Skip to content

Commit 2647cf9

Browse files
PanZezhong1725PanZezhong1725
committed
issue/1031 fix T1-1-41
1 parent 6431be3 commit 2647cf9

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

43 files changed

+993
-398
lines changed

‎include/infiniop/ops/erf.h‎

Lines changed: 12 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -5,20 +5,20 @@
55

66
typedef struct InfiniopDescriptor *infiniopErfDescriptor_t;
77

8-
__C __export infiniStatus_t infiniopCreateErfDescriptor(infiniopHandle_t handle,
9-
infiniopErfDescriptor_t *desc_ptr,
10-
infiniopTensorDescriptor_t y,
11-
infiniopTensorDescriptor_t x);
8+
__INFINI_C __export infiniStatus_t infiniopCreateErfDescriptor(infiniopHandle_t handle,
9+
infiniopErfDescriptor_t *desc_ptr,
10+
infiniopTensorDescriptor_t y,
11+
infiniopTensorDescriptor_t x);
1212

13-
__C __export infiniStatus_t infiniopGetErfWorkspaceSize(infiniopErfDescriptor_t desc, size_t *size);
13+
__INFINI_C __export infiniStatus_t infiniopGetErfWorkspaceSize(infiniopErfDescriptor_t desc, size_t *size);
1414

15-
__C __export infiniStatus_t infiniopErf(infiniopErfDescriptor_t desc,
16-
void *workspace,
17-
size_t workspace_size,
18-
void *y,
19-
const void *x,
20-
void *stream);
15+
__INFINI_C __export infiniStatus_t infiniopErf(infiniopErfDescriptor_t desc,
16+
void *workspace,
17+
size_t workspace_size,
18+
void *y,
19+
const void *x,
20+
void *stream);
2121

22-
__C __export infiniStatus_t infiniopDestroyErfDescriptor(infiniopErfDescriptor_t desc);
22+
__INFINI_C __export infiniStatus_t infiniopDestroyErfDescriptor(infiniopErfDescriptor_t desc);
2323

2424
#endif

‎include/infiniop/ops/erfc.h‎

Lines changed: 12 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -5,20 +5,20 @@
55

66
typedef struct InfiniopDescriptor *infiniopErfcDescriptor_t;
77

8-
__C __export infiniStatus_t infiniopCreateErfcDescriptor(infiniopHandle_t handle,
9-
infiniopErfcDescriptor_t *desc_ptr,
10-
infiniopTensorDescriptor_t y,
11-
infiniopTensorDescriptor_t x);
8+
__INFINI_C __export infiniStatus_t infiniopCreateErfcDescriptor(infiniopHandle_t handle,
9+
infiniopErfcDescriptor_t *desc_ptr,
10+
infiniopTensorDescriptor_t y,
11+
infiniopTensorDescriptor_t x);
1212

13-
__C __export infiniStatus_t infiniopGetErfcWorkspaceSize(infiniopErfcDescriptor_t desc, size_t *size);
13+
__INFINI_C __export infiniStatus_t infiniopGetErfcWorkspaceSize(infiniopErfcDescriptor_t desc, size_t *size);
1414

15-
__C __export infiniStatus_t infiniopErfc(infiniopErfcDescriptor_t desc,
16-
void *workspace,
17-
size_t workspace_size,
18-
void *y,
19-
const void *x,
20-
void *stream);
15+
__INFINI_C __export infiniStatus_t infiniopErfc(infiniopErfcDescriptor_t desc,
16+
void *workspace,
17+
size_t workspace_size,
18+
void *y,
19+
const void *x,
20+
void *stream);
2121

22-
__C __export infiniStatus_t infiniopDestroyErfcDescriptor(infiniopErfcDescriptor_t desc);
22+
__INFINI_C __export infiniStatus_t infiniopDestroyErfcDescriptor(infiniopErfcDescriptor_t desc);
2323

2424
#endif

‎include/infiniop/ops/erfinv.h‎

Lines changed: 12 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -5,20 +5,20 @@
55

66
typedef struct InfiniopDescriptor *infiniopErfinvDescriptor_t;
77

8-
__C __export infiniStatus_t infiniopCreateErfinvDescriptor(infiniopHandle_t handle,
9-
infiniopErfinvDescriptor_t *desc_ptr,
10-
infiniopTensorDescriptor_t y,
11-
infiniopTensorDescriptor_t x);
8+
__INFINI_C __export infiniStatus_t infiniopCreateErfinvDescriptor(infiniopHandle_t handle,
9+
infiniopErfinvDescriptor_t *desc_ptr,
10+
infiniopTensorDescriptor_t y,
11+
infiniopTensorDescriptor_t x);
1212

13-
__C __export infiniStatus_t infiniopGetErfinvWorkspaceSize(infiniopErfinvDescriptor_t desc, size_t *size);
13+
__INFINI_C __export infiniStatus_t infiniopGetErfinvWorkspaceSize(infiniopErfinvDescriptor_t desc, size_t *size);
1414

15-
__C __export infiniStatus_t infiniopErfinv(infiniopErfinvDescriptor_t desc,
16-
void *workspace,
17-
size_t workspace_size,
18-
void *y,
19-
const void *x,
20-
void *stream);
15+
__INFINI_C __export infiniStatus_t infiniopErfinv(infiniopErfinvDescriptor_t desc,
16+
void *workspace,
17+
size_t workspace_size,
18+
void *y,
19+
const void *x,
20+
void *stream);
2121

22-
__C __export infiniStatus_t infiniopDestroyErfinvDescriptor(infiniopErfinvDescriptor_t desc);
22+
__INFINI_C __export infiniStatus_t infiniopDestroyErfinvDescriptor(infiniopErfinvDescriptor_t desc);
2323

2424
#endif

‎include/infiniop/ops/matrix_power.h‎

Lines changed: 13 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -5,21 +5,21 @@
55

66
typedef struct InfiniopDescriptor *infiniopMatrixPowerDescriptor_t;
77

8-
__C __export infiniStatus_t infiniopCreateMatrixPowerDescriptor(infiniopHandle_t handle,
9-
infiniopMatrixPowerDescriptor_t *desc_ptr,
10-
infiniopTensorDescriptor_t y,
11-
infiniopTensorDescriptor_t x,
12-
int n);
8+
__INFINI_C __export infiniStatus_t infiniopCreateMatrixPowerDescriptor(infiniopHandle_t handle,
9+
infiniopMatrixPowerDescriptor_t *desc_ptr,
10+
infiniopTensorDescriptor_t y,
11+
infiniopTensorDescriptor_t x,
12+
int n);
1313

14-
__C __export infiniStatus_t infiniopGetMatrixPowerWorkspaceSize(infiniopMatrixPowerDescriptor_t desc, size_t *size);
14+
__INFINI_C __export infiniStatus_t infiniopGetMatrixPowerWorkspaceSize(infiniopMatrixPowerDescriptor_t desc, size_t *size);
1515

16-
__C __export infiniStatus_t infiniopMatrixPower(infiniopMatrixPowerDescriptor_t desc,
17-
void *workspace,
18-
size_t workspace_size,
19-
void *y,
20-
const void *x,
21-
void *stream);
16+
__INFINI_C __export infiniStatus_t infiniopMatrixPower(infiniopMatrixPowerDescriptor_t desc,
17+
void *workspace,
18+
size_t workspace_size,
19+
void *y,
20+
const void *x,
21+
void *stream);
2222

23-
__C __export infiniStatus_t infiniopDestroyMatrixPowerDescriptor(infiniopMatrixPowerDescriptor_t desc);
23+
__INFINI_C __export infiniStatus_t infiniopDestroyMatrixPowerDescriptor(infiniopMatrixPowerDescriptor_t desc);
2424

2525
#endif

‎include/infiniop/ops/pixel_shuffle.h‎

Lines changed: 13 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -5,21 +5,21 @@
55

66
typedef struct InfiniopDescriptor *infiniopPixelShuffleDescriptor_t;
77

8-
__C __export infiniStatus_t infiniopCreatePixelShuffleDescriptor(infiniopHandle_t handle,
9-
infiniopPixelShuffleDescriptor_t *desc_ptr,
10-
infiniopTensorDescriptor_t y,
11-
infiniopTensorDescriptor_t x,
12-
int upscale_factor);
8+
__INFINI_C __export infiniStatus_t infiniopCreatePixelShuffleDescriptor(infiniopHandle_t handle,
9+
infiniopPixelShuffleDescriptor_t *desc_ptr,
10+
infiniopTensorDescriptor_t y,
11+
infiniopTensorDescriptor_t x,
12+
int upscale_factor);
1313

14-
__C __export infiniStatus_t infiniopGetPixelShuffleWorkspaceSize(infiniopPixelShuffleDescriptor_t desc, size_t *size);
14+
__INFINI_C __export infiniStatus_t infiniopGetPixelShuffleWorkspaceSize(infiniopPixelShuffleDescriptor_t desc, size_t *size);
1515

16-
__C __export infiniStatus_t infiniopPixelShuffle(infiniopPixelShuffleDescriptor_t desc,
17-
void *workspace,
18-
size_t workspace_size,
19-
void *y,
20-
const void *x,
21-
void *stream);
16+
__INFINI_C __export infiniStatus_t infiniopPixelShuffle(infiniopPixelShuffleDescriptor_t desc,
17+
void *workspace,
18+
size_t workspace_size,
19+
void *y,
20+
const void *x,
21+
void *stream);
2222

23-
__C __export infiniStatus_t infiniopDestroyPixelShuffleDescriptor(infiniopPixelShuffleDescriptor_t desc);
23+
__INFINI_C __export infiniStatus_t infiniopDestroyPixelShuffleDescriptor(infiniopPixelShuffleDescriptor_t desc);
2424

2525
#endif

‎src/infiniop/ops/erf/cuda/kernel.cuh‎

Lines changed: 1 addition & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1,8 +1,5 @@
11
#pragma once
22
#include <cmath>
3-
#include <cuda_bf16.h>
4-
#include <cuda_fp16.h>
5-
#include <cuda_runtime.h>
63
#include <type_traits>
74

85
namespace op::cuda {
@@ -15,7 +12,7 @@ struct ErfOp {
1512
if constexpr (std::is_same_v<T, float>) {
1613
return erff(x);
1714
} else if constexpr (std::is_same_v<T, double>) {
18-
return erf(x);
15+
return ::erf(x);
1916
} else {
2017
// For F16/BF16: promote to float, compute, then cast back
2118
float xf;

‎src/infiniop/ops/erf/moore/erf_moore.mu‎

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@
22

33
#include "../../../elementwise/moore/elementwise_moore.h"
44

5-
#include "erf_moore_kernel.h"
5+
#include "../cuda/kernel.cuh"
66

77
namespace op::erf::moore {
88

@@ -43,13 +43,13 @@ infiniStatus_t Descriptor::calculate(
4343

4444
switch (_dtype) {
4545
case INFINI_DTYPE_BF16:
46-
return _device_info->calculate<256, moore::ErfOp, cuda_bfloat16>(_info, workspace, output, inputs, stream);
46+
return _device_info->calculate<256, cuda::ErfOp, cuda_bfloat16>(_info, workspace, output, inputs, stream);
4747
case INFINI_DTYPE_F16:
48-
return _device_info->calculate<256, moore::ErfOp, half>(_info, workspace, output, inputs, stream);
48+
return _device_info->calculate<256, cuda::ErfOp, half>(_info, workspace, output, inputs, stream);
4949
case INFINI_DTYPE_F32:
50-
return _device_info->calculate<256, moore::ErfOp, float>(_info, workspace, output, inputs, stream);
50+
return _device_info->calculate<256, cuda::ErfOp, float>(_info, workspace, output, inputs, stream);
5151
case INFINI_DTYPE_F64:
52-
return _device_info->calculate<256, moore::ErfOp, double>(_info, workspace, output, inputs, stream);
52+
return _device_info->calculate<256, cuda::ErfOp, double>(_info, workspace, output, inputs, stream);
5353
default:
5454
return INFINI_STATUS_BAD_TENSOR_DTYPE;
5555
}

‎src/infiniop/ops/erf/moore/erf_moore_kernel.h‎

Lines changed: 0 additions & 36 deletions
This file was deleted.

‎src/infiniop/ops/erf/operator.cc‎

Lines changed: 17 additions & 17 deletions
Original file line numberDiff line numberDiff line change
@@ -15,18 +15,18 @@
1515
#include "moore/erf_moore.h"
1616
#endif
1717

18-
__C __export infiniStatus_t infiniopCreateErfDescriptor(
18+
__INFINI_C __export infiniStatus_t infiniopCreateErfDescriptor(
1919
infiniopHandle_t handle,
2020
infiniopErfDescriptor_t *desc_ptr,
2121
infiniopTensorDescriptor_t y_desc,
2222
infiniopTensorDescriptor_t x_desc) {
2323

24-
#define CREATE(CASE, NAMESPACE) \
25-
case CASE: \
26-
return op::erf::NAMESPACE::Descriptor::create( \
27-
handle, \
28-
reinterpret_cast<op::erf::NAMESPACE::Descriptor **>(desc_ptr), \
29-
y_desc, \
24+
#define CREATE(CASE, NAMESPACE) \
25+
case CASE: \
26+
return op::erf::NAMESPACE::Descriptor::create( \
27+
handle, \
28+
reinterpret_cast<op::erf::NAMESPACE::Descriptor **>(desc_ptr), \
29+
y_desc, \
3030
{x_desc})
3131

3232
switch (handle->device) {
@@ -54,10 +54,10 @@ __C __export infiniStatus_t infiniopCreateErfDescriptor(
5454
#undef CREATE
5555
}
5656

57-
__C __export infiniStatus_t infiniopGetErfWorkspaceSize(infiniopErfDescriptor_t desc, size_t *size) {
57+
__INFINI_C __export infiniStatus_t infiniopGetErfWorkspaceSize(infiniopErfDescriptor_t desc, size_t *size) {
5858

59-
#define GET(CASE, NAMESPACE) \
60-
case CASE: \
59+
#define GET(CASE, NAMESPACE) \
60+
case CASE: \
6161
*size = reinterpret_cast<op::erf::NAMESPACE::Descriptor *>(desc)->workspaceSize(); \
6262
return INFINI_STATUS_SUCCESS;
6363

@@ -85,17 +85,17 @@ __C __export infiniStatus_t infiniopGetErfWorkspaceSize(infiniopErfDescriptor_t
8585
return INFINI_STATUS_DEVICE_TYPE_NOT_SUPPORTED;
8686
}
8787

88-
__C __export infiniStatus_t infiniopErf(
88+
__INFINI_C __export infiniStatus_t infiniopErf(
8989
infiniopErfDescriptor_t desc,
9090
void *workspace,
9191
size_t workspace_size,
9292
void *y,
9393
const void *x,
9494
void *stream) {
9595

96-
#define CALCULATE(CASE, NAMESPACE) \
97-
case CASE: \
98-
return reinterpret_cast<const op::erf::NAMESPACE::Descriptor *>(desc) \
96+
#define CALCULATE(CASE, NAMESPACE) \
97+
case CASE: \
98+
return reinterpret_cast<const op::erf::NAMESPACE::Descriptor *>(desc) \
9999
->calculate(workspace, workspace_size, y, {x}, stream)
100100

101101
switch (desc->device_type) {
@@ -123,11 +123,11 @@ __C __export infiniStatus_t infiniopErf(
123123
#undef CALCULATE
124124
}
125125

126-
__C __export infiniStatus_t
126+
__INFINI_C __export infiniStatus_t
127127
infiniopDestroyErfDescriptor(infiniopErfDescriptor_t desc) {
128128

129-
#define DELETE(CASE, NAMESPACE) \
130-
case CASE: \
129+
#define DELETE(CASE, NAMESPACE) \
130+
case CASE: \
131131
delete reinterpret_cast<const op::erf::NAMESPACE::Descriptor *>(desc); \
132132
return INFINI_STATUS_SUCCESS;
133133

‎src/infiniop/ops/erfc/cuda/kernel.cuh‎

Lines changed: 1 addition & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1,8 +1,5 @@
11
#pragma once
22
#include <cmath>
3-
#include <cuda_bf16.h>
4-
#include <cuda_fp16.h>
5-
#include <cuda_runtime.h>
63
#include <type_traits>
74

85
namespace op::cuda {
@@ -15,7 +12,7 @@ struct ErfcOp {
1512
if constexpr (std::is_same_v<T, float>) {
1613
return erfcf(x);
1714
} else if constexpr (std::is_same_v<T, double>) {
18-
return erfc(x);
15+
return ::erfc(x);
1916
} else {
2017
// For F16/BF16: promote to float, compute, then cast back
2118
float xf;

0 commit comments

Comments
 (0)