[llvm-branch-commits] [clang] [llvm] AMDGPU: Add v_smfmac_f32_16x16x128_bf8_fp8 for gfx950 (PR #117233)

Matt Arsenault via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Thu Nov 21 15:03:52 PST 2024


https://github.com/arsenm updated https://github.com/llvm/llvm-project/pull/117233

>From 26c1f2d322e0ff8cbeb4d79052b26e5858a77332 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Fri, 2 Feb 2024 23:21:12 +0530
Subject: [PATCH] AMDGPU: Add v_smfmac_f32_16x16x128_bf8_fp8 for gfx950

---
 clang/include/clang/Basic/BuiltinsAMDGPU.def  |   1 +
 .../CodeGenOpenCL/builtins-amdgcn-mfma.cl     |   7 +
 .../builtins-amdgcn-error-gfx950-param.cl     |   6 +
 .../builtins-amdgcn-error-gfx950.cl           |   1 +
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td      |   1 +
 .../AMDGPU/AMDGPUInstructionSelector.cpp      |   4 +
 .../Target/AMDGPU/AMDGPURegisterBankInfo.cpp  |   3 +-
 llvm/lib/Target/AMDGPU/VOP3PInstructions.td   |   2 +
 .../UniformityAnalysis/AMDGPU/intrinsics.ll   |   9 +
 .../AMDGPU/llvm.amdgcn.smfmac.gfx950.ll       | 215 ++++++++++++++++++
 llvm/test/MC/AMDGPU/mai-gfx950.s              |  36 +++
 .../MC/Disassembler/AMDGPU/gfx950_mai.txt     |  23 ++
 12 files changed, 307 insertions(+), 1 deletion(-)

diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 5029e1a6153f9a..4c9ce45aa311a9 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -451,6 +451,7 @@ TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x32_bf16, "V16fV8yV16yV16fiIiIi"
 TARGET_BUILTIN(__builtin_amdgcn_smfmac_i32_16x16x128_i8, "V4iV4iV8iV4iiIiIi", "nc", "gfx950-insts")
 TARGET_BUILTIN(__builtin_amdgcn_smfmac_i32_32x32x64_i8, "V16iV4iV8iV16iiIiIi", "nc", "gfx950-insts")
 TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x128_bf8_bf8, "V4fV4iV8iV4fiIiIi", "nc", "gfx950-insts")
+TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x128_bf8_fp8, "V4fV4iV8iV4fiIiIi", "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 f07bec5ec91eb3..a9450e5f322241 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl
@@ -517,4 +517,11 @@ void test_smfmac_f32_16x16x128_bf8_bf8(global v4f* out, v4i a, v8i b, v4f c, int
   *out = __builtin_amdgcn_smfmac_f32_16x16x128_bf8_bf8(a, b, c, idx, 0, 0);
 }
 
+// CHECK-GFX950-LABEL: @test_smfmac_f32_16x16x128_bf8_fp8
+// CHECK-GFX950: call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x128.bf8.fp8(<4 x i32> %a, <8 x i32> %b, <4 x float> %c, i32 %idx, i32 0, i32 0)
+void test_smfmac_f32_16x16x128_bf8_fp8(global v4f* out, v4i a, v8i b, v4f c, int idx)
+{
+  *out = __builtin_amdgcn_smfmac_f32_16x16x128_bf8_fp8(a, b, c, idx, 0, 0);
+}
+
 #endif
diff --git a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl
index e2235cec42d84b..6b5ea7f7efd8d6 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl
@@ -106,3 +106,9 @@ void test_smfmac_f32_16x16x128_bf8_bf8(global float4* out, int4 a, int8 b, float
   *out = __builtin_amdgcn_smfmac_f32_16x16x128_bf8_bf8(a, b, c, idx, d, 0); // expected-error{{argument to '__builtin_amdgcn_smfmac_f32_16x16x128_bf8_bf8' must be a constant integer}}
   *out = __builtin_amdgcn_smfmac_f32_16x16x128_bf8_bf8(a, b, c, idx, 0, d); // expected-error{{argument to '__builtin_amdgcn_smfmac_f32_16x16x128_bf8_bf8' must be a constant integer}}
 }
+
+void test_smfmac_f32_16x16x128_bf8_fp8(global float4* out, int4 a, int8 b, float4 c, int idx, int d)
+{
+  *out = __builtin_amdgcn_smfmac_f32_16x16x128_bf8_fp8(a, b, c, idx, d, 0); // expected-error{{argument to '__builtin_amdgcn_smfmac_f32_16x16x128_bf8_fp8' must be a constant integer}}
+  *out = __builtin_amdgcn_smfmac_f32_16x16x128_bf8_fp8(a, b, c, idx, 0, d); // expected-error{{argument to '__builtin_amdgcn_smfmac_f32_16x16x128_bf8_fp8' 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 f65abeb2b58c4d..1ed60530fabb4d 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950.cl
@@ -41,6 +41,7 @@ void test(__global float4* out0, half8 a0, half8 b0, float4 c0,
   *out10 = __builtin_amdgcn_smfmac_i32_16x16x128_i8(a10, b10, c10, 0, 0, 0); // expected-error{{'__builtin_amdgcn_smfmac_i32_16x16x128_i8' needs target feature gfx950-insts}}
   *out11 = __builtin_amdgcn_smfmac_i32_32x32x64_i8(a11, b11, c11, 0, 0, 0); // expected-error{{'__builtin_amdgcn_smfmac_i32_32x32x64_i8' needs target feature gfx950-insts}}
   *out12 = __builtin_amdgcn_smfmac_f32_16x16x128_bf8_bf8(a12, b12, c12, 0, 0, 0); // expected-error{{'__builtin_amdgcn_smfmac_f32_16x16x128_bf8_bf8' needs target feature gfx950-insts}}
+  *out12 = __builtin_amdgcn_smfmac_f32_16x16x128_bf8_fp8(a12, b12, c12, 0, 0, 0); // expected-error{{'__builtin_amdgcn_smfmac_f32_16x16x128_bf8_fp8' 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 7ad23afb7010de..51330566a3a62d 100644
--- a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
+++ b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
@@ -3159,6 +3159,7 @@ def int_amdgcn_smfmac_f32_32x32x32_bf16 : AMDGPUMSmfmacIntrinsic<llvm_v16f32_ty,
 def int_amdgcn_smfmac_i32_16x16x128_i8 : AMDGPUMSmfmacIntrinsic<llvm_v4i32_ty,  llvm_v4i32_ty, llvm_v8i32_ty>;
 def int_amdgcn_smfmac_i32_32x32x64_i8 : AMDGPUMSmfmacIntrinsic<llvm_v16i32_ty, llvm_v4i32_ty, llvm_v8i32_ty>;
 def int_amdgcn_smfmac_f32_16x16x128_bf8_bf8 : AMDGPUMSmfmacIntrinsic<llvm_v4f32_ty, llvm_v4i32_ty, llvm_v8i32_ty>;
+def int_amdgcn_smfmac_f32_16x16x128_bf8_fp8 : AMDGPUMSmfmacIntrinsic<llvm_v4f32_ty, llvm_v4i32_ty, llvm_v8i32_ty>;
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
index 909fc349eca302..f5dae490f2cb9c 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
@@ -1097,6 +1097,7 @@ bool AMDGPUInstructionSelector::selectG_INTRINSIC(MachineInstr &I) const {
   case Intrinsic::amdgcn_smfmac_i32_16x16x128_i8:
   case Intrinsic::amdgcn_smfmac_i32_32x32x64_i8:
   case Intrinsic::amdgcn_smfmac_f32_16x16x128_bf8_bf8:
+  case Intrinsic::amdgcn_smfmac_f32_16x16x128_bf8_fp8:
     return selectSMFMACIntrin(I);
   default:
     return selectImpl(I, *CoverageInfo);
@@ -3510,6 +3511,9 @@ bool AMDGPUInstructionSelector::selectSMFMACIntrin(MachineInstr &MI) const {
   case Intrinsic::amdgcn_smfmac_f32_16x16x128_bf8_bf8:
     Opc = AMDGPU::V_SMFMAC_F32_16X16X128_BF8_BF8_e64;
     break;
+  case Intrinsic::amdgcn_smfmac_f32_16x16x128_bf8_fp8:
+    Opc = AMDGPU::V_SMFMAC_F32_16X16X128_BF8_FP8_e64;
+    break;
   default:
     llvm_unreachable("unhandled smfmac intrinsic");
   }
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
index 7cf0dc17a25e7a..6a56989259571b 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -4811,7 +4811,8 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
     case Intrinsic::amdgcn_smfmac_f32_32x32x32_bf16:
     case Intrinsic::amdgcn_smfmac_i32_16x16x128_i8:
     case Intrinsic::amdgcn_smfmac_i32_32x32x64_i8:
-    case Intrinsic::amdgcn_smfmac_f32_16x16x128_bf8_bf8: {
+    case Intrinsic::amdgcn_smfmac_f32_16x16x128_bf8_bf8:
+    case Intrinsic::amdgcn_smfmac_f32_16x16x128_bf8_fp8: {
       // vdst, srcA, srcB, srcC, idx
       OpdsMapping[0] = getAGPROpMapping(MI.getOperand(0).getReg(), MRI, *TRI);
       OpdsMapping[2] = getVGPROpMapping(MI.getOperand(2).getReg(), MRI, *TRI);
diff --git a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
index 8be84c250166ba..94ce3f0938ffbc 100644
--- a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -1058,6 +1058,7 @@ defm V_SMFMAC_F32_32X32X32_BF16    : SMFMACInst<"v_smfmac_f32_32x32x32_bf16",
 defm V_SMFMAC_I32_16X16X128_I8     : SMFMACInst<"v_smfmac_i32_16x16x128_i8",     "I32_16X16X128_I8", int_amdgcn_smfmac_i32_16x16x128_i8>;
 defm V_SMFMAC_I32_32X32X64_I8      : SMFMACInst<"v_smfmac_i32_32x32x64_i8",      "I32_32X32X64_I8", int_amdgcn_smfmac_i32_32x32x64_i8>;
 defm V_SMFMAC_F32_16X16X128_BF8_BF8 : SMFMACInst<"v_smfmac_f32_16x16x128_bf8_bf8", "F32_16X16X128_F8", int_amdgcn_smfmac_f32_16x16x128_bf8_bf8>;
+defm V_SMFMAC_F32_16X16X128_BF8_FP8 : SMFMACInst<"v_smfmac_f32_16x16x128_bf8_fp8", "F32_16X16X128_F8", int_amdgcn_smfmac_f32_16x16x128_bf8_fp8>;
 }
 
 def MAIInstInfoTable : GenericTable {
@@ -2157,6 +2158,7 @@ defm V_SMFMAC_I32_16X16X128_I8     : VOP3P_Real_SMFMAC <0x3a, "v_smfmac_i32_16x1
 defm V_SMFMAC_I32_32X32X64_I8      : VOP3P_Real_SMFMAC <0x47, "v_smfmac_i32_32x32x64i8">;
 
 defm V_SMFMAC_F32_16X16X128_BF8_BF8 : VOP3P_Real_SMFMAC <0x3b, "v_smfmac_f32_16x16x128bf8bf8">;
+defm V_SMFMAC_F32_16X16X128_BF8_FP8 : VOP3P_Real_SMFMAC <0x3c, "v_smfmac_f32_16x16x128bf8fp8">;
 
 defm V_PK_FMA_F32 : VOP3P_Real_vi <0x30>;
 defm V_PK_MUL_F32 : VOP3P_Real_vi <0x31>;
diff --git a/llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll b/llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll
index bea81706b652a0..2d6ed237fe52ba 100644
--- a/llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll
+++ b/llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll
@@ -377,6 +377,15 @@ define amdgpu_kernel void @smfmac_f32_16x16x128_bf8_bf8(<4 x i32> %arg0, <8 x i3
   ret void
 }
 
+declare <4 x float> @llvm.amdgcn.smfmac.f32.16x16x128.bf8.fp8(<4 x i32>, <8 x i32>, <4 x float>, i32, i32, i32)
+
+; CHECK: DIVERGENT:  %result = call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x128.bf8.fp8(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3, i32 1, i32 2)
+define amdgpu_kernel void @smfmac_f32_16x16x128_bf8_fp8(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3, ptr addrspace(1) %out) {
+  %result = call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x128.bf8.fp8(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3, i32 1, i32 2)
+  store <4 x float> %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.smfmac.gfx950.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll
index 98b65803887129..9ec0a51417f3ce 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll
@@ -2325,6 +2325,221 @@ define <4 x float> @test_smfmac_f32_16x16x128_bf8_bf8__sgpr(<4 x i32> inreg %arg
   ret <4 x float> %result
 }
 
+; --------------------------------------------------------------------
+; llvm.amdgcn.smfmac.f32.16x16x128.bf8.fp8
+; --------------------------------------------------------------------
+
+declare <4 x float> @llvm.amdgcn.smfmac.f32.16x16x128.bf8.fp8(<4 x i32>, <8 x i32>, <4 x float>, i32, i32 immarg, i32 immarg)
+
+define amdgpu_kernel void @test_smfmac_f32_16x16x128_bf8_fp8__vgpr(ptr addrspace(1) %arg, <4 x i32> %a, <8 x i32> %b, i32 %idx) #0 {
+; SDAG-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__vgpr:
+; SDAG:       ; %bb.0: ; %bb
+; SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
+; SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
+; SDAG-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
+; SDAG-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    global_load_dwordx4 v[8:11], v0, s[6:7]
+; SDAG-NEXT:    s_load_dword s16, s[4:5], 0x64
+; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
+; SDAG-NEXT:    v_mov_b32_e32 v12, s8
+; SDAG-NEXT:    v_mov_b32_e32 v13, s9
+; SDAG-NEXT:    v_mov_b32_e32 v14, s10
+; SDAG-NEXT:    v_mov_b32_e32 v15, s11
+; 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_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b32_e32 v4, s0
+; SDAG-NEXT:    v_mov_b32_e32 v5, s1
+; SDAG-NEXT:    v_mov_b32_e32 v6, s2
+; SDAG-NEXT:    v_mov_b32_e32 v7, s3
+; SDAG-NEXT:    v_mov_b32_e32 v17, s16
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_smfmac_f32_16x16x128_bf8_fp8 v[8:11], v[12:15], v[0:7], v17 cbsz:1 abid:2
+; SDAG-NEXT:    s_nop 6
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[6:7]
+; SDAG-NEXT:    s_endpgm
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__vgpr:
+; GISEL:       ; %bb.0: ; %bb
+; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
+; GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
+; GISEL-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
+; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-NEXT:    global_load_dwordx4 v[8:11], v0, s[0:1]
+; GISEL-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
+; GISEL-NEXT:    s_load_dwordx4 s[16:19], s[4:5], 0x54
+; GISEL-NEXT:    s_load_dword s2, s[4:5], 0x64
+; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[8:9]
+; 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:    v_mov_b64_e32 v[4:5], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[18:19]
+; GISEL-NEXT:    v_mov_b32_e32 v16, s2
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_smfmac_f32_16x16x128_bf8_fp8 v[8:11], v[12:15], v[0:7], v16 cbsz:1 abid:2
+; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    s_nop 5
+; GISEL-NEXT:    global_store_dwordx4 v0, v[8:11], s[0:1]
+; GISEL-NEXT:    s_endpgm
+bb:
+  %id = call i32 @llvm.amdgcn.workitem.id.x()
+  %gep = getelementptr <4 x float>, ptr addrspace(1) %arg, i32 %id
+  %in.1 = load <4 x float>, ptr addrspace(1) %gep
+  %mai.1 = tail call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x128.bf8.fp8(<4 x i32> %a, <8 x i32> %b, <4 x float> %in.1, i32 %idx, i32 1, i32 2)
+  store <4 x float> %mai.1, ptr addrspace(1) %arg
+  ret void
+}
+
+define <4 x float> @test_smfmac_f32_16x16x128_bf8_fp8(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) {
+; SDAG-LABEL: test_smfmac_f32_16x16x128_bf8_fp8:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT:    v_accvgpr_write_b32 a0, v12
+; SDAG-NEXT:    v_accvgpr_write_b32 a1, v13
+; SDAG-NEXT:    v_accvgpr_write_b32 a2, v14
+; SDAG-NEXT:    v_accvgpr_write_b32 a3, v15
+; SDAG-NEXT:    s_nop 1
+; SDAG-NEXT:    v_smfmac_f32_16x16x128_bf8_fp8 a[0:3], v[0:3], v[4:11], v16
+; SDAG-NEXT:    s_nop 6
+; SDAG-NEXT:    v_accvgpr_read_b32 v0, a0
+; SDAG-NEXT:    v_accvgpr_read_b32 v1, a1
+; SDAG-NEXT:    v_accvgpr_read_b32 v2, a2
+; SDAG-NEXT:    v_accvgpr_read_b32 v3, a3
+; SDAG-NEXT:    s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_fp8:
+; GISEL:       ; %bb.0:
+; GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT:    v_smfmac_f32_16x16x128_bf8_fp8 v[12:15], v[0:3], v[4:11], v16
+; GISEL-NEXT:    s_nop 6
+; GISEL-NEXT:    v_mov_b32_e32 v0, v12
+; GISEL-NEXT:    v_mov_b32_e32 v1, v13
+; GISEL-NEXT:    v_mov_b32_e32 v2, v14
+; GISEL-NEXT:    v_mov_b32_e32 v3, v15
+; GISEL-NEXT:    s_setpc_b64 s[30:31]
+  %result = call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x128.bf8.fp8(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3, i32 immarg 0, i32 immarg 0)
+  ret <4 x float> %result
+}
+
+define <4 x float> @test_smfmac_f32_16x16x128_bf8_fp8__flags0(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) {
+; SDAG-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__flags0:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT:    v_accvgpr_write_b32 a0, v12
+; SDAG-NEXT:    v_accvgpr_write_b32 a1, v13
+; SDAG-NEXT:    v_accvgpr_write_b32 a2, v14
+; SDAG-NEXT:    v_accvgpr_write_b32 a3, v15
+; SDAG-NEXT:    s_nop 1
+; SDAG-NEXT:    v_smfmac_f32_16x16x128_bf8_fp8 a[0:3], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; SDAG-NEXT:    s_nop 6
+; SDAG-NEXT:    v_accvgpr_read_b32 v0, a0
+; SDAG-NEXT:    v_accvgpr_read_b32 v1, a1
+; SDAG-NEXT:    v_accvgpr_read_b32 v2, a2
+; SDAG-NEXT:    v_accvgpr_read_b32 v3, a3
+; SDAG-NEXT:    s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__flags0:
+; GISEL:       ; %bb.0:
+; GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT:    v_smfmac_f32_16x16x128_bf8_fp8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; GISEL-NEXT:    s_nop 6
+; GISEL-NEXT:    v_mov_b32_e32 v0, v12
+; GISEL-NEXT:    v_mov_b32_e32 v1, v13
+; GISEL-NEXT:    v_mov_b32_e32 v2, v14
+; GISEL-NEXT:    v_mov_b32_e32 v3, v15
+; GISEL-NEXT:    s_setpc_b64 s[30:31]
+  %result = call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x128.bf8.fp8(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3, i32 immarg 1, i32 immarg 3)
+  ret <4 x float> %result
+}
+
+define <4 x float> @test_smfmac_f32_16x16x128_bf8_fp8__flags1(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) {
+; SDAG-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__flags1:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT:    v_accvgpr_write_b32 a0, v12
+; SDAG-NEXT:    v_accvgpr_write_b32 a1, v13
+; SDAG-NEXT:    v_accvgpr_write_b32 a2, v14
+; SDAG-NEXT:    v_accvgpr_write_b32 a3, v15
+; SDAG-NEXT:    s_nop 1
+; SDAG-NEXT:    v_smfmac_f32_16x16x128_bf8_fp8 a[0:3], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; SDAG-NEXT:    s_nop 6
+; SDAG-NEXT:    v_accvgpr_read_b32 v0, a0
+; SDAG-NEXT:    v_accvgpr_read_b32 v1, a1
+; SDAG-NEXT:    v_accvgpr_read_b32 v2, a2
+; SDAG-NEXT:    v_accvgpr_read_b32 v3, a3
+; SDAG-NEXT:    s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__flags1:
+; GISEL:       ; %bb.0:
+; GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT:    v_smfmac_f32_16x16x128_bf8_fp8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; GISEL-NEXT:    s_nop 6
+; GISEL-NEXT:    v_mov_b32_e32 v0, v12
+; GISEL-NEXT:    v_mov_b32_e32 v1, v13
+; GISEL-NEXT:    v_mov_b32_e32 v2, v14
+; GISEL-NEXT:    v_mov_b32_e32 v3, v15
+; GISEL-NEXT:    s_setpc_b64 s[30:31]
+  %result = call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x128.bf8.fp8(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3, i32 immarg 3, i32 immarg 1)
+  ret <4 x float> %result
+}
+
+define <4 x float> @test_smfmac_f32_16x16x128_bf8_fp8__sgpr(<4 x i32> inreg %arg0, <8 x i32> inreg %arg1, <4 x float> inreg %arg2, i32 inreg %arg3) {
+; SDAG-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__sgpr:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b32_e32 v8, s0
+; SDAG-NEXT:    v_mov_b32_e32 v9, s1
+; SDAG-NEXT:    v_mov_b32_e32 v10, s2
+; SDAG-NEXT:    v_mov_b32_e32 v11, s3
+; 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:    v_mov_b32_e32 v4, s20
+; SDAG-NEXT:    v_mov_b32_e32 v5, s21
+; SDAG-NEXT:    v_mov_b32_e32 v6, s22
+; SDAG-NEXT:    v_mov_b32_e32 v7, s23
+; SDAG-NEXT:    v_accvgpr_write_b32 a0, s24
+; SDAG-NEXT:    v_accvgpr_write_b32 a1, s25
+; SDAG-NEXT:    v_accvgpr_write_b32 a2, s26
+; SDAG-NEXT:    v_accvgpr_write_b32 a3, s27
+; SDAG-NEXT:    v_mov_b32_e32 v12, s28
+; SDAG-NEXT:    s_nop 1
+; SDAG-NEXT:    v_smfmac_f32_16x16x128_bf8_fp8 a[0:3], v[8:11], v[0:7], v12
+; SDAG-NEXT:    s_nop 6
+; SDAG-NEXT:    v_accvgpr_read_b32 v0, a0
+; SDAG-NEXT:    v_accvgpr_read_b32 v1, a1
+; SDAG-NEXT:    v_accvgpr_read_b32 v2, a2
+; SDAG-NEXT:    v_accvgpr_read_b32 v3, a3
+; SDAG-NEXT:    s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__sgpr:
+; GISEL:       ; %bb.0:
+; GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[2:3]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[0:1]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
+; GISEL-NEXT:    v_mov_b32_e32 v16, s28
+; GISEL-NEXT:    s_nop 1
+; GISEL-NEXT:    v_smfmac_f32_16x16x128_bf8_fp8 v[0:3], v[12:15], v[4:11], v16
+; GISEL-NEXT:    s_setpc_b64 s[30:31]
+  %result = call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x128.bf8.fp8(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3, i32 immarg 0, i32 immarg 0)
+  ret <4 x float> %result
+}
+
 attributes #0 = { "amdgpu-flat-work-group-size"="1,256" }
 ;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
 ; GCN: {{.*}}
diff --git a/llvm/test/MC/AMDGPU/mai-gfx950.s b/llvm/test/MC/AMDGPU/mai-gfx950.s
index cee66f5274761d..d83c7b5f62d24d 100644
--- a/llvm/test/MC/AMDGPU/mai-gfx950.s
+++ b/llvm/test/MC/AMDGPU/mai-gfx950.s
@@ -1308,3 +1308,39 @@ v_smfmac_f32_16x16x128_bf8_bf8 a[10:13], a[2:5], a[6:13], v2 cbsz:3 abid:1
 // GFX950: v_smfmac_f32_16x16x128_bf8_bf8 a[10:13], a[2:5], a[6:13], v3 cbsz:1 abid:3 ; encoding: [0x0a,0x99,0xbb,0xd3,0x02,0x0d,0x0e,0x1c]
 // ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
 v_smfmac_f32_16x16x128_bf8_bf8 a[10:13], a[2:5], a[6:13], v3 cbsz:1 abid:3
+
+//===----------------------------------------------------------------------===//
+// v_smfmac_f32_16x16x128_bf8_fp8
+//===----------------------------------------------------------------------===//
+
+// GFX950: v_smfmac_f32_16x16x128_bf8_fp8 v[10:13], a[2:5], v[4:11], v3 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xbc,0xd3,0x02,0x09,0x0e,0x0c]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x128_bf8_fp8 v[10:13], a[2:5], v[4:11], v3 cbsz:3 abid:1
+
+// GFX950: v_smfmac_f32_16x16x128_bf8_fp8 v[10:13], a[2:5], v[4:11], v3 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xbc,0xd3,0x02,0x09,0x0e,0x0c]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x128bf8fp8 v[10:13], a[2:5], v[4:11], v3 cbsz:3 abid:1
+
+// GFX950: v_smfmac_f32_16x16x128_bf8_fp8 a[10:13], v[2:5], a[4:11], v1 ; encoding: [0x0a,0x80,0xbc,0xd3,0x02,0x09,0x06,0x14]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x128_bf8_fp8 a[10:13], v[2:5], a[4:11], v1
+
+// GFX950: v_smfmac_f32_16x16x128_bf8_fp8 v[10:13], a[2:5], v[4:11], v2 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xbc,0xd3,0x02,0x09,0x0a,0x0c]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x128_bf8_fp8 v[10:13], a[2:5], v[4:11], v2 cbsz:3 abid:1
+
+// GFX950: v_smfmac_f32_16x16x128_bf8_fp8 a[10:13], v[2:5], a[4:11], v3 ; encoding: [0x0a,0x80,0xbc,0xd3,0x02,0x09,0x0e,0x14]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x128_bf8_fp8 a[10:13], v[2:5], a[4:11], v3
+
+// GFX950: v_smfmac_f32_16x16x128_bf8_fp8 v[10:13], v[2:5], v[6:13], v2 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xbc,0xd3,0x02,0x0d,0x0a,0x04]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x128_bf8_fp8 v[10:13], v[2:5], v[6:13], v2 cbsz:3 abid:1
+
+// GFX950: v_smfmac_f32_16x16x128_bf8_fp8 a[10:13], a[2:5], a[6:13], v2 cbsz:3 abid:1 ; encoding: [0x0a,0x8b,0xbc,0xd3,0x02,0x0d,0x0a,0x1c]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x128_bf8_fp8 a[10:13], a[2:5], a[6:13], v2 cbsz:3 abid:1
+
+// GFX950: v_smfmac_f32_16x16x128_bf8_fp8 a[10:13], a[2:5], a[6:13], v3 cbsz:1 abid:3 ; encoding: [0x0a,0x99,0xbc,0xd3,0x02,0x0d,0x0e,0x1c]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x128_bf8_fp8 a[10:13], a[2:5], a[6:13], v3 cbsz:1 abid:3
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx950_mai.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx950_mai.txt
index 94f1fc66ae8fc2..6d90177211d2f6 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx950_mai.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx950_mai.txt
@@ -821,3 +821,26 @@
 
 # GFX950:   v_smfmac_f32_16x16x128_bf8_bf8 v[10:13], v[2:5], v[6:13], v2 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xbb,0xd3,0x02,0x0d,0x0a,0x04]
 0x0a,0x0b,0xbb,0xd3,0x02,0x0d,0x0a,0x04
+
+
+
+# GFX950:   v_smfmac_f32_16x16x128_bf8_fp8 a[10:13], a[2:5], a[6:13], v2 cbsz:3 abid:1 ; encoding: [0x0a,0x8b,0xbc,0xd3,0x02,0x0d,0x0a,0x1c]
+0x0a,0x8b,0xbc,0xd3,0x02,0x0d,0x0a,0x1c
+
+# GFX950:   v_smfmac_f32_16x16x128_bf8_fp8 a[10:13], a[2:5], a[6:13], v3 cbsz:1 abid:3 ; encoding: [0x0a,0x99,0xbc,0xd3,0x02,0x0d,0x0e,0x1c]
+0x0a,0x99,0xbc,0xd3,0x02,0x0d,0x0e,0x1c
+
+# GFX950:   v_smfmac_f32_16x16x128_bf8_fp8 a[10:13], v[2:5], a[4:11], v1 ; encoding: [0x0a,0x80,0xbc,0xd3,0x02,0x09,0x06,0x14]
+0x0a,0x80,0xbc,0xd3,0x02,0x09,0x06,0x14
+
+# GFX950:   v_smfmac_f32_16x16x128_bf8_fp8 a[10:13], v[2:5], a[4:11], v3 ; encoding: [0x0a,0x80,0xbc,0xd3,0x02,0x09,0x0e,0x14]
+0x0a,0x80,0xbc,0xd3,0x02,0x09,0x0e,0x14
+
+# GFX950:   v_smfmac_f32_16x16x128_bf8_fp8 v[10:13], a[2:5], v[4:11], v2 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xbc,0xd3,0x02,0x09,0x0a,0x0c]
+0x0a,0x0b,0xbc,0xd3,0x02,0x09,0x0a,0x0c
+
+# GFX950:   v_smfmac_f32_16x16x128_bf8_fp8 v[10:13], a[2:5], v[4:11], v3 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xbc,0xd3,0x02,0x09,0x0e,0x0c]
+0x0a,0x0b,0xbc,0xd3,0x02,0x09,0x0e,0x0c
+
+# GFX950:   v_smfmac_f32_16x16x128_bf8_fp8 v[10:13], v[2:5], v[6:13], v2 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xbc,0xd3,0x02,0x0d,0x0a,0x04]
+0x0a,0x0b,0xbc,0xd3,0x02,0x0d,0x0a,0x04



More information about the llvm-branch-commits mailing list