Skip to content

Commit 775d068

Browse files
MrSidimsMrSidims
authored
[SYCL][Matrix] Fix __spirv_JointMatrixINTEL signature (#6957)
Default SYCL_EXT_ONEAPI_MATRIX to 1 Added SYCL_EXT_ONEAPI_MATRIX_VERSION to differentiate API versions. Only add 'Use' parameter if testing macro SYCL_EXT_ONEAPI_MATRIX_VERSION = 2 is defined. Test change: intel/llvm-test-suite#1311 Signed-off-by: Sidorov, Dmitry [email protected]
1 parent a6222ba commit 775d068

23 files changed

+88
-73
lines changed

‎sycl/include/CL/__spirv/spirv_ops.hpp‎

Lines changed: 42 additions & 29 deletions
Original file line numberDiff line numberDiff line change
@@ -22,11 +22,20 @@
2222
#endif
2323

2424
#ifdef __SYCL_DEVICE_ONLY__
25+
26+
#if (SYCL_EXT_ONEAPI_MATRIX_VERSION > 1)
27+
#define JOINT_MATRIX_INTEL(T, R, C, L, S, U) \
28+
__spv::__spirv_JointMatrixINTEL<T, R, C, L, S, U>
29+
#else
30+
#define JOINT_MATRIX_INTEL(T, R, C, L, S, U) \
31+
__spv::__spirv_JointMatrixINTEL<T, R, C, L, S>
32+
#endif // SYCL_EXT_ONEAPI_MATRIX_VERSION
33+
2534
template <typename T, std::size_t R, std::size_t C,
2635
__spv::MatrixUse U = __spv::MatrixUse::Unnecessary,
2736
__spv::MatrixLayout L = __spv::MatrixLayout::RowMajor,
2837
__spv::Scope::Flag S = __spv::Scope::Flag::Subgroup>
29-
extern SYCL_EXTERNAL __spv::__spirv_JointMatrixINTEL<T, R, C, L, S, U> *
38+
extern SYCL_EXTERNAL JOINT_MATRIX_INTEL(T, R, C, L, S, U) *
3039
__spirv_JointMatrixLoadINTEL(T *Ptr, std::size_t Stride,
3140
__spv::MatrixLayout Layout = L,
3241
__spv::Scope::Flag Sc = S, int MemOperand = 0);
@@ -36,7 +45,7 @@ template <typename T, std::size_t R, std::size_t C,
3645
__spv::MatrixLayout L = __spv::MatrixLayout::RowMajor,
3746
__spv::Scope::Flag S = __spv::Scope::Flag::Subgroup>
3847
extern SYCL_EXTERNAL void __spirv_JointMatrixStoreINTEL(
39-
T *Ptr, __spv::__spirv_JointMatrixINTEL<T, R, C, L, S, U> *Object,
48+
T *Ptr, JOINT_MATRIX_INTEL(T, R, C, L, S, U) *Object,
4049
std::size_t Stride, __spv::MatrixLayout Layout = L,
4150
__spv::Scope::Flag Sc = S, int MemOperand = 0);
4251

@@ -48,11 +57,11 @@ template <typename T1, typename T2, std::size_t M, std::size_t K, std::size_t N,
4857
__spv::MatrixLayout LB = __spv::MatrixLayout::RowMajor,
4958
__spv::MatrixLayout LC = __spv::MatrixLayout::RowMajor,
5059
__spv::Scope::Flag S = __spv::Scope::Flag::Subgroup>
51-
extern SYCL_EXTERNAL __spv::__spirv_JointMatrixINTEL<T2, M, N, LC, S, UC> *
60+
extern SYCL_EXTERNAL JOINT_MATRIX_INTEL(T2, M, N, LC, S, UC) *
5261
__spirv_JointMatrixMadINTEL(
53-
__spv::__spirv_JointMatrixINTEL<T1, M, K, LA, S, UA> *A,
54-
__spv::__spirv_JointMatrixINTEL<T1, K, N, LB, S, UB> *B,
55-
__spv::__spirv_JointMatrixINTEL<T2, M, N, LC, S, UC> *C,
62+
JOINT_MATRIX_INTEL(T1, M, K, LA, S, UA) *A,
63+
JOINT_MATRIX_INTEL(T1, K, N, LB, S, UB) *B,
64+
JOINT_MATRIX_INTEL(T2, M, N, LC, S, UC) *C,
5665
__spv::Scope::Flag Sc = __spv::Scope::Flag::Subgroup);
5766

5867
template <typename T1, typename T2, typename T3, std::size_t M, std::size_t K,
@@ -63,11 +72,11 @@ template <typename T1, typename T2, typename T3, std::size_t M, std::size_t K,
6372
__spv::MatrixLayout LB = __spv::MatrixLayout::RowMajor,
6473
__spv::MatrixLayout LC = __spv::MatrixLayout::RowMajor,
6574
__spv::Scope::Flag S = __spv::Scope::Flag::Subgroup>
66-
extern SYCL_EXTERNAL __spv::__spirv_JointMatrixINTEL<T3, M, N, LC, S, UC> *
75+
extern SYCL_EXTERNAL JOINT_MATRIX_INTEL(T2, M, N, LC, S, UC) *
6776
__spirv_JointMatrixUUMadINTEL(
68-
__spv::__spirv_JointMatrixINTEL<T1, M, K, LA, S, UA> *A,
69-
__spv::__spirv_JointMatrixINTEL<T2, K, N, LB, S, UB> *B,
70-
__spv::__spirv_JointMatrixINTEL<T3, M, N, LC, S, UC> *C,
77+
JOINT_MATRIX_INTEL(T1, M, K, LA, S, UA) *A,
78+
JOINT_MATRIX_INTEL(T2, K, N, LB, S, UB) *B,
79+
JOINT_MATRIX_INTEL(T3, M, N, LC, S, UC) *C,
7180
__spv::Scope::Flag Sc = __spv::Scope::Flag::Subgroup);
7281

7382
template <typename T1, typename T2, typename T3, std::size_t M, std::size_t K,
@@ -78,11 +87,11 @@ template <typename T1, typename T2, typename T3, std::size_t M, std::size_t K,
7887
__spv::MatrixLayout LB = __spv::MatrixLayout::RowMajor,
7988
__spv::MatrixLayout LC = __spv::MatrixLayout::RowMajor,
8089
__spv::Scope::Flag S = __spv::Scope::Flag::Subgroup>
81-
extern SYCL_EXTERNAL __spv::__spirv_JointMatrixINTEL<T3, M, N, LC, S, UC> *
90+
extern SYCL_EXTERNAL JOINT_MATRIX_INTEL(T3, M, N, LC, S, UC) *
8291
__spirv_JointMatrixUSMadINTEL(
83-
__spv::__spirv_JointMatrixINTEL<T1, M, K, LA, S, UA> *A,
84-
__spv::__spirv_JointMatrixINTEL<T2, K, N, LB, S, UB> *B,
85-
__spv::__spirv_JointMatrixINTEL<T3, M, N, LC, S, UC> *C,
92+
JOINT_MATRIX_INTEL(T1, M, K, LA, S, UA) *A,
93+
JOINT_MATRIX_INTEL(T2, K, N, LB, S, UB) *B,
94+
JOINT_MATRIX_INTEL(T3, M, N, LC, S, UC) *C,
8695
__spv::Scope::Flag Sc = __spv::Scope::Flag::Subgroup);
8796

8897
template <typename T1, typename T2, typename T3, std::size_t M, std::size_t K,
@@ -93,38 +102,42 @@ template <typename T1, typename T2, typename T3, std::size_t M, std::size_t K,
93102
__spv::MatrixLayout LB = __spv::MatrixLayout::RowMajor,
94103
__spv::MatrixLayout LC = __spv::MatrixLayout::RowMajor,
95104
__spv::Scope::Flag S = __spv::Scope::Flag::Subgroup>
96-
extern SYCL_EXTERNAL __spv::__spirv_JointMatrixINTEL<T3, M, N, LC, S, UC> *
105+
extern SYCL_EXTERNAL JOINT_MATRIX_INTEL(T3, M, N, LC, S, UC) *
97106
__spirv_JointMatrixSUMadINTEL(
98-
__spv::__spirv_JointMatrixINTEL<T1, M, K, LA, S, UA> *A,
99-
__spv::__spirv_JointMatrixINTEL<T2, K, N, LB, S, UB> *B,
100-
__spv::__spirv_JointMatrixINTEL<T3, M, N, LC, S, UC> *C,
107+
JOINT_MATRIX_INTEL(T1, M, K, LA, S, UA) *A,
108+
JOINT_MATRIX_INTEL(T2, K, N, LB, S, UB) *B,
109+
JOINT_MATRIX_INTEL(T3, M, N, LC, S, UC) *C,
101110
__spv::Scope::Flag Sc = __spv::Scope::Flag::Subgroup);
102111

103112
template <typename T, std::size_t R, std::size_t C,
104113
__spv::MatrixUse U = __spv::MatrixUse::Unnecessary,
105114
__spv::MatrixLayout L = __spv::MatrixLayout::RowMajor,
106115
__spv::Scope::Flag S = __spv::Scope::Flag::Subgroup>
107-
extern SYCL_EXTERNAL __spv::__spirv_JointMatrixINTEL<T, R, C, L, S, U> *
116+
extern SYCL_EXTERNAL JOINT_MATRIX_INTEL(T, R, C, L, S, U) *
108117
__spirv_CompositeConstruct(const T v);
109118

110-
template <typename T, std::size_t R, std::size_t C, __spv::MatrixUse U,
111-
__spv::MatrixLayout L,
119+
template <typename T, std::size_t R, std::size_t C,
120+
__spv::MatrixUse U = __spv::MatrixUse::Unnecessary,
121+
__spv::MatrixLayout L = __spv::MatrixLayout::RowMajor,
112122
__spv::Scope::Flag S = __spv::Scope::Flag::Subgroup>
113123
extern SYCL_EXTERNAL size_t __spirv_JointMatrixWorkItemLengthINTEL(
114-
__spv::__spirv_JointMatrixINTEL<T, R, C, L, S, U> *);
124+
JOINT_MATRIX_INTEL(T, R, C, L, S, U) *);
115125

116-
template <typename T, std::size_t R, std::size_t C, __spv::MatrixUse U,
117-
__spv::MatrixLayout L,
126+
template <typename T, std::size_t R, std::size_t C,
127+
__spv::MatrixUse U = __spv::MatrixUse::Unnecessary,
128+
__spv::MatrixLayout L = __spv::MatrixLayout::RowMajor,
118129
__spv::Scope::Flag S = __spv::Scope::Flag::Subgroup>
119130
extern SYCL_EXTERNAL T __spirv_VectorExtractDynamic(
120-
__spv::__spirv_JointMatrixINTEL<T, R, C, L, S, U> *, size_t i);
131+
JOINT_MATRIX_INTEL(T, R, C, L, S, U) *, size_t i);
121132

122-
template <typename T, std::size_t R, std::size_t C, __spv::MatrixUse U,
123-
__spv::MatrixLayout L,
133+
template <typename T, std::size_t R, std::size_t C,
134+
__spv::MatrixUse U = __spv::MatrixUse::Unnecessary,
135+
__spv::MatrixLayout L = __spv::MatrixLayout::RowMajor,
124136
__spv::Scope::Flag S = __spv::Scope::Flag::Subgroup>
125-
extern SYCL_EXTERNAL __spv::__spirv_JointMatrixINTEL<T, R, C, L, S, U> *
126-
__spirv_VectorInsertDynamic(__spv::__spirv_JointMatrixINTEL<T, R, C, L, S, U> *,
137+
extern SYCL_EXTERNAL JOINT_MATRIX_INTEL(T, R, C, L, S, U) *
138+
__spirv_VectorInsertDynamic(JOINT_MATRIX_INTEL(T, R, C, L, S, U) *,
127139
T val, size_t i);
140+
#undef JOINT_MATRIX_INTEL
128141

129142
#ifndef __SPIRV_BUILTIN_DECLARATIONS__
130143
#error \

‎sycl/include/CL/__spirv/spirv_types.hpp‎

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -137,6 +137,7 @@ enum class MatrixUse : uint32_t {
137137
// information to SPIRV translator.
138138
// The long term solution would be to introduce a matrix type in Clang and use
139139
// it instead of this member.
140+
#if (SYCL_EXT_ONEAPI_MATRIX_VERSION > 1)
140141
template <typename T, std::size_t R, std::size_t C, MatrixLayout L,
141142
Scope::Flag S = Scope::Flag::Subgroup,
142143
MatrixUse U = MatrixUse::Unnecessary>
@@ -145,6 +146,14 @@ struct __spirv_JointMatrixINTEL {
145146
[R][C][static_cast<size_t>(L) + 1][static_cast<size_t>(S) + 1]
146147
[static_cast<size_t>(U) + 1];
147148
};
149+
#else
150+
template <typename T, std::size_t R, std::size_t C, MatrixLayout L,
151+
Scope::Flag S = Scope::Flag::Subgroup>
152+
struct __spirv_JointMatrixINTEL {
153+
T(*Value)
154+
[R][C][static_cast<size_t>(L) + 1][static_cast<size_t>(S) + 1];
155+
};
156+
#endif // SYCL_EXT_ONEAPI_MATRIX_VERSION
148157

149158
} // namespace __spv
150159

‎sycl/include/sycl/ext/oneapi/matrix/matrix-jit.hpp‎

Lines changed: 1 addition & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -73,8 +73,7 @@ struct joint_matrix {
7373
public:
7474
__spv::__spirv_JointMatrixINTEL<
7575
T, NumRows, NumCols, spv_matrix_layout_traits<Layout>::value,
76-
spv_scope_traits<Group>::value,
77-
spv_matrix_use_traits<matrix_use::unnecessary>::value> *spvm;
76+
spv_scope_traits<Group>::value> *spvm;
7877
joint_matrix(Group sg) {
7978
#ifndef __SYCL_DEVICE_ONLY__
8079
(void)sg;

‎sycl/include/sycl/ext/oneapi/matrix/matrix-tensorcore.hpp‎

Lines changed: 6 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -714,7 +714,7 @@ void joint_matrix_load(
714714
std::ignore = src;
715715
std::ignore = stride;
716716
throw runtime_error(
717-
"When using SYCL_EXT_ONEAPI_MATRIX=3 joint_matrix_load is "
717+
"When using SYCL_EXT_ONEAPI_MATRIX_VERSION=3 joint_matrix_load is "
718718
"only supported by CUDA devices",
719719
PI_ERROR_INVALID_DEVICE);
720720
#endif // defined(__SYCL_DEVICE_ONLY__) && defined(__NVPTX__)
@@ -736,7 +736,7 @@ void joint_matrix_store(Group sg,
736736
std::ignore = dst;
737737
std::ignore = stride;
738738
throw runtime_error(
739-
"When using SYCL_EXT_ONEAPI_MATRIX=3 joint_matrix_store is "
739+
"When using SYCL_EXT_ONEAPI_MATRIX_VERSION=3 joint_matrix_store is "
740740
"only supported by CUDA devices",
741741
PI_ERROR_INVALID_DEVICE);
742742
#endif // defined(__SYCL_DEVICE_ONLY__) && defined(__NVPTX__)
@@ -759,9 +759,10 @@ joint_matrix_mad(
759759
std::ignore = A;
760760
std::ignore = B;
761761
std::ignore = C;
762-
throw runtime_error("When using SYCL_EXT_ONEAPI_MATRIX=3 joint_matrix_mad is "
763-
"only supported by CUDA devices",
764-
PI_ERROR_INVALID_DEVICE);
762+
throw runtime_error(
763+
"When using SYCL_EXT_ONEAPI_MATRIX_VERSION=3 joint_matrix_mad is "
764+
"only supported by CUDA devices",
765+
PI_ERROR_INVALID_DEVICE);
765766
#endif // defined(__SYCL_DEVICE_ONLY__) && defined(__NVPTX__)
766767
}
767768

‎sycl/include/sycl/ext/oneapi/matrix/matrix.hpp‎

Lines changed: 9 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -14,18 +14,18 @@
1414

1515
#pragma once
1616

17-
#include <sycl/feature_test.hpp>
17+
#ifndef SYCL_EXT_ONEAPI_MATRIX_VERSION
18+
#define SYCL_EXT_ONEAPI_MATRIX_VERSION 1
19+
#endif // SYCL_EXT_ONEAPI_MATRIX_VERSION
1820

19-
// the default is matrix-jit-use but existing tests in llvm-test-suite won't
20-
// fail because we have the "unnecessary" use value
21-
#if (SYCL_EXT_ONEAPI_MATRIX == 1)
21+
#if (SYCL_EXT_ONEAPI_MATRIX_VERSION == 1)
2222
#include <sycl/ext/oneapi/matrix/matrix-jit.hpp>
2323
#include <sycl/ext/oneapi/matrix/static-query.hpp>
24-
#endif
25-
#if (SYCL_EXT_ONEAPI_MATRIX == 2)
24+
#endif // SYCL_EXT_ONEAPI_MATRIX_VERSION
25+
#if (SYCL_EXT_ONEAPI_MATRIX_VERSION == 2)
2626
#include <sycl/ext/oneapi/matrix/matrix-jit-use.hpp>
2727
#include <sycl/ext/oneapi/matrix/static-query-use.hpp>
28-
#endif
29-
#if (SYCL_EXT_ONEAPI_MATRIX == 3)
28+
#endif // SYCL_EXT_ONEAPI_MATRIX_VERSION
29+
#if (SYCL_EXT_ONEAPI_MATRIX_VERSION == 3)
3030
#include <sycl/ext/oneapi/matrix/matrix-tensorcore.hpp>
31-
#endif
31+
#endif // SYCL_EXT_ONEAPI_MATRIX_VERSION

‎sycl/include/sycl/feature_test.hpp.in‎

Lines changed: 1 addition & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -32,14 +32,7 @@ __SYCL_INLINE_VER_NAMESPACE(_V1) {
3232
#define SYCL_EXT_INTEL_DEVICE_INFO 3
3333
#define SYCL_EXT_ONEAPI_SUB_GROUP_MASK 1
3434
#define SYCL_EXT_ONEAPI_LOCAL_MEMORY 1
35-
// As for SYCL_EXT_ONEAPI_MATRIX:
36-
// 1- provides AOT initial implementation for AMX for the experimental matrix
37-
// extension
38-
// 2- provides JIT implementation (target agnostic) for the
39-
// experimental matrix extension
40-
#ifndef SYCL_EXT_ONEAPI_MATRIX
41-
#define SYCL_EXT_ONEAPI_MATRIX 2
42-
#endif
35+
#define SYCL_EXT_ONEAPI_MATRIX 1
4336
#define SYCL_EXT_ONEAPI_ASSERT 1
4437
#define SYCL_EXT_ONEAPI_COMPLEX_ALGORITHMS 1
4538
#define SYCL_EXT_ONEAPI_DISCARD_QUEUE_EVENTS 1

‎sycl/test/check_device_code/matrix/matrix-nvptx-bf16-test.cpp‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,6 @@
11
// REQUIRES: cuda
22

3-
// RUN: %clangxx -fsycl-device-only -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend --cuda-gpu-arch=sm_80 -DSYCL_EXT_ONEAPI_MATRIX=3 -S -Xclang -emit-llvm %s -o -| FileCheck %s
3+
// RUN: %clangxx -fsycl-device-only -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend --cuda-gpu-arch=sm_80 -DSYCL_EXT_ONEAPI_MATRIX_VERSION=3 -S -Xclang -emit-llvm %s -o -| FileCheck %s
44

55
#include <sycl/sycl.hpp>
66

‎sycl/test/check_device_code/matrix/matrix-nvptx-bfloat16-test.cpp‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,6 @@
11
// REQUIRES: cuda
22

3-
// RUN: %clangxx -fsycl-device-only -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend --cuda-gpu-arch=sm_80 -DSYCL_EXT_ONEAPI_MATRIX=3 -S -Xclang -emit-llvm %s -o -| FileCheck %s
3+
// RUN: %clangxx -fsycl-device-only -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend --cuda-gpu-arch=sm_80 -DSYCL_EXT_ONEAPI_MATRIX_VERSION=3 -S -Xclang -emit-llvm %s -o -| FileCheck %s
44

55
#include <sycl/sycl.hpp>
66

‎sycl/test/check_device_code/matrix/matrix-nvptx-double-test.cpp‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,6 @@
11
// REQUIRES: cuda
22

3-
// RUN: %clangxx -fsycl-device-only -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend --cuda-gpu-arch=sm_80 -DSYCL_EXT_ONEAPI_MATRIX=3 -S -Xclang -emit-llvm %s -o -| FileCheck %s
3+
// RUN: %clangxx -fsycl-device-only -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend --cuda-gpu-arch=sm_80 -DSYCL_EXT_ONEAPI_MATRIX_VERSION=3 -S -Xclang -emit-llvm %s -o -| FileCheck %s
44

55
#include <sycl/sycl.hpp>
66

‎sycl/test/check_device_code/matrix/matrix-nvptx-half-float-test.cpp‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,6 @@
11
// REQUIRES: cuda
22

3-
// RUN: %clangxx -fsycl-device-only -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend --cuda-gpu-arch=sm_70 -DSYCL_EXT_ONEAPI_MATRIX=3 -S -Xclang -emit-llvm %s -o -| FileCheck %s
3+
// RUN: %clangxx -fsycl-device-only -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend --cuda-gpu-arch=sm_70 -DSYCL_EXT_ONEAPI_MATRIX_VERSION=3 -S -Xclang -emit-llvm %s -o -| FileCheck %s
44

55
#include <sycl/sycl.hpp>
66

0 commit comments

Comments
 (0)