builtins-amdgcn-gfx12-wmma-w64.cl | Explore in Territory

// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
// REQUIRES: amdgpu-registered-target
// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu gfx1200 -target-feature +wavefrontsize64 -emit-llvm -o - %s | FileCheck %s --check-prefix=CHECK-GFX1200

typedef float  v4f   __attribute__((ext_vector_type(4)));
typedef half   v4h   __attribute__((ext_vector_type(4)));
typedef short  v4s   __attribute__((ext_vector_type(4)));
typedef int    v4i   __attribute__((ext_vector_type(4)));

// Wave64

//
// amdgcn_wmma_f32_16x16x16_f16
//

// CHECK-GFX1200-LABEL: @test_amdgcn_wmma_f32_16x16x16_f16_w64(
// CHECK-GFX1200-NEXT:  entry:
// CHECK-GFX1200-NEXT:    [[TMP0:%.*]] = tail call <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16.v4f32.v4f16(<4 x half> [[A:%.*]], <4 x half> [[B:%.*]], <4 x float> [[C:%.*]])
// CHECK-GFX1200-NEXT:    store <4 x float> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 16, !tbaa [[TBAA4:![0-9]+]]
// CHECK-GFX1200-NEXT:    ret void
//
void test_amdgcn_wmma_f32_16x16x16_f16_w64(global v4f* out, v4h a, v4h b, v4f c)
{
  *out = __builtin_amdgcn_wmma_f32_16x16x16_f16_w64_gfx12(a, b, c);
}

//
// amdgcn_wmma_f32_16x16x16_bf16
//

// CHECK-GFX1200-LABEL: @test_amdgcn_wmma_f32_16x16x16_bf16_w64(
// CHECK-GFX1200-NEXT:  entry:
// CHECK-GFX1200-NEXT:    [[TMP0:%.*]] = tail call <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16.v4f32.v4i16(<4 x i16> [[A:%.*]], <4 x i16> [[B:%.*]], <4 x float> [[C:%.*]])
// CHECK-GFX1200-NEXT:    store <4 x float> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 16, !tbaa [[TBAA4]]
// CHECK-GFX1200-NEXT:    ret void
//
void test_amdgcn_wmma_f32_16x16x16_bf16_w64(global v4f* out, v4s a, v4s b, v4f c)
{
  *out = __builtin_amdgcn_wmma_f32_16x16x16_bf16_w64_gfx12(a, b, c);
}

//
// amdgcn_wmma_f16_16x16x16_f16
//

// CHECK-GFX1200-LABEL: @test_amdgcn_wmma_f16_16x16x16_f16_w64(
// CHECK-GFX1200-NEXT:  entry:
// CHECK-GFX1200-NEXT:    [[TMP0:%.*]] = tail call <4 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16.v4f16.v4f16(<4 x half> [[A:%.*]], <4 x half> [[B:%.*]], <4 x half> [[C:%.*]], i1 false)
// CHECK-GFX1200-NEXT:    store <4 x half> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 8, !tbaa [[TBAA4]]
// CHECK-GFX1200-NEXT:    ret void
//
void test_amdgcn_wmma_f16_16x16x16_f16_w64(global v4h* out, v4h a, v4h b, v4h c)
{
  *out = __builtin_amdgcn_wmma_f16_16x16x16_f16_w64_gfx12(a, b, c);
}

//
// amdgcn_wmma_bf16_16x16x16_bf16
//

// CHECK-GFX1200-LABEL: @test_amdgcn_wmma_bf16_16x16x16_bf16_w64(
// CHECK-GFX1200-NEXT:  entry:
// CHECK-GFX1200-NEXT:    [[TMP0:%.*]] = tail call <4 x i16> @llvm.amdgcn.wmma.bf16.16x16x16.bf16.v4i16.v4i16(<4 x i16> [[A:%.*]], <4 x i16> [[B:%.*]], <4 x i16> [[C:%.*]], i1 false)
// CHECK-GFX1200-NEXT:    store <4 x i16> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 8, !tbaa [[TBAA4]]
// CHECK-GFX1200-NEXT:    ret void
//
void test_amdgcn_wmma_bf16_16x16x16_bf16_w64(global v4s* out, v4s a, v4s b, v4s c)
{
  *out = __builtin_amdgcn_wmma_bf16_16x16x16_bf16_w64_gfx12(a, b, c);
}

//
// amdgcn_wmma_i32_16x16x16_iu8
//

// CHECK-GFX1200-LABEL: @test_amdgcn_wmma_i32_16x16x16_iu8_w64(
// CHECK-GFX1200-NEXT:  entry:
// CHECK-GFX1200-NEXT:    [[TMP0:%.*]] = tail call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8.v4i32.i32(i1 true, i32 [[A:%.*]], i1 true, i32 [[B:%.*]], <4 x i32> [[C:%.*]], i1 false)
// CHECK-GFX1200-NEXT:    store <4 x i32> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 16, !tbaa [[TBAA4]]
// CHECK-GFX1200-NEXT:    ret void
//
void test_amdgcn_wmma_i32_16x16x16_iu8_w64(global v4i* out, int a, int b, v4i c)
{
  *out = __builtin_amdgcn_wmma_i32_16x16x16_iu8_w64_gfx12(true, a, true, b, c, false);
}

//
// amdgcn_wmma_i32_16x16x16_iu4
//

// CHECK-GFX1200-LABEL: @test_amdgcn_wmma_i32_16x16x16_iu4_w64(
// CHECK-GFX1200-NEXT:  entry:
// CHECK-GFX1200-NEXT:    [[TMP0:%.*]] = tail call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4.v4i32.i32(i1 true, i32 [[A:%.*]], i1 true, i32 [[B:%.*]], <4 x i32> [[C:%.*]], i1 false)
// CHECK-GFX1200-NEXT:    store <4 x i32> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 16, !tbaa [[TBAA4]]
// CHECK-GFX1200-NEXT:    ret void
//
void test_amdgcn_wmma_i32_16x16x16_iu4_w64(global v4i* out, int a, int b, v4i c)
{
  *out = __builtin_amdgcn_wmma_i32_16x16x16_iu4_w64_gfx12(true, a, true, b, c, false);
}

// CHECK-GFX1200-LABEL: @test_amdgcn_wmma_f32_16x16x16_fp8_fp8_w32(
// CHECK-GFX1200-NEXT:  entry:
// CHECK-GFX1200-NEXT:    [[TMP0:%.*]] = tail call <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.fp8.fp8.v4f32.i32(i32 [[A:%.*]], i32 [[B:%.*]], <4 x float> [[C:%.*]])
// CHECK-GFX1200-NEXT:    store <4 x float> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 16, !tbaa [[TBAA4]]
// CHECK-GFX1200-NEXT:    ret void
//
void test_amdgcn_wmma_f32_16x16x16_fp8_fp8_w32(global v4f* out, int a, int b, v4f c)
{
  *out = __builtin_amdgcn_wmma_f32_16x16x16_fp8_fp8_w64_gfx12(a, b, c);
}

// CHECK-GFX1200-LABEL: @test_amdgcn_wmma_f32_16x16x16_fp8_bf8_w32(
// CHECK-GFX1200-NEXT:  entry:
// CHECK-GFX1200-NEXT:    [[TMP0:%.*]] = tail call <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.fp8.bf8.v4f32.i32(i32 [[A:%.*]], i32 [[B:%.*]], <4 x float> [[C:%.*]])
// CHECK-GFX1200-NEXT:    store <4 x float> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 16, !tbaa [[TBAA4]]
// CHECK-GFX1200-NEXT:    ret void
//
void test_amdgcn_wmma_f32_16x16x16_fp8_bf8_w32(global v4f* out, int a, int b, v4f c)
{
  *out = __builtin_amdgcn_wmma_f32_16x16x16_fp8_bf8_w64_gfx12(a, b, c);
}

// CHECK-GFX1200-LABEL: @test_amdgcn_wmma_f32_16x16x16_bf8_fp8_w32(
// CHECK-GFX1200-NEXT:  entry:
// CHECK-GFX1200-NEXT:    [[TMP0:%.*]] = tail call <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf8.fp8.v4f32.i32(i32 [[A:%.*]], i32 [[B:%.*]], <4 x float> [[C:%.*]])
// CHECK-GFX1200-NEXT:    store <4 x float> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 16, !tbaa [[TBAA4]]
// CHECK-GFX1200-NEXT:    ret void
//
void test_amdgcn_wmma_f32_16x16x16_bf8_fp8_w32(global v4f* out, int a, int b, v4f c)
{
  *out = __builtin_amdgcn_wmma_f32_16x16x16_bf8_fp8_w64_gfx12(a, b, c);
}

// CHECK-GFX1200-LABEL: @test_amdgcn_wmma_f32_16x16x16_bf8_bf8_w32(
// CHECK-GFX1200-NEXT:  entry:
// CHECK-GFX1200-NEXT:    [[TMP0:%.*]] = tail call <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf8.bf8.v4f32.i32(i32 [[A:%.*]], i32 [[B:%.*]], <4 x float> [[C:%.*]])
// CHECK-GFX1200-NEXT:    store <4 x float> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 16, !tbaa [[TBAA4]]
// CHECK-GFX1200-NEXT:    ret void
//
void test_amdgcn_wmma_f32_16x16x16_bf8_bf8_w32(global v4f* out, int a, int b, v4f c)
{
  *out = __builtin_amdgcn_wmma_f32_16x16x16_bf8_bf8_w64_gfx12(a, b, c);
}

// CHECK-GFX1200-LABEL: @test_amdgcn_wmma_i32_16x16x32_iu4_w32(
// CHECK-GFX1200-NEXT:  entry:
// CHECK-GFX1200-NEXT:    [[TMP0:%.*]] = tail call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x32.iu4.v4i32.i32(i1 true, i32 [[A:%.*]], i1 true, i32 [[B:%.*]], <4 x i32> [[C:%.*]], i1 false)
// CHECK-GFX1200-NEXT:    store <4 x i32> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 16, !tbaa [[TBAA4]]
// CHECK-GFX1200-NEXT:    ret void
//
void test_amdgcn_wmma_i32_16x16x32_iu4_w32(global v4i* out, int a, int b, v4i c)
{
  *out = __builtin_amdgcn_wmma_i32_16x16x32_iu4_w64_gfx12(true, a, true, b, c, false);
}
llvm/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx12-wmma-w64.cl
