Magic-Mirror/llvm-capstone
1
0
Fork 0
mirror of https://github.com/capstone-engine/llvm-capstone.git synced 2024-11-23 05:40:09 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[AMDGPU] Add cross-project-tests for WMMA builtins

Browse Source 
Add a few tests to make sure we get the expected instruction for the
WMMA builtins (and generally that our builtins and intrinsics are on the
same page and won't blow up).

Differential Revision: https://reviews.llvm.org/D144176
This commit is contained in:
Diana Picus 2023-02-16 11:10:54 +01:00
parent a048d7394b
commit 58dada5f0a
5 changed files with 151 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
cross-project-tests/CMakeLists.txt
View File

@ -86,6 +86,13 @@ add_lit_testsuite(check-intrinsic-headers "Running intrinsic header tests"
DEPENDS ${CROSS_PROJECT_TEST_DEPS}
)
# AMDGPU tests.
add_lit_testsuite(check-cross-amdgpu "Running AMDGPU cross-project tests"
${CMAKE_CURRENT_BINARY_DIR}/amdgpu
EXCLUDE_FROM_CHECK_ALL
DEPENDS clang
)
# Add check-cross-project-* targets.
add_lit_testsuites(CROSS_PROJECT ${CMAKE_CURRENT_SOURCE_DIR}
DEPENDS ${CROSS_PROJECT_TEST_DEPS}
@ -94,3 +101,4 @@ add_lit_testsuites(CROSS_PROJECT ${CMAKE_CURRENT_SOURCE_DIR}
set_target_properties(check-cross-project PROPERTIES FOLDER "Tests")
set_target_properties(check-debuginfo PROPERTIES FOLDER "Tests")
set_target_properties(check-intrinsic-headers PROPERTIES FOLDER "Tests")
set_target_properties(check-cross-amdgpu PROPERTIES FOLDER "Tests")

68
cross-project-tests/amdgpu/builtins-amdgcn-wmma-w32.cl Normal file
View File

@ -0,0 +1,68 @@
// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu gfx1100 -target-feature +wavefrontsize32 -DWMMA_GFX1100_TESTS -S -o - %s | FileCheck %s --check-prefix=CHECK-GFX1100
typedef float v4f __attribute__((ext_vector_type(4)));
typedef float v8f __attribute__((ext_vector_type(8)));
typedef half v16h __attribute__((ext_vector_type(16)));
typedef int v2i __attribute__((ext_vector_type(2)));
typedef int v4i __attribute__((ext_vector_type(4)));
typedef int v8i __attribute__((ext_vector_type(8)));
typedef short v16s __attribute__((ext_vector_type(16)));
#ifdef WMMA_GFX1100_TESTS
// Wave32
// CHECK-GFX1100-LABEL: test_amdgcn_wmma_f32_16x16x16_f16_w32:
// CHECK-GFX1100: v_wmma_f32_16x16x16_f16 v[{{.*}}], v[{{.*}} v[{{.*}}], v[{{.*}}]
//
void test_amdgcn_wmma_f32_16x16x16_f16_w32(global v8f* out, v16h a, v16h b, v8f c)
{
*out = __builtin_amdgcn_wmma_f32_16x16x16_f16_w32(a, b, c);
}
// CHECK-GFX1100-LABEL: test_amdgcn_wmma_f32_16x16x16_bf16_w32:
// CHECK-GFX1100: v_wmma_f32_16x16x16_bf16 v[{{.*}}], v[{{.*}} v[{{.*}}], v[{{.*}}]
//
void test_amdgcn_wmma_f32_16x16x16_bf16_w32(global v8f* out, v16s a, v16s b, v8f c)
{
*out = __builtin_amdgcn_wmma_f32_16x16x16_bf16_w32(a, b, c);
}
// CHECK-GFX1100-LABEL: test_amdgcn_wmma_f16_16x16x16_f16_w32:
// CHECK-GFX1100: v_wmma_f16_16x16x16_f16 v[{{.*}}], v[{{.*}} v[{{.*}}], v[{{.*}}] op_sel:[0,0,1]
//
void test_amdgcn_wmma_f16_16x16x16_f16_w32(global v16h* out, v16h a, v16h b, v16h c)
{
*out = __builtin_amdgcn_wmma_f16_16x16x16_f16_w32(a, b, c, true);
}
// CHECK-GFX1100-LABEL: test_amdgcn_wmma_bf16_16x16x16_bf16_w32:
// CHECK-GFX1100: v_wmma_bf16_16x16x16_bf16 v[{{.*}}], v[{{.*}} v[{{.*}}], v[{{.*}}] op_sel:[0,0,1]
//
void test_amdgcn_wmma_bf16_16x16x16_bf16_w32(global v16s* out, v16s a, v16s b, v16s c)
{
*out = __builtin_amdgcn_wmma_bf16_16x16x16_bf16_w32(a, b, c, true);
}
// CHECK-GFX1100-LABEL: test_amdgcn_wmma_i32_16x16x16_iu8_w32:
// CHECK-GFX1100: v_wmma_i32_16x16x16_iu8 v[{{.*}}], v[{{.*}} v[{{.*}} v[{{.*}}] neg_lo:[1,1,0]
//
void test_amdgcn_wmma_i32_16x16x16_iu8_w32(global v8i* out, v4i a, v4i b, v8i c)
{
*out = __builtin_amdgcn_wmma_i32_16x16x16_iu8_w32(true, a, true, b, c, false);
}
// CHECK-GFX1100-LABEL: test_amdgcn_wmma_i32_16x16x16_iu4_w32:
// CHECK-GFX1100: v_wmma_i32_16x16x16_iu4 v[{{.*}}, v[{{.*}} v[{{.*}} v[{{.*}} neg_lo:[1,1,0]
void test_amdgcn_wmma_i32_16x16x16_iu4_w32(global v8i* out, v2i a, v2i b, v8i c)
{
*out = __builtin_amdgcn_wmma_i32_16x16x16_iu4_w32(true, a, true, b, c, false);
}
#endif

72
cross-project-tests/amdgpu/builtins-amdgcn-wmma-w64.cl Normal file
View File

@ -0,0 +1,72 @@
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu gfx1100 -target-feature +wavefrontsize64 -DWMMA_GFX1100_TESTS -S -o - %s | FileCheck %s --check-prefix=CHECK-GFX1100
typedef float v4f __attribute__((ext_vector_type(4)));
typedef float v8f __attribute__((ext_vector_type(8)));
typedef half v8h __attribute__((ext_vector_type(8)));
typedef half v16h __attribute__((ext_vector_type(16)));
typedef int v2i __attribute__((ext_vector_type(2)));
typedef int v4i __attribute__((ext_vector_type(4)));
typedef int v8i __attribute__((ext_vector_type(8)));
typedef short v8s __attribute__((ext_vector_type(8)));
typedef short v16s __attribute__((ext_vector_type(16)));
#ifdef WMMA_GFX1100_TESTS
// Wave64
// CHECK-GFX1100-LABEL: test_amdgcn_wmma_f32_16x16x16_f16_w64:
// CHECK-GFX1100: v_wmma_f32_16x16x16_f16 v[{{.*}}], v[{{.*}} v[{{.*}}], v[{{.*}}]
//
void test_amdgcn_wmma_f32_16x16x16_f16_w64(global v4f* out, v16h a, v16h b, v4f c)
{
*out = __builtin_amdgcn_wmma_f32_16x16x16_f16_w64(a, b, c);
}
// CHECK-GFX1100-LABEL: test_amdgcn_wmma_f32_16x16x16_bf16_w64:
// CHECK-GFX1100: v_wmma_f32_16x16x16_bf16 v[{{.*}}], v[{{.*}} v[{{.*}}], v[{{.*}}]
//
void test_amdgcn_wmma_f32_16x16x16_bf16_w64(global v4f* out, v16s a, v16s b, v4f c)
{
*out = __builtin_amdgcn_wmma_f32_16x16x16_bf16_w64(a, b, c);
}
// CHECK-GFX1100-LABEL: test_amdgcn_wmma_f16_16x16x16_f16_w64:
// CHECK-GFX1100: v_wmma_f16_16x16x16_f16 v[{{.*}}], v[{{.*}} v[{{.*}}], v[{{.*}}] op_sel:[0,0,1]
//
void test_amdgcn_wmma_f16_16x16x16_f16_w64(global v8h* out, v16h a, v16h b, v8h c)
{
*out = __builtin_amdgcn_wmma_f16_16x16x16_f16_w64(a, b, c, true);
}
// CHECK-GFX1100-LABEL: test_amdgcn_wmma_bf16_16x16x16_bf16_w64:
// CHECK-GFX1100: v_wmma_bf16_16x16x16_bf16 v[{{.*}}], v[{{.*}} v[{{.*}}], v[{{.*}}] op_sel:[0,0,1]
//
void test_amdgcn_wmma_bf16_16x16x16_bf16_w64(global v8s* out, v16s a, v16s b, v8s c)
{
*out = __builtin_amdgcn_wmma_bf16_16x16x16_bf16_w64(a, b, c, true);
}
// CHECK-GFX1100-LABEL: test_amdgcn_wmma_i32_16x16x16_iu8_w64:
// CHECK-GFX1100: v_wmma_i32_16x16x16_iu8 v[{{.*}}], v[{{.*}} v[{{.*}} v[{{.*}}] neg_lo:[1,1,0]
//
void test_amdgcn_wmma_i32_16x16x16_iu8_w64(global v4i* out, v4i a, v4i b, v4i c)
{
*out = __builtin_amdgcn_wmma_i32_16x16x16_iu8_w64(true, a, true, b, c, false);
}
// CHECK-GFX1100-LABEL: test_amdgcn_wmma_i32_16x16x16_iu4_w64:
// CHECK-GFX1100: v_wmma_i32_16x16x16_iu4 v[{{.*}} v[{{.*}} v[{{.*}} v[{{.*}}neg_lo:[1,1,0]
//
void test_amdgcn_wmma_i32_16x16x16_iu4_w64(global v4i* out, v2i a, v2i b, v4i c)
{
*out = __builtin_amdgcn_wmma_i32_16x16x16_iu4_w64(true, a, true, b, c, false);
}
#endif

2
cross-project-tests/amdgpu/lit.local.cfg Normal file
View File

@ -0,0 +1,2 @@
if 'clang' not in config.available_features or 'AMDGPU' not in config.targets_to_build:
config.unsupported = True

2
cross-project-tests/lit.cfg.py
View File

@ -22,7 +22,7 @@ config.name = 'cross-project-tests'
config.test_format = lit.formats.ShTest(not llvm_config.use_lit_shell)
# suffixes: A list of file extensions to treat as test files.
config.suffixes = ['.c', '.cpp', '.m']
config.suffixes = ['.c', '.cl', '.cpp', '.m']
# excludes: A list of directories to exclude from the testsuite. The 'Inputs'
# subdirectories contain auxiliary inputs for various tests in their parent