v[8:9], 16
+; SDAG-NEXT:    v_mov_b64_e32 v[10:11], 0
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b32_e32 v0, s20
+; SDAG-NEXT:    v_accvgpr_write_b32 a31, s19
+; SDAG-NEXT:    v_mov_b32_e32 v1, s21
+; SDAG-NEXT:    v_mov_b32_e32 v2, s22
+; SDAG-NEXT:    v_mov_b32_e32 v3, s23
+; SDAG-NEXT:    v_mov_b32_e32 v4, s24
+; SDAG-NEXT:    v_mov_b32_e32 v5, s25
+; SDAG-NEXT:    v_mov_b32_e32 v6, s26
+; SDAG-NEXT:    v_mov_b32_e32 v7, s27
+; SDAG-NEXT:    v_accvgpr_write_b32 a30, s18
+; SDAG-NEXT:    v_accvgpr_write_b32 a29, s17
+; SDAG-NEXT:    v_accvgpr_write_b32 a28, s16
+; SDAG-NEXT:    v_accvgpr_write_b32 a27, s15
+; SDAG-NEXT:    v_accvgpr_write_b32 a26, s14
+; SDAG-NEXT:    v_accvgpr_write_b32 a25, s13
+; SDAG-NEXT:    v_accvgpr_write_b32 a24, s12
+; SDAG-NEXT:    v_accvgpr_write_b32 a23, s11
+; SDAG-NEXT:    v_accvgpr_write_b32 a22, s10
+; SDAG-NEXT:    v_accvgpr_write_b32 a21, s9
+; SDAG-NEXT:    v_accvgpr_write_b32 a20, s8
+; SDAG-NEXT:    v_accvgpr_write_b32 a19, s7
+; SDAG-NEXT:    v_accvgpr_write_b32 a18, s6
+; SDAG-NEXT:    v_accvgpr_write_b32 a17, s5
+; SDAG-NEXT:    v_accvgpr_write_b32 a16, s4
+; SDAG-NEXT:    s_nop 1
+; SDAG-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[0:3], v[4:7], a[16:31]
+; SDAG-NEXT:    v_mov_b64_e32 v[4:5], 48
+; SDAG-NEXT:    v_mov_b64_e32 v[6:7], 32
+; SDAG-NEXT:    v_mov_b32_e32 v0, s12
+; SDAG-NEXT:    v_mov_b32_e32 v1, s13
+; SDAG-NEXT:    v_mov_b32_e32 v2, s14
+; SDAG-NEXT:    v_mov_b32_e32 v3, s15
+; SDAG-NEXT:    s_nop 7
+; SDAG-NEXT:    s_nop 4
+; SDAG-NEXT:    global_store_dwordx4 v[4:5], a[12:15], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v[6:7], a[8:11], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v[8:9], a[4:7], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v[10:11], a[0:3], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v[6:7], v[0:3], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mov_b32_e32 v0, s16
+; SDAG-NEXT:    v_mov_b32_e32 v1, s17
+; SDAG-NEXT:    v_mov_b32_e32 v2, s18
+; SDAG-NEXT:    v_mov_b32_e32 v3, s19
+; SDAG-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mov_b32_e32 v0, s4
+; SDAG-NEXT:    v_mov_b32_e32 v1, s5
+; SDAG-NEXT:    v_mov_b32_e32 v2, s6
+; SDAG-NEXT:    v_mov_b32_e32 v3, s7
+; SDAG-NEXT:    global_store_dwordx4 v[10:11], v[0:3], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mov_b32_e32 v0, s8
+; SDAG-NEXT:    v_mov_b32_e32 v1, s9
+; SDAG-NEXT:    v_mov_b32_e32 v2, s10
+; SDAG-NEXT:    v_mov_b32_e32 v3, s11
+; SDAG-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_endpgm
+;
+; GISEL-LABEL: test_mfma_i32_32x32x32_i8:
+; 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:    v_mov_b64_e32 v[8:9], 0
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], 48
+; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[24:25]
+; GISEL-NEXT:    v_accvgpr_write_b32 a31, s19
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[26:27]
+; GISEL-NEXT:    v_accvgpr_write_b32 a30, s18
+; GISEL-NEXT:    v_accvgpr_write_b32 a29, s17
+; GISEL-NEXT:    v_accvgpr_write_b32 a28, s16
+; GISEL-NEXT:    v_accvgpr_write_b32 a27, s15
+; GISEL-NEXT:    v_accvgpr_write_b32 a26, s14
+; GISEL-NEXT:    v_accvgpr_write_b32 a25, s13
+; GISEL-NEXT:    v_accvgpr_write_b32 a24, s12
+; GISEL-NEXT:    v_accvgpr_write_b32 a23, s11
+; GISEL-NEXT:    v_accvgpr_write_b32 a22, s10
+; GISEL-NEXT:    v_accvgpr_write_b32 a21, s9
+; GISEL-NEXT:    v_accvgpr_write_b32 a20, s8
+; GISEL-NEXT:    v_accvgpr_write_b32 a19, s7
+; GISEL-NEXT:    v_accvgpr_write_b32 a18, s6
+; GISEL-NEXT:    v_accvgpr_write_b32 a17, s5
+; GISEL-NEXT:    v_accvgpr_write_b32 a16, s4
+; GISEL-NEXT:    s_nop 1
+; GISEL-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[0:3], v[4:7], a[16:31]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[4:5]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], 16
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], 32
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[6:7]
+; GISEL-NEXT:    s_nop 7
+; GISEL-NEXT:    s_nop 6
+; GISEL-NEXT:    global_store_dwordx4 v[8:9], a[0:3], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v[4:5], a[4:7], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v[6:7], a[8:11], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v[10:11], a[12:15], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; GISEL-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[14:15]
+; GISEL-NEXT:    global_store_dwordx4 v[6:7], v[0:3], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[18:19]
+; GISEL-NEXT:    global_store_dwordx4 v[10:11], v[0:3], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_endpgm
+  %result = call <16 x i32> @llvm.amdgcn.mfma.i32.32x32x32.i8(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, i32 0, i32 0, i32 0)
+  store volatile <16 x i32> %result, ptr addrspace(1) null
+  store volatile <16 x i32> %arg2, ptr addrspace(1) null
+  ret void
+}
+
+define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__flags(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2) #1 {
+; SDAG-LABEL: test_mfma_i32_32x32x32_i8__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:    v_mov_b64_e32 v[8:9], 16
+; SDAG-NEXT:    v_mov_b64_e32 v[10:11], 0
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b32_e32 v0, s20
+; SDAG-NEXT:    v_accvgpr_write_b32 a31, s19
+; SDAG-NEXT:    v_mov_b32_e32 v1, s21
+; SDAG-NEXT:    v_mov_b32_e32 v2, s22
+; SDAG-NEXT:    v_mov_b32_e32 v3, s23
+; SDAG-NEXT:    v_mov_b32_e32 v4, s24
+; SDAG-NEXT:    v_mov_b32_e32 v5, s25
+; SDAG-NEXT:    v_mov_b32_e32 v6, s26
+; SDAG-NEXT:    v_mov_b32_e32 v7, s27
+; SDAG-NEXT:    v_accvgpr_write_b32 a30, s18
+; SDAG-NEXT:    v_accvgpr_write_b32 a29, s17
+; SDAG-NEXT:    v_accvgpr_write_b32 a28, s16
+; SDAG-NEXT:    v_accvgpr_write_b32 a27, s15
+; SDAG-NEXT:    v_accvgpr_write_b32 a26, s14
+; SDAG-NEXT:    v_accvgpr_write_b32 a25, s13
+; SDAG-NEXT:    v_accvgpr_write_b32 a24, s12
+; SDAG-NEXT:    v_accvgpr_write_b32 a23, s11
+; SDAG-NEXT:    v_accvgpr_write_b32 a22, s10
+; SDAG-NEXT:    v_accvgpr_write_b32 a21, s9
+; SDAG-NEXT:    v_accvgpr_write_b32 a20, s8
+; SDAG-NEXT:    v_accvgpr_write_b32 a19, s7
+; SDAG-NEXT:    v_accvgpr_write_b32 a18, s6
+; SDAG-NEXT:    v_accvgpr_write_b32 a17, s5
+; SDAG-NEXT:    v_accvgpr_write_b32 a16, s4
+; SDAG-NEXT:    s_nop 1
+; SDAG-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[0:3], v[4:7], a[16:31] cbsz:2 abid:3 blgp:1
+; SDAG-NEXT:    v_mov_b64_e32 v[4:5], 48
+; SDAG-NEXT:    v_mov_b64_e32 v[6:7], 32
+; SDAG-NEXT:    v_mov_b32_e32 v0, s12
+; SDAG-NEXT:    v_mov_b32_e32 v1, s13
+; SDAG-NEXT:    v_mov_b32_e32 v2, s14
+; SDAG-NEXT:    v_mov_b32_e32 v3, s15
+; SDAG-NEXT:    s_nop 7
+; SDAG-NEXT:    s_nop 4
+; SDAG-NEXT:    global_store_dwordx4 v[4:5], a[12:15], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v[6:7], a[8:11], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v[8:9], a[4:7], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v[10:11], a[0:3], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v[6:7], v[0:3], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mov_b32_e32 v0, s16
+; SDAG-NEXT:    v_mov_b32_e32 v1, s17
+; SDAG-NEXT:    v_mov_b32_e32 v2, s18
+; SDAG-NEXT:    v_mov_b32_e32 v3, s19
+; SDAG-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mov_b32_e32 v0, s4
+; SDAG-NEXT:    v_mov_b32_e32 v1, s5
+; SDAG-NEXT:    v_mov_b32_e32 v2, s6
+; SDAG-NEXT:    v_mov_b32_e32 v3, s7
+; SDAG-NEXT:    global_store_dwordx4 v[10:11], v[0:3], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mov_b32_e32 v0, s8
+; SDAG-NEXT:    v_mov_b32_e32 v1, s9
+; SDAG-NEXT:    v_mov_b32_e32 v2, s10
+; SDAG-NEXT:    v_mov_b32_e32 v3, s11
+; SDAG-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_endpgm
+;
+; GISEL-LABEL: test_mfma_i32_32x32x32_i8__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:    v_mov_b64_e32 v[8:9], 0
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], 48
+; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[24:25]
+; GISEL-NEXT:    v_accvgpr_write_b32 a31, s19
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[26:27]
+; GISEL-NEXT:    v_accvgpr_write_b32 a30, s18
+; GISEL-NEXT:    v_accvgpr_write_b32 a29, s17
+; GISEL-NEXT:    v_accvgpr_write_b32 a28, s16
+; GISEL-NEXT:    v_accvgpr_write_b32 a27, s15
+; GISEL-NEXT:    v_accvgpr_write_b32 a26, s14
+; GISEL-NEXT:    v_accvgpr_write_b32 a25, s13
+; GISEL-NEXT:    v_accvgpr_write_b32 a24, s12
+; GISEL-NEXT:    v_accvgpr_write_b32 a23, s11
+; GISEL-NEXT:    v_accvgpr_write_b32 a22, s10
+; GISEL-NEXT:    v_accvgpr_write_b32 a21, s9
+; GISEL-NEXT:    v_accvgpr_write_b32 a20, s8
+; GISEL-NEXT:    v_accvgpr_write_b32 a19, s7
+; GISEL-NEXT:    v_accvgpr_write_b32 a18, s6
+; GISEL-NEXT:    v_accvgpr_write_b32 a17, s5
+; GISEL-NEXT:    v_accvgpr_write_b32 a16, s4
+; GISEL-NEXT:    s_nop 1
+; GISEL-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[0:3], v[4:7], a[16:31] cbsz:2 abid:3 blgp:1
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[4:5]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], 16
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], 32
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[6:7]
+; GISEL-NEXT:    s_nop 7
+; GISEL-NEXT:    s_nop 6
+; GISEL-NEXT:    global_store_dwordx4 v[8:9], a[0:3], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v[4:5], a[4:7], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v[6:7], a[8:11], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v[10:11], a[12:15], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; GISEL-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[14:15]
+; GISEL-NEXT:    global_store_dwordx4 v[6:7], v[0:3], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[18:19]
+; GISEL-NEXT:    global_store_dwordx4 v[10:11], v[0:3], off sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_endpgm
+  %result = call <16 x i32> @llvm.amdgcn.mfma.i32.32x32x32.i8(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, i32 2, i32 3, i32 1)
+  store volatile <16 x i32> %result, ptr addrspace(1) null
+  store volatile <16 x i32> %arg2, ptr addrspace(1) null
+  ret void
+}
+
+define <16 x i32> @test_mfma_i32_32x32x32_i8__mac(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2) {
+; GCN-LABEL: test_mfma_i32_32x32x32_i8__mac:
+; 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:    v_accvgpr_write_b32 a4, v12
+; GCN-NEXT:    v_accvgpr_write_b32 a5, v13
+; GCN-NEXT:    v_accvgpr_write_b32 a6, v14
+; GCN-NEXT:    v_accvgpr_write_b32 a7, v15
+; GCN-NEXT:    v_accvgpr_write_b32 a8, v16
+; GCN-NEXT:    v_accvgpr_write_b32 a9, v17
+; GCN-NEXT:    v_accvgpr_write_b32 a10, v18
+; GCN-NEXT:    v_accvgpr_write_b32 a11, v19
+; GCN-NEXT:    v_accvgpr_write_b32 a12, v20
+; GCN-NEXT:    v_accvgpr_write_b32 a13, v21
+; GCN-NEXT:    v_accvgpr_write_b32 a14, v22
+; GCN-NEXT:    v_accvgpr_write_b32 a15, v23
+; GCN-NEXT:    s_nop 1
+; GCN-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[0:3], v[4:7], a[0:15]
+; 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:    v_accvgpr_read_b32 v4, a4
+; GCN-NEXT:    v_accvgpr_read_b32 v5, a5
+; GCN-NEXT:    v_accvgpr_read_b32 v6, a6
+; GCN-NEXT:    v_accvgpr_read_b32 v7, a7
+; GCN-NEXT:    v_accvgpr_read_b32 v8, a8
+; GCN-NEXT:    v_accvgpr_read_b32 v9, a9
+; GCN-NEXT:    v_accvgpr_read_b32 v10, a10
+; GCN-NEXT:    v_accvgpr_read_b32 v11, a11
+; GCN-NEXT:    v_accvgpr_read_b32 v12, a12
+; GCN-NEXT:    v_accvgpr_read_b32 v13, a13
+; GCN-NEXT:    v_accvgpr_read_b32 v14, a14
+; GCN-NEXT:    v_accvgpr_read_b32 v15, a15
+; GCN-NEXT:    s_setpc_b64 s[30:31]
+  %result = call <16 x i32> @llvm.amdgcn.mfma.i32.32x32x32.i8(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, i32 0, i32 0, i32 0)
+  ret <16 x i32> %result
+}
+
+define <16 x i32> @test_mfma_i32_32x32x32_i8__mac__flags(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2) {
+; GCN-LABEL: test_mfma_i32_32x32x32_i8__mac__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:    v_accvgpr_write_b32 a4, v12
+; GCN-NEXT:    v_accvgpr_write_b32 a5, v13
+; GCN-NEXT:    v_accvgpr_write_b32 a6, v14
+; GCN-NEXT:    v_accvgpr_write_b32 a7, v15
+; GCN-NEXT:    v_accvgpr_write_b32 a8, v16
+; GCN-NEXT:    v_accvgpr_write_b32 a9, v17
+; GCN-NEXT:    v_accvgpr_write_b32 a10, v18
+; GCN-NEXT:    v_accvgpr_write_b32 a11, v19
+; GCN-NEXT:    v_accvgpr_write_b32 a12, v20
+; GCN-NEXT:    v_accvgpr_write_b32 a13, v21
+; GCN-NEXT:    v_accvgpr_write_b32 a14, v22
+; GCN-NEXT:    v_accvgpr_write_b32 a15, v23
+; GCN-NEXT:    s_nop 1
+; GCN-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[0:3], v[4:7], a[0:15] 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:    v_accvgpr_read_b32 v4, a4
+; GCN-NEXT:    v_accvgpr_read_b32 v5, a5
+; GCN-NEXT:    v_accvgpr_read_b32 v6, a6
+; GCN-NEXT:    v_accvgpr_read_b32 v7, a7
+; GCN-NEXT:    v_accvgpr_read_b32 v8, a8
+; GCN-NEXT:    v_accvgpr_read_b32 v9, a9
+; GCN-NEXT:    v_accvgpr_read_b32 v10, a10
+; GCN-NEXT:    v_accvgpr_read_b32 v11, a11
+; GCN-NEXT:    v_accvgpr_read_b32 v12, a12
+; GCN-NEXT:    v_accvgpr_read_b32 v13, a13
+; GCN-NEXT:    v_accvgpr_read_b32 v14, a14
+; GCN-NEXT:    v_accvgpr_read_b32 v15, a15
+; GCN-NEXT:    s_setpc_b64 s[30:31]
+  %result = call <16 x i32> @llvm.amdgcn.mfma.i32.32x32x32.i8(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, i32 1, i32 1, i32 1)
+  ret <16 x i32> %result
+}
+
+define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, ptr addrspace(1) %out) #0 {
+; SDAG-LABEL: test_mfma_i32_32x32x32_i8__vgprcd:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_load_dwordx8 s[16:23], s[0:1], 0x24
+; SDAG-NEXT:    v_mov_b32_e32 v40, 0
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b32_e32 v32, s16
+; SDAG-NEXT:    v_mov_b32_e32 v33, s17
+; SDAG-NEXT:    v_mov_b32_e32 v34, s18
+; SDAG-NEXT:    v_mov_b32_e32 v35, s19
+; SDAG-NEXT:    s_load_dwordx16 s[4:19], s[0:1], 0x64
+; SDAG-NEXT:    v_mov_b32_e32 v36, s20
+; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xa4
+; SDAG-NEXT:    v_mov_b32_e32 v37, s21
+; SDAG-NEXT:    v_mov_b32_e32 v38, s22
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b64_e32 v[30:31], s[18:19]
+; SDAG-NEXT:    v_mov_b32_e32 v39, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[28:29], s[16:17]
+; SDAG-NEXT:    v_mov_b64_e32 v[26:27], s[14:15]
+; SDAG-NEXT:    v_mov_b64_e32 v[24:25], s[12:13]
+; SDAG-NEXT:    v_mov_b64_e32 v[22:23], s[10:11]
+; SDAG-NEXT:    v_mov_b64_e32 v[20:21], s[8:9]
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[6:7]
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[4:5]
+; SDAG-NEXT:    s_nop 1
+; SDAG-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[32:35], v[36:39], v[16:31]
+; SDAG-NEXT:    s_nop 7
+; SDAG-NEXT:    s_nop 6
+; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v17, s17
+; SDAG-NEXT:    v_mov_b32_e32 v18, s18
+; SDAG-NEXT:    v_mov_b32_e32 v19, s19
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:48 sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mov_b32_e32 v16, s12
+; SDAG-NEXT:    v_mov_b32_e32 v17, s13
+; SDAG-NEXT:    v_mov_b32_e32 v18, s14
+; SDAG-NEXT:    v_mov_b32_e32 v19, s15
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:32 sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mov_b32_e32 v16, s8
+; SDAG-NEXT:    v_mov_b32_e32 v17, s9
+; SDAG-NEXT:    v_mov_b32_e32 v18, s10
+; SDAG-NEXT:    v_mov_b32_e32 v19, s11
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:16 sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mov_b32_e32 v16, s4
+; SDAG-NEXT:    v_mov_b32_e32 v17, s5
+; SDAG-NEXT:    v_mov_b32_e32 v18, s6
+; SDAG-NEXT:    v_mov_b32_e32 v19, s7
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v40, v[8:11], s[0:1] offset:32 sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v40, v[12:15], s[0:1] offset:48 sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v40, v[0:3], s[0:1] sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v40, v[4:7], s[0:1] offset:16 sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_endpgm
+;
+; GISEL-LABEL: test_mfma_i32_32x32x32_i8__vgprcd:
+; 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_load_dwordx2 s[2:3], s[0:1], 0xa4
+; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-NEXT:    v_mov_b64_e32 v[34:35], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[38:39], s[26:27]
+; GISEL-NEXT:    v_mov_b64_e32 v[30:31], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[32:33], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[36:37], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[28:29], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[26:27], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[24:25], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[6:7]
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[4:5]
+; GISEL-NEXT:    s_nop 1
+; GISEL-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[32:35], v[36:39], v[16:31]
+; GISEL-NEXT:    s_nop 7
+; GISEL-NEXT:    s_nop 6
+; GISEL-NEXT:    v_mov_b32_e32 v20, 0
+; GISEL-NEXT:    global_store_dwordx4 v20, v[16:19], s[2:3] sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
+; GISEL-NEXT:    global_store_dwordx4 v20, v[16:19], s[2:3] offset:16 sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[12:13]
+; GISEL-NEXT:    global_store_dwordx4 v20, v[16:19], s[2:3] offset:32 sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[18:19]
+; GISEL-NEXT:    global_store_dwordx4 v20, v[16:19], s[2:3] offset:48 sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v20, v[0:3], s[2:3] sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v20, v[4:7], s[2:3] offset:16 sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v20, v[8:11], s[2:3] offset:32 sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v20, v[12:15], s[2:3] offset:48 sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_endpgm
+  %result = call <16 x i32> @llvm.amdgcn.mfma.i32.32x32x32.i8(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, i32 0, i32 0, i32 0)
+  store volatile <16 x i32> %arg2, ptr addrspace(1) %out
+  store volatile <16 x i32> %result, ptr addrspace(1) %out
+  ret void
+}
+
+define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd__flags(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, ptr addrspace(1) %out) #0 {
+; SDAG-LABEL: test_mfma_i32_32x32x32_i8__vgprcd__flags:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_load_dwordx8 s[16:23], s[0:1], 0x24
+; SDAG-NEXT:    v_mov_b32_e32 v40, 0
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b32_e32 v32, s16
+; SDAG-NEXT:    v_mov_b32_e32 v33, s17
+; SDAG-NEXT:    v_mov_b32_e32 v34, s18
+; SDAG-NEXT:    v_mov_b32_e32 v35, s19
+; SDAG-NEXT:    s_load_dwordx16 s[4:19], s[0:1], 0x64
+; SDAG-NEXT:    v_mov_b32_e32 v36, s20
+; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xa4
+; SDAG-NEXT:    v_mov_b32_e32 v37, s21
+; SDAG-NEXT:    v_mov_b32_e32 v38, s22
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b64_e32 v[30:31], s[18:19]
+; SDAG-NEXT:    v_mov_b32_e32 v39, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[28:29], s[16:17]
+; SDAG-NEXT:    v_mov_b64_e32 v[26:27], s[14:15]
+; SDAG-NEXT:    v_mov_b64_e32 v[24:25], s[12:13]
+; SDAG-NEXT:    v_mov_b64_e32 v[22:23], s[10:11]
+; SDAG-NEXT:    v_mov_b64_e32 v[20:21], s[8:9]
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[6:7]
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[4:5]
+; SDAG-NEXT:    s_nop 1
+; SDAG-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[32:35], v[36:39], v[16:31] cbsz:1 abid:2 blgp:3
+; SDAG-NEXT:    s_nop 7
+; SDAG-NEXT:    s_nop 6
+; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v17, s17
+; SDAG-NEXT:    v_mov_b32_e32 v18, s18
+; SDAG-NEXT:    v_mov_b32_e32 v19, s19
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:48 sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mov_b32_e32 v16, s12
+; SDAG-NEXT:    v_mov_b32_e32 v17, s13
+; SDAG-NEXT:    v_mov_b32_e32 v18, s14
+; SDAG-NEXT:    v_mov_b32_e32 v19, s15
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:32 sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mov_b32_e32 v16, s8
+; SDAG-NEXT:    v_mov_b32_e32 v17, s9
+; SDAG-NEXT:    v_mov_b32_e32 v18, s10
+; SDAG-NEXT:    v_mov_b32_e32 v19, s11
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:16 sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mov_b32_e32 v16, s4
+; SDAG-NEXT:    v_mov_b32_e32 v17, s5
+; SDAG-NEXT:    v_mov_b32_e32 v18, s6
+; SDAG-NEXT:    v_mov_b32_e32 v19, s7
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v40, v[8:11], s[0:1] offset:32 sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v40, v[12:15], s[0:1] offset:48 sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v40, v[0:3], s[0:1] sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    global_store_dwordx4 v40, v[4:7], s[0:1] offset:16 sc0 sc1
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_endpgm
+;
+; GISEL-LABEL: test_mfma_i32_32x32x32_i8__vgprcd__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_load_dwordx2 s[2:3], s[0:1], 0xa4
+; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-NEXT:    v_mov_b64_e32 v[34:35], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[38:39], s[26:27]
+; GISEL-NEXT:    v_mov_b64_e32 v[30:31], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[32:33], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[36:37], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[28:29], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[26:27], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[24:25], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[6:7]
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[4:5]
+; GISEL-NEXT:    s_nop 1
+; GISEL-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[32:35], v[36:39], v[16:31] cbsz:1 abid:2 blgp:3
+; GISEL-NEXT:    s_nop 7
+; GISEL-NEXT:    s_nop 6
+; GISEL-NEXT:    v_mov_b32_e32 v20, 0
+; GISEL-NEXT:    global_store_dwordx4 v20, v[16:19], s[2:3] sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
+; GISEL-NEXT:    global_store_dwordx4 v20, v[16:19], s[2:3] offset:16 sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[12:13]
+; GISEL-NEXT:    global_store_dwordx4 v20, v[16:19], s[2:3] offset:32 sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[18:19]
+; GISEL-NEXT:    global_store_dwordx4 v20, v[16:19], s[2:3] offset:48 sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v20, v[0:3], s[2:3] sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v20, v[4:7], s[2:3] offset:16 sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v20, v[8:11], s[2:3] offset:32 sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    global_store_dwordx4 v20, v[12:15], s[2:3] offset:48 sc0 sc1
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_endpgm
+  %result = call <16 x i32> @llvm.amdgcn.mfma.i32.32x32x32.i8(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, i32 1, i32 2, i32 3)
+  store volatile <16 x i32> %arg2, ptr addrspace(1) %out
+  store volatile <16 x i32> %result, ptr addrspace(1) %out
+  ret void
+}
+
+define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd_mac(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, ptr addrspace(1) %out) #0 {
+; SDAG-LABEL: test_mfma_i32_32x32x32_i8__vgprcd_mac:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_load_dwordx8 s[16:23], s[0:1], 0x24
+; SDAG-NEXT:    v_mov_b32_e32 v24, 0
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v17, s17
+; SDAG-NEXT:    v_mov_b32_e32 v18, s18
+; SDAG-NEXT:    v_mov_b32_e32 v19, s19
+; SDAG-NEXT:    s_load_dwordx16 s[4:19], s[0:1], 0x64
+; SDAG-NEXT:    v_mov_b32_e32 v20, s20
+; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xa4
+; SDAG-NEXT:    v_mov_b32_e32 v21, s21
+; SDAG-NEXT:    v_mov_b32_e32 v22, s22
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[4:5]
+; SDAG-NEXT:    v_mov_b32_e32 v23, s23
+; 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_nop 1
+; SDAG-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[16:19], v[20:23], v[0:15]
+; SDAG-NEXT:    s_nop 7
+; SDAG-NEXT:    s_nop 7
+; SDAG-NEXT:    s_nop 2
+; SDAG-NEXT:    global_store_dwordx4 v24, v[12:15], s[0:1] offset:48
+; SDAG-NEXT:    global_store_dwordx4 v24, v[8:11], s[0:1] offset:32
+; SDAG-NEXT:    global_store_dwordx4 v24, v[4:7], s[0:1] offset:16
+; SDAG-NEXT:    global_store_dwordx4 v24, v[0:3], s[0:1]
+; SDAG-NEXT:    s_endpgm
+;
+; GISEL-LABEL: test_mfma_i32_32x32x32_i8__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_load_dwordx2 s[2:3], s[0:1], 0xa4
+; 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[20:21], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[4:5]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[22:23]
+; 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_nop 1
+; GISEL-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[16:19], v[20:23], v[0:15]
+; GISEL-NEXT:    v_mov_b32_e32 v16, 0
+; GISEL-NEXT:    s_nop 7
+; GISEL-NEXT:    s_nop 7
+; GISEL-NEXT:    s_nop 1
+; GISEL-NEXT:    global_store_dwordx4 v16, v[0:3], s[2:3]
+; GISEL-NEXT:    global_store_dwordx4 v16, v[4:7], s[2:3] offset:16
+; GISEL-NEXT:    global_store_dwordx4 v16, v[8:11], s[2:3] offset:32
+; GISEL-NEXT:    global_store_dwordx4 v16, v[12:15], s[2:3] offset:48
+; GISEL-NEXT:    s_endpgm
+  %result = call <16 x i32> @llvm.amdgcn.mfma.i32.32x32x32.i8(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, i32 0, i32 0, i32 0)
+  store <16 x i32> %result, ptr addrspace(1) %out
+  ret void
+}
+
+define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd_mac_flags(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, ptr addrspace(1) %out) #0 {
+; SDAG-LABEL: test_mfma_i32_32x32x32_i8__vgprcd_mac_flags:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_load_dwordx8 s[16:23], s[0:1], 0x24
+; SDAG-NEXT:    v_mov_b32_e32 v24, 0
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v17, s17
+; SDAG-NEXT:    v_mov_b32_e32 v18, s18
+; SDAG-NEXT:    v_mov_b32_e32 v19, s19
+; SDAG-NEXT:    s_load_dwordx16 s[4:19], s[0:1], 0x64
+; SDAG-NEXT:    v_mov_b32_e32 v20, s20
+; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xa4
+; SDAG-NEXT:    v_mov_b32_e32 v21, s21
+; SDAG-NEXT:    v_mov_b32_e32 v22, s22
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[4:5]
+; SDAG-NEXT:    v_mov_b32_e32 v23, s23
+; 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_nop 1
+; SDAG-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[16:19], v[20:23], v[0:15] cbsz:3 abid:2 blgp:1
+; SDAG-NEXT:    s_nop 7
+; SDAG-NEXT:    s_nop 7
+; SDAG-NEXT:    s_nop 2
+; SDAG-NEXT:    global_store_dwordx4 v24, v[12:15], s[0:1] offset:48
+; SDAG-NEXT:    global_store_dwordx4 v24, v[8:11], s[0:1] offset:32
+; SDAG-NEXT:    global_store_dwordx4 v24, v[4:7], s[0:1] offset:16
+; SDAG-NEXT:    global_store_dwordx4 v24, v[0:3], s[0:1]
+; SDAG-NEXT:    s_endpgm
+;
+; GISEL-LABEL: test_mfma_i32_32x32x32_i8__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_load_dwordx2 s[2:3], s[0:1], 0xa4
+; 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[20:21], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[4:5]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[22:23]
+; 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_nop 1
+; GISEL-NEXT:    v_mfma_i32_32x32x32_i8 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_nop 7
+; GISEL-NEXT:    s_nop 7
+; GISEL-NEXT:    s_nop 1
+; GISEL-NEXT:    global_store_dwordx4 v16, v[0:3], s[2:3]
+; GISEL-NEXT:    global_store_dwordx4 v16, v[4:7], s[2:3] offset:16
+; GISEL-NEXT:    global_store_dwordx4 v16, v[8:11], s[2:3] offset:32
+; GISEL-NEXT:    global_store_dwordx4 v16, v[12:15], s[2:3] offset:48
+; GISEL-NEXT:    s_endpgm
+  %result = call <16 x i32> @llvm.amdgcn.mfma.i32.32x32x32.i8(<4 x i32> %arg0, <4 x i32> %arg1, <16 x i32> %arg2, i32 3, i32 2, i32 1)
+  store <16 x i32> %result, ptr addrspace(1) %out
+  ret void
+}
+
 attributes #0 = { "amdgpu-flat-work-group-size"="512,512" }
 attributes #1 = { "amdgpu-flat-work-group-size"="1,64" }
diff --git a/llvm/test/MC/AMDGPU/mai-gfx950.s b/llvm/test/MC/AMDGPU/mai-gfx950.s
index 3e9d3ef7fa1648..a2ea2ff6f68d0d 100644
--- a/llvm/test/MC/AMDGPU/mai-gfx950.s
+++ b/llvm/test/MC/AMDGPU/mai-gfx950.s
@@ -946,3 +946,51 @@ v_mfma_i32_16x16x64_i8 a[0:3], v[0:3], v[0:3], a[4:7]
 // GFX950: v_mfma_i32_16x16x64_i8 v[0:3], a[0:3], a[0:3], v[4:7] ; encoding: [0x00,0x00,0xb6,0xd3,0x00,0x01,0x12,0x1c]
 // ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
 v_mfma_i32_16x16x64_i8 v[0:3], a[0:3], a[0:3], v[4:7]
+
+//===----------------------------------------------------------------------===//
+// v_mfma_i32_32x32x32_i8
+//===----------------------------------------------------------------------===//
+
+// GFX950: v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15] ; encoding: [0x00,0x00,0xb8,0xd3,0x00,0x01,0x02,0x04]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15]
+
+// GFX950: v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], a[0:15] ; encoding: [0x00,0x80,0xb8,0xd3,0x00,0x01,0x02,0x1c]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], a[0:15]
+
+// GFX950:  v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15] ; encoding: [0x00,0x00,0xb8,0xd3,0x00,0x01,0x02,0x04]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_mfma_i32_32x32x32i8 v[0:15], v[0:3], v[0:3], v[0:15]
+
+// GFX950: v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], a[0:15] ; encoding: [0x00,0x80,0xb8,0xd3,0x00,0x01,0x02,0x1c]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_mfma_i32_32x32x32i8 a[0:15], a[0:3], a[0:3], a[0:15]
+
+// GFX950: v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], 1.0 ; encoding: [0x00,0x00,0xb8,0xd3,0x00,0x01,0xca,0x03]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], 1.0
+
+// GFX950: v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], 1.0 ; encoding: [0x00,0x80,0xb8,0xd3,0x00,0x01,0xca,0x1b]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], 1.0
+
+// GFX950:  v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15] blgp:5 ; encoding: [0x00,0x00,0xb8,0xd3,0x00,0x01,0x02,0xa4]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15] blgp:5
+
+// GFX950: v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], a[0:15] blgp:2 ; encoding: [0x00,0x80,0xb8,0xd3,0x00,0x01,0x02,0x5c]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], a[0:15] blgp:2
+
+// GFX950: v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15] cbsz:3 ; encoding: [0x00,0x03,0xb8,0xd3,0x00,0x01,0x02,0x04]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15] cbsz:3
+
+// GFX950: v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15] abid:1 ; encoding: [0x00,0x08,0xb8,0xd3,0x00,0x01,0x02,0x04]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15] abid:1
+
+// GFX950: v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], a[0:15] cbsz:3 abid:1 ; encoding: [0x00,0x8b,0xb8,0xd3,0x00,0x01,0x02,0x1c]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], a[0:15] cbsz:3 abid:1
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx950_mai.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx950_mai.txt
index 715a71919181f5..88bf255a0637aa 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx950_mai.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx950_mai.txt
@@ -602,3 +602,31 @@
 
 # GFX950:   v_mfma_i32_16x16x64_i8 v[0:3], v[0:3], v[0:3], v[0:3] blgp:5 ; encoding: [0x00,0x00,0xb6,0xd3,0x00,0x01,0x02,0xa4]
 0x00,0x00,0xb6,0xd3,0x00,0x01,0x02,0xa4
+
+
+# GFX950:   v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], 1.0 ; encoding: [0x00,0x80,0xb8,0xd3,0x00,0x01,0xca,0x1b]
+0x00,0x80,0xb8,0xd3,0x00,0x01,0xca,0x1b
+
+# GFX950:   v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], a[0:15] ; encoding: [0x00,0x80,0xb8,0xd3,0x00,0x01,0x02,0x1c]
+0x00,0x80,0xb8,0xd3,0x00,0x01,0x02,0x1c
+
+# GFX950:   v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], a[0:15] blgp:2 ; encoding: [0x00,0x80,0xb8,0xd3,0x00,0x01,0x02,0x5c]
+0x00,0x80,0xb8,0xd3,0x00,0x01,0x02,0x5c
+
+# GFX950:   v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], a[0:15] cbsz:3 abid:1 ; encoding: [0x00,0x8b,0xb8,0xd3,0x00,0x01,0x02,0x1c]
+0x00,0x8b,0xb8,0xd3,0x00,0x01,0x02,0x1c
+
+# GFX950:   v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], 1.0 ; encoding: [0x00,0x00,0xb8,0xd3,0x00,0x01,0xca,0x03]
+0x00,0x00,0xb8,0xd3,0x00,0x01,0xca,0x03
+
+# GFX950:   v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15] ; encoding: [0x00,0x00,0xb8,0xd3,0x00,0x01,0x02,0x04]
+0x00,0x00,0xb8,0xd3,0x00,0x01,0x02,0x04
+
+# GFX950:   v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15] abid:1 ; encoding: [0x00,0x08,0xb8,0xd3,0x00,0x01,0x02,0x04]
+0x00,0x08,0xb8,0xd3,0x00,0x01,0x02,0x04
+
+# GFX950:   v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15] blgp:5 ; encoding: [0x00,0x00,0xb8,0xd3,0x00,0x01,0x02,0xa4]
+0x00,0x00,0xb8,0xd3,0x00,0x01,0x02,0xa4
+
+# GFX950:   v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15] cbsz:3 ; encoding: [0x00,0x03,0xb8,0xd3,0x00,0x01,0x02,0x04]
+0x00,0x03,0xb8,0xd3,0x00,0x01,0x02,0x04
diff --git a/llvm/test/tools/llvm-mca/AMDGPU/gfx950.s b/llvm/test/tools/llvm-mca/AMDGPU/gfx950.s
index 5d5bb4eb55060e..760a20731e2166 100644
--- a/llvm/test/tools/llvm-mca/AMDGPU/gfx950.s
+++ b/llvm/test/tools/llvm-mca/AMDGPU/gfx950.s
@@ -1,9 +1,9 @@
 # RUN: llvm-mca -mtriple=amdgcn -mcpu=gfx950 --timeline --iterations=1 --timeline-max-cycles=0 < %s | FileCheck %s
 
 # CHECK: Iterations:        1
-# CHECK: Instructions:      11
-# CHECK: Total Cycles:      46
-# CHECK: Total uOps:        11
+# CHECK: Instructions:      13
+# CHECK: Total Cycles:      48
+# CHECK: Total uOps:        13
 
 v_mfma_ld_scale_b32 v0, v0
 v_mfma_f32_16x16x128_f8f6f4 v[0:3], v[4:11], v[4:11], v[0:3]
@@ -18,6 +18,8 @@ v_mfma_f32_32x32x16_bf16 a[0:15], a[0:3], a[0:3], a[0:15] blgp:2
 
 v_mfma_i32_16x16x64_i8 a[0:3], a[0:3], a[0:3], a[0:3] blgp:1
 v_mfma_i32_16x16x64_i8 a[0:3], v[0:3], v[0:3], a[4:7]
+v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15]
+v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], a[0:15] blgp:2
 
 
 # CHECK:     [0]    [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
@@ -32,3 +34,5 @@ v_mfma_i32_16x16x64_i8 a[0:3], v[0:3], v[0:3], a[4:7]
 # CHECK-NEXT: -      -      -      -      -      -     8.00  v_mfma_f32_32x32x16_bf16 a[0:15], a[0:3], a[0:3], a[0:15] blgp:2
 # CHECK-NEXT: -      -      -      -     1.00    -      -    v_mfma_i32_16x16x64_i8 a[0:3], a[0:3], a[0:3], a[0:3] blgp:1
 # CHECK-NEXT: -      -      -      -     1.00    -      -    v_mfma_i32_16x16x64_i8 a[0:3], v[0:3], v[0:3], a[4:7]
+# CHECK-NEXT: -      -      -      -     1.00    -      -    v_mfma_i32_32x32x32_i8 v[0:15], v[0:3], v[0:3], v[0:15]
+# CHECK-NEXT: -      -      -      -     1.00    -      -    v_mfma_i32_32x32x32_i8 a[0:15], a[0:3], a[0:3], a[0:15] blgp:2



More information about the llvm-branch-commits mailing list