Skip to content

Avoid memory allocations and deallocations when creating NVTETensor#1813

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ptrendx merged 19 commits into from
May 29, 2025
Merged

Avoid memory allocations and deallocations when creating NVTETensor #1813

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
19 commits
Select commit Hold shift + click to select a range
500533b
Changed the Tensor allocation strategy
ptrendx May 19, 2025
c410dc1
Fixes
ptrendx May 20, 2025
7a2b7c9
Disable debug flag
ptrendx May 20, 2025
795c8a7
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] May 21, 2025
7eb0000
Merge branch 'main' into pr_cpp_avoid_allocs
ptrendx May 27, 2025
cfbb136
Fix the double free error
ptrendx May 27, 2025
8fbe02a
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] May 27, 2025
56c32e6
Fix
ptrendx May 27, 2025
dd79cc6
Fixed pyTorch recipe extension
ptrendx May 27, 2025
bbfec4a
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] May 27, 2025
81a33fa
Fix
ptrendx May 27, 2025
13d9937
Fix
ptrendx May 27, 2025
5d6b374
Hide TensorAllocator and fix the usage in LayerNorm
ptrendx May 27, 2025
e7a70bd
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] May 27, 2025
da5c790
Cleaning
ptrendx May 27, 2025
6596719
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] May 27, 2025
7d4bc70
Fix
ptrendx May 27, 2025
db7621c
Fix permutation
ptrendx May 28, 2025
c6f4f40
Merge branch 'main' into pr_cpp_avoid_allocs
ptrendx May 28, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion transformer_engine/common/common.cu
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -200,7 +200,7 @@ std::vector<std::vector<Tensor *>> convert_tensor_array(NVTETensor **nvte_tensor
for (size_t i = 0; i < outer_size; ++i) {
ret.emplace_back();
for (size_t j = 0; j < inner_size; ++j) {
ret.back().push_back(reinterpret_cast<Tensor *>(nvte_tensors[i][j]));
ret.back().push_back(convertNVTETensor(nvte_tensors[i][j]));
}
}
return ret;
Expand Down
26 changes: 24 additions & 2 deletions transformer_engine/common/common.h
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -89,18 +89,25 @@ struct SimpleTensor {
}
return acc;
}

void clear() {
dptr = nullptr;
shape.resize(0);
dtype = DType::kFloat32;
}
};

struct Tensor {
public:
SimpleTensor data;
SimpleTensor columnwise_data;
SimpleTensor amax;
SimpleTensor scale;
SimpleTensor scale_inv;
SimpleTensor columnwise_scale_inv;

public:
NVTEScalingMode scaling_mode;
NVTETensor nvte_tensor;

Tensor()
: data(),
Expand All@@ -109,7 +116,20 @@ struct Tensor {
scale(nullptr, {1}, DType::kFloat32),
scale_inv(nullptr, {1}, DType::kFloat32),
columnwise_scale_inv(nullptr, {1}, DType::kFloat32),
scaling_mode(NVTE_DELAYED_TENSOR_SCALING) {}
scaling_mode(NVTE_DELAYED_TENSOR_SCALING),
nvte_tensor(0) {}

void clear() {
data.clear();
columnwise_data.clear();
amax.clear();
scale.clear();
scale_inv.clear();
columnwise_scale_inv.clear();
scaling_mode = NVTE_DELAYED_TENSOR_SCALING;
}

explicit operator NVTETensor() const noexcept { return nvte_tensor; }

size_t numel() const {
size_t acc = 1;
Expand DownExpand Up@@ -620,6 +640,8 @@ bool is_supported_by_CC_100();
std::vector<std::vector<Tensor *>> convert_tensor_array(NVTETensor **nvte_tensors,
size_t outer_size, size_t inner_size);

Tensor *convertNVTETensor(const NVTETensor tensor);
Tensor *convertNVTETensorCheck(const NVTETensor tensor);
} // namespace transformer_engine

#endif // TRANSFORMER_ENGINE_COMMON_COMMON_H_
28 changes: 14 additions & 14 deletions transformer_engine/common/fused_attn/context_parallel.cu
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -677,9 +677,9 @@ void nvte_cp_thd_read_half_tensor(const NVTETensor &tensor, const NVTETensor &cu
NVTE_API_CALL(nvte_thd_read_half_tensor);
using namespace transformer_engine;

context_parallel::thd_read_half_tensor(*reinterpret_cast<Tensor *>(tensor),
*reinterpret_cast<Tensor *>(cu_seqlens),
*reinterpret_cast<Tensor *>(half), half_idx, stream);
context_parallel::thd_read_half_tensor(*convertNVTETensorCheck(tensor),
*convertNVTETensorCheck(cu_seqlens),
*convertNVTETensorCheck(half), half_idx, stream);
}

void nvte_cp_thd_second_half_lse_correction(NVTETensor lse, const NVTETensor &lse_per_step,
Expand All@@ -689,8 +689,8 @@ void nvte_cp_thd_second_half_lse_correction(NVTETensor lse, const NVTETensor &ls
using namespace transformer_engine;

context_parallel::thd_second_half_lse_correction(
*reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(lse_per_step),
*reinterpret_cast<Tensor *>(cu_seqlens), lse_packed, stream);
*convertNVTETensorCheck(lse), *convertNVTETensorCheck(lse_per_step),
*convertNVTETensorCheck(cu_seqlens), lse_packed, stream);
}

void nvte_cp_thd_read_second_half_lse(const NVTETensor &lse, const NVTETensor &cu_seqlens,
Expand All@@ -700,8 +700,8 @@ void nvte_cp_thd_read_second_half_lse(const NVTETensor &lse, const NVTETensor &c
using namespace transformer_engine;

context_parallel::thd_read_second_half_lse(
*reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(cu_seqlens),
*reinterpret_cast<Tensor *>(half_lse), lse_packed, second_half_lse_seqlen, stream);
*convertNVTETensorCheck(lse), *convertNVTETensorCheck(cu_seqlens),
*convertNVTETensorCheck(half_lse), lse_packed, second_half_lse_seqlen, stream);
}

void nvte_cp_thd_out_correction(NVTETensor out, const NVTETensor &out_per_step,
Expand All@@ -712,9 +712,9 @@ void nvte_cp_thd_out_correction(NVTETensor out, const NVTETensor &out_per_step,
using namespace transformer_engine;

context_parallel::thd_out_correction(
*reinterpret_cast<Tensor *>(out), *reinterpret_cast<Tensor *>(out_per_step),
*reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(lse_per_step),
*reinterpret_cast<Tensor *>(cu_seqlens), only_second_half, lse_packed, stream);
*convertNVTETensorCheck(out), *convertNVTETensorCheck(out_per_step),
*convertNVTETensorCheck(lse), *convertNVTETensorCheck(lse_per_step),
*convertNVTETensorCheck(cu_seqlens), only_second_half, lse_packed, stream);
}

void nvte_cp_thd_grad_correction(NVTETensor grad, const NVTETensor &grad_per_step,
Expand All@@ -727,8 +727,8 @@ void nvte_cp_thd_grad_correction(NVTETensor grad, const NVTETensor &grad_per_ste
std::string second_half_str(second_half);

context_parallel::thd_grad_correction(
*reinterpret_cast<Tensor *>(grad), *reinterpret_cast<Tensor *>(grad_per_step),
*reinterpret_cast<Tensor *>(cu_seqlens), first_half_str, second_half_str, stream);
*convertNVTETensorCheck(grad), *convertNVTETensorCheck(grad_per_step),
*convertNVTETensorCheck(cu_seqlens), first_half_str, second_half_str, stream);
}

void nvte_cp_thd_get_partitioned_indices(const NVTETensor &cu_seqlens, NVTETensor output,
Expand All@@ -737,7 +737,7 @@ void nvte_cp_thd_get_partitioned_indices(const NVTETensor &cu_seqlens, NVTETenso
NVTE_API_CALL(nvte_thd_get_partitioned_indices);
using namespace transformer_engine;

context_parallel::thd_get_partitioned_indices(*reinterpret_cast<Tensor *>(cu_seqlens),
*reinterpret_cast<Tensor *>(output), total_tokens,
context_parallel::thd_get_partitioned_indices(*convertNVTETensorCheck(cu_seqlens),
*convertNVTETensorCheck(output), total_tokens,
world_size, rank, stream);
}
10 changes: 5 additions & 5 deletions transformer_engine/common/fused_attn/flash_attn.cu
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -138,16 +138,16 @@ void nvte_prepare_flash_attn_fwd(NVTETensor qkvi, NVTETensor qkv, cudaStream_t s
NVTE_API_CALL(nvte_prepare_flash_attn_fwd);
using namespace transformer_engine;

flash_attention::prepare_flash_attn_fwd(*reinterpret_cast<Tensor *>(qkvi),
*reinterpret_cast<Tensor *>(qkv), stream);
flash_attention::prepare_flash_attn_fwd(*convertNVTETensorCheck(qkvi),
*convertNVTETensorCheck(qkv), stream);
}

void nvte_prepare_flash_attn_bwd(NVTETensor q, NVTETensor k, NVTETensor v, NVTETensor qkv,
cudaStream_t stream) {
NVTE_API_CALL(nvte_prepare_flash_attn_bwd);
using namespace transformer_engine;

flash_attention::prepare_flash_attn_bwd(
*reinterpret_cast<Tensor *>(q), *reinterpret_cast<Tensor *>(k),
*reinterpret_cast<Tensor *>(v), *reinterpret_cast<Tensor *>(qkv), stream);
flash_attention::prepare_flash_attn_bwd(*convertNVTETensorCheck(q), *convertNVTETensorCheck(k),
*convertNVTETensorCheck(v), *convertNVTETensorCheck(qkv),
stream);
}
Loading
, 'i'); if (__m === '*' || __re.test(location.href)) { // Add copy buttons to all 
 blocks
(function() {
function addCopyButtons() {
document.querySelectorAll('pre code').forEach(function(codeBlock) {
if (codeBlock.parentElement.hasAttribute('data-copy-added')) return;
codeBlock.parentElement.setAttribute('data-copy-added', 'true');
var btn = document.createElement('button');
btn.textContent = 'Copy';
btn.style.cssText = 'position:absolute;top:4px;right:4px;padding:2px 8px;font-size:11px;background:#4ecdc4;border:none;border-radius:4px;color:#1a1a2e;cursor:pointer;opacity:0.7;transition:opacity 0.2s;';
btn.onmouseover = function() { this.style.opacity = '1'; };
btn.onmouseout = function() { this.style.opacity = '0.7'; };
btn.onclick = function() {
navigator.clipboard.writeText(codeBlock.textContent).then(function() {
btn.textContent = 'Copied!';
setTimeout(function() { btn.textContent = 'Copy'; }, 1500);
});
};
codeBlock.parentElement.style.position = 'relative';
codeBlock.parentElement.appendChild(btn);
});
}
addCopyButtons();
// Re-run on dynamic content
var observer = new MutationObserver(addCopyButtons);
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Add Copy Buttons to Code Blocks]', __e); }
})();
(function(){
try {
var __m = "github.com";
var __re = new RegExp('^' + "github\\.com" + '
Avoid memory allocations and deallocations when creating NVTETensor by ptrendx · Pull Request #1813 · NVIDIA/TransformerEngine · GitHub
Skip to content
Avoid memory allocations and deallocations when creating NVTETensor#1813

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

 Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and
privacy statement. We’ll occasionally send you account related emails.

Already on GitHub?
Sign in
to your account

 Merged

ptrendx
merged 19 commits into

from

May 29, 2025
 Merged


Avoid memory allocations and deallocations when creating NVTETensor
#1813
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


19 commits


Select commit
Hold shift + click to select a range
500533b

Changed the Tensor allocation strategy


ptrendx May 19, 2025
c410dc1

Fixes


ptrendx May 20, 2025
7a2b7c9

Disable debug flag


ptrendx May 20, 2025
795c8a7

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 21, 2025
7eb0000

Merge branch 'main' into pr_cpp_avoid_allocs


ptrendx May 27, 2025
cfbb136

Fix the double free error


ptrendx May 27, 2025
8fbe02a

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
56c32e6

Fix


ptrendx May 27, 2025
dd79cc6

Fixed pyTorch recipe extension


ptrendx May 27, 2025
bbfec4a

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
81a33fa

Fix


ptrendx May 27, 2025
13d9937

Fix


ptrendx May 27, 2025
5d6b374

Hide TensorAllocator and fix the usage in LayerNorm


ptrendx May 27, 2025
e7a70bd

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
da5c790

Cleaning


ptrendx May 27, 2025
6596719

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
7d4bc70

Fix


ptrendx May 27, 2025
db7621c

Fix permutation


ptrendx May 28, 2025
c6f4f40

Merge branch 'main' into pr_cpp_avoid_allocs


ptrendx May 28, 2025

File filter

Filter by extension
Filter by extension
Conversations

Failed to load comments. 
Loading
Jump to

Jump to file


Failed to load files. 
Loading
 Diff view

 Diff view


2 changes: 1 addition & 1 deletion
transformer_engine/common/common.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -200,7 +200,7 @@ std::vector<std::vector<Tensor *>> convert_tensor_array(NVTETensor **nvte_tensor
 for (size_t i = 0; i < outer_size; ++i) {
 ret.emplace_back();
 for (size_t j = 0; j < inner_size; ++j) {
 ret.back().push_back(reinterpret_cast<Tensor *>(nvte_tensors[i][j]));
 ret.back().push_back(convertNVTETensor(nvte_tensors[i][j]));
 }
 }
 return ret;
Expand Down

26 changes: 24 additions & 2 deletions
transformer_engine/common/common.h

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -89,18 +89,25 @@ struct SimpleTensor {
 }
 return acc;
 }

 void clear() {
 dptr = nullptr;
 shape.resize(0);
 dtype = DType::kFloat32;
 }
};

struct Tensor {
 public:
 SimpleTensor data;
 SimpleTensor columnwise_data;
 SimpleTensor amax;
 SimpleTensor scale;
 SimpleTensor scale_inv;
 SimpleTensor columnwise_scale_inv;

 public:
 NVTEScalingMode scaling_mode;
 NVTETensor nvte_tensor;

 Tensor()
 : data(),
Expand All@@ -109,7 +116,20 @@ struct Tensor {
 scale(nullptr, {1}, DType::kFloat32),
 scale_inv(nullptr, {1}, DType::kFloat32),
 columnwise_scale_inv(nullptr, {1}, DType::kFloat32),
 scaling_mode(NVTE_DELAYED_TENSOR_SCALING) {}
 scaling_mode(NVTE_DELAYED_TENSOR_SCALING),
 nvte_tensor(0) {}

 void clear() {
 data.clear();
 columnwise_data.clear();
 amax.clear();
 scale.clear();
 scale_inv.clear();
 columnwise_scale_inv.clear();
 scaling_mode = NVTE_DELAYED_TENSOR_SCALING;
 }

 explicit operator NVTETensor() const noexcept { return nvte_tensor; }

 size_t numel() const {
 size_t acc = 1;
Expand DownExpand Up@@ -620,6 +640,8 @@ bool is_supported_by_CC_100();
std::vector<std::vector<Tensor *>> convert_tensor_array(NVTETensor **nvte_tensors,
 size_t outer_size, size_t inner_size);

Tensor *convertNVTETensor(const NVTETensor tensor);
Tensor *convertNVTETensorCheck(const NVTETensor tensor);
} // namespace transformer_engine

#endif // TRANSFORMER_ENGINE_COMMON_COMMON_H_

28 changes: 14 additions & 14 deletions
transformer_engine/common/fused_attn/context_parallel.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -677,9 +677,9 @@ void nvte_cp_thd_read_half_tensor(const NVTETensor &tensor, const NVTETensor &cu
 NVTE_API_CALL(nvte_thd_read_half_tensor);
 using namespace transformer_engine;

 context_parallel::thd_read_half_tensor(*reinterpret_cast<Tensor *>(tensor),
 *reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(half), half_idx, stream);
 context_parallel::thd_read_half_tensor(*convertNVTETensorCheck(tensor),
 *convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(half), half_idx, stream);
}

void nvte_cp_thd_second_half_lse_correction(NVTETensor lse, const NVTETensor &lse_per_step,
Expand All@@ -689,8 +689,8 @@ void nvte_cp_thd_second_half_lse_correction(NVTETensor lse, const NVTETensor &ls
 using namespace transformer_engine;

 context_parallel::thd_second_half_lse_correction(
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(lse_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), lse_packed, stream);
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(lse_per_step),
 *convertNVTETensorCheck(cu_seqlens), lse_packed, stream);
}

void nvte_cp_thd_read_second_half_lse(const NVTETensor &lse, const NVTETensor &cu_seqlens,
Expand All@@ -700,8 +700,8 @@ void nvte_cp_thd_read_second_half_lse(const NVTETensor &lse, const NVTETensor &c
 using namespace transformer_engine;

 context_parallel::thd_read_second_half_lse(
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(half_lse), lse_packed, second_half_lse_seqlen, stream);
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(half_lse), lse_packed, second_half_lse_seqlen, stream);
}

void nvte_cp_thd_out_correction(NVTETensor out, const NVTETensor &out_per_step,
Expand All@@ -712,9 +712,9 @@ void nvte_cp_thd_out_correction(NVTETensor out, const NVTETensor &out_per_step,
 using namespace transformer_engine;

 context_parallel::thd_out_correction(
 *reinterpret_cast<Tensor *>(out), *reinterpret_cast<Tensor *>(out_per_step),
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(lse_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), only_second_half, lse_packed, stream);
 *convertNVTETensorCheck(out), *convertNVTETensorCheck(out_per_step),
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(lse_per_step),
 *convertNVTETensorCheck(cu_seqlens), only_second_half, lse_packed, stream);
}

void nvte_cp_thd_grad_correction(NVTETensor grad, const NVTETensor &grad_per_step,
Expand All@@ -727,8 +727,8 @@ void nvte_cp_thd_grad_correction(NVTETensor grad, const NVTETensor &grad_per_ste
 std::string second_half_str(second_half);

 context_parallel::thd_grad_correction(
 *reinterpret_cast<Tensor *>(grad), *reinterpret_cast<Tensor *>(grad_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), first_half_str, second_half_str, stream);
 *convertNVTETensorCheck(grad), *convertNVTETensorCheck(grad_per_step),
 *convertNVTETensorCheck(cu_seqlens), first_half_str, second_half_str, stream);
}

void nvte_cp_thd_get_partitioned_indices(const NVTETensor &cu_seqlens, NVTETensor output,
Expand All@@ -737,7 +737,7 @@ void nvte_cp_thd_get_partitioned_indices(const NVTETensor &cu_seqlens, NVTETenso
 NVTE_API_CALL(nvte_thd_get_partitioned_indices);
 using namespace transformer_engine;

 context_parallel::thd_get_partitioned_indices(*reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(output), total_tokens,
 context_parallel::thd_get_partitioned_indices(*convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(output), total_tokens,
 world_size, rank, stream);
}

10 changes: 5 additions & 5 deletions
transformer_engine/common/fused_attn/flash_attn.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -138,16 +138,16 @@ void nvte_prepare_flash_attn_fwd(NVTETensor qkvi, NVTETensor qkv, cudaStream_t s
 NVTE_API_CALL(nvte_prepare_flash_attn_fwd);
 using namespace transformer_engine;

 flash_attention::prepare_flash_attn_fwd(*reinterpret_cast<Tensor *>(qkvi),
 *reinterpret_cast<Tensor *>(qkv), stream);
 flash_attention::prepare_flash_attn_fwd(*convertNVTETensorCheck(qkvi),
 *convertNVTETensorCheck(qkv), stream);
}

void nvte_prepare_flash_attn_bwd(NVTETensor q, NVTETensor k, NVTETensor v, NVTETensor qkv,
 cudaStream_t stream) {
 NVTE_API_CALL(nvte_prepare_flash_attn_bwd);
 using namespace transformer_engine;

 flash_attention::prepare_flash_attn_bwd(
*reinterpret_cast<Tensor *>(q), *reinterpret_cast<Tensor *>(k),
*reinterpret_cast<Tensor *>(v), *reinterpret_cast<Tensor *>(qkv), stream);
 flash_attention::prepare_flash_attn_bwd(*convertNVTETensorCheck(q), *convertNVTETensorCheck(k),
 *convertNVTETensorCheck(v), *convertNVTETensorCheck(qkv),
 stream);
}
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Force GitHub README to respect dark mode
(function() {
var style = document.createElement('style');
style.textContent = '
.markdown-body {
color-scheme: dark light;
}
.markdown-body pre { background: #161b22 !important; }
.markdown-body code { background: rgba(110, 118, 129, 0.4) !important; }
.markdown-body table th, .markdown-body table td { border-color: #30363d !important; }
.markdown-body img { background: #0d1117; }
.markdown-body blockquote { border-left-color: #8b949e; }
.markdown-body hr { border-color: #30363d; }
';
document.head.appendChild(style);
})();
}
} catch(__e) { console.warn('[Userscript:GitHub Dark Mode README Fix]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
Avoid memory allocations and deallocations when creating NVTETensor by ptrendx · Pull Request #1813 · NVIDIA/TransformerEngine · GitHub
Skip to content
Avoid memory allocations and deallocations when creating NVTETensor#1813

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

 Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and
privacy statement. We’ll occasionally send you account related emails.

Already on GitHub?
Sign in
to your account

 Merged

ptrendx
merged 19 commits into

from

May 29, 2025
 Merged


Avoid memory allocations and deallocations when creating NVTETensor
#1813
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


19 commits


Select commit
Hold shift + click to select a range
500533b

Changed the Tensor allocation strategy


ptrendx May 19, 2025
c410dc1

Fixes


ptrendx May 20, 2025
7a2b7c9

Disable debug flag


ptrendx May 20, 2025
795c8a7

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 21, 2025
7eb0000

Merge branch 'main' into pr_cpp_avoid_allocs


ptrendx May 27, 2025
cfbb136

Fix the double free error


ptrendx May 27, 2025
8fbe02a

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
56c32e6

Fix


ptrendx May 27, 2025
dd79cc6

Fixed pyTorch recipe extension


ptrendx May 27, 2025
bbfec4a

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
81a33fa

Fix


ptrendx May 27, 2025
13d9937

Fix


ptrendx May 27, 2025
5d6b374

Hide TensorAllocator and fix the usage in LayerNorm


ptrendx May 27, 2025
e7a70bd

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
da5c790

Cleaning


ptrendx May 27, 2025
6596719

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
7d4bc70

Fix


ptrendx May 27, 2025
db7621c

Fix permutation


ptrendx May 28, 2025
c6f4f40

Merge branch 'main' into pr_cpp_avoid_allocs


ptrendx May 28, 2025

File filter

Filter by extension
Filter by extension
Conversations

Failed to load comments. 
Loading
Jump to

Jump to file


Failed to load files. 
Loading
 Diff view

 Diff view


2 changes: 1 addition & 1 deletion
transformer_engine/common/common.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -200,7 +200,7 @@ std::vector<std::vector<Tensor *>> convert_tensor_array(NVTETensor **nvte_tensor
 for (size_t i = 0; i < outer_size; ++i) {
 ret.emplace_back();
 for (size_t j = 0; j < inner_size; ++j) {
 ret.back().push_back(reinterpret_cast<Tensor *>(nvte_tensors[i][j]));
 ret.back().push_back(convertNVTETensor(nvte_tensors[i][j]));
 }
 }
 return ret;
Expand Down

26 changes: 24 additions & 2 deletions
transformer_engine/common/common.h

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -89,18 +89,25 @@ struct SimpleTensor {
 }
 return acc;
 }

 void clear() {
 dptr = nullptr;
 shape.resize(0);
 dtype = DType::kFloat32;
 }
};

struct Tensor {
 public:
 SimpleTensor data;
 SimpleTensor columnwise_data;
 SimpleTensor amax;
 SimpleTensor scale;
 SimpleTensor scale_inv;
 SimpleTensor columnwise_scale_inv;

 public:
 NVTEScalingMode scaling_mode;
 NVTETensor nvte_tensor;

 Tensor()
 : data(),
Expand All@@ -109,7 +116,20 @@ struct Tensor {
 scale(nullptr, {1}, DType::kFloat32),
 scale_inv(nullptr, {1}, DType::kFloat32),
 columnwise_scale_inv(nullptr, {1}, DType::kFloat32),
 scaling_mode(NVTE_DELAYED_TENSOR_SCALING) {}
 scaling_mode(NVTE_DELAYED_TENSOR_SCALING),
 nvte_tensor(0) {}

 void clear() {
 data.clear();
 columnwise_data.clear();
 amax.clear();
 scale.clear();
 scale_inv.clear();
 columnwise_scale_inv.clear();
 scaling_mode = NVTE_DELAYED_TENSOR_SCALING;
 }

 explicit operator NVTETensor() const noexcept { return nvte_tensor; }

 size_t numel() const {
 size_t acc = 1;
Expand DownExpand Up@@ -620,6 +640,8 @@ bool is_supported_by_CC_100();
std::vector<std::vector<Tensor *>> convert_tensor_array(NVTETensor **nvte_tensors,
 size_t outer_size, size_t inner_size);

Tensor *convertNVTETensor(const NVTETensor tensor);
Tensor *convertNVTETensorCheck(const NVTETensor tensor);
} // namespace transformer_engine

#endif // TRANSFORMER_ENGINE_COMMON_COMMON_H_

28 changes: 14 additions & 14 deletions
transformer_engine/common/fused_attn/context_parallel.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -677,9 +677,9 @@ void nvte_cp_thd_read_half_tensor(const NVTETensor &tensor, const NVTETensor &cu
 NVTE_API_CALL(nvte_thd_read_half_tensor);
 using namespace transformer_engine;

 context_parallel::thd_read_half_tensor(*reinterpret_cast<Tensor *>(tensor),
 *reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(half), half_idx, stream);
 context_parallel::thd_read_half_tensor(*convertNVTETensorCheck(tensor),
 *convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(half), half_idx, stream);
}

void nvte_cp_thd_second_half_lse_correction(NVTETensor lse, const NVTETensor &lse_per_step,
Expand All@@ -689,8 +689,8 @@ void nvte_cp_thd_second_half_lse_correction(NVTETensor lse, const NVTETensor &ls
 using namespace transformer_engine;

 context_parallel::thd_second_half_lse_correction(
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(lse_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), lse_packed, stream);
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(lse_per_step),
 *convertNVTETensorCheck(cu_seqlens), lse_packed, stream);
}

void nvte_cp_thd_read_second_half_lse(const NVTETensor &lse, const NVTETensor &cu_seqlens,
Expand All@@ -700,8 +700,8 @@ void nvte_cp_thd_read_second_half_lse(const NVTETensor &lse, const NVTETensor &c
 using namespace transformer_engine;

 context_parallel::thd_read_second_half_lse(
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(half_lse), lse_packed, second_half_lse_seqlen, stream);
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(half_lse), lse_packed, second_half_lse_seqlen, stream);
}

void nvte_cp_thd_out_correction(NVTETensor out, const NVTETensor &out_per_step,
Expand All@@ -712,9 +712,9 @@ void nvte_cp_thd_out_correction(NVTETensor out, const NVTETensor &out_per_step,
 using namespace transformer_engine;

 context_parallel::thd_out_correction(
 *reinterpret_cast<Tensor *>(out), *reinterpret_cast<Tensor *>(out_per_step),
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(lse_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), only_second_half, lse_packed, stream);
 *convertNVTETensorCheck(out), *convertNVTETensorCheck(out_per_step),
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(lse_per_step),
 *convertNVTETensorCheck(cu_seqlens), only_second_half, lse_packed, stream);
}

void nvte_cp_thd_grad_correction(NVTETensor grad, const NVTETensor &grad_per_step,
Expand All@@ -727,8 +727,8 @@ void nvte_cp_thd_grad_correction(NVTETensor grad, const NVTETensor &grad_per_ste
 std::string second_half_str(second_half);

 context_parallel::thd_grad_correction(
 *reinterpret_cast<Tensor *>(grad), *reinterpret_cast<Tensor *>(grad_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), first_half_str, second_half_str, stream);
 *convertNVTETensorCheck(grad), *convertNVTETensorCheck(grad_per_step),
 *convertNVTETensorCheck(cu_seqlens), first_half_str, second_half_str, stream);
}

void nvte_cp_thd_get_partitioned_indices(const NVTETensor &cu_seqlens, NVTETensor output,
Expand All@@ -737,7 +737,7 @@ void nvte_cp_thd_get_partitioned_indices(const NVTETensor &cu_seqlens, NVTETenso
 NVTE_API_CALL(nvte_thd_get_partitioned_indices);
 using namespace transformer_engine;

 context_parallel::thd_get_partitioned_indices(*reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(output), total_tokens,
 context_parallel::thd_get_partitioned_indices(*convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(output), total_tokens,
 world_size, rank, stream);
}

10 changes: 5 additions & 5 deletions
transformer_engine/common/fused_attn/flash_attn.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -138,16 +138,16 @@ void nvte_prepare_flash_attn_fwd(NVTETensor qkvi, NVTETensor qkv, cudaStream_t s
 NVTE_API_CALL(nvte_prepare_flash_attn_fwd);
 using namespace transformer_engine;

 flash_attention::prepare_flash_attn_fwd(*reinterpret_cast<Tensor *>(qkvi),
 *reinterpret_cast<Tensor *>(qkv), stream);
 flash_attention::prepare_flash_attn_fwd(*convertNVTETensorCheck(qkvi),
 *convertNVTETensorCheck(qkv), stream);
}

void nvte_prepare_flash_attn_bwd(NVTETensor q, NVTETensor k, NVTETensor v, NVTETensor qkv,
 cudaStream_t stream) {
 NVTE_API_CALL(nvte_prepare_flash_attn_bwd);
 using namespace transformer_engine;

 flash_attention::prepare_flash_attn_bwd(
*reinterpret_cast<Tensor *>(q), *reinterpret_cast<Tensor *>(k),
*reinterpret_cast<Tensor *>(v), *reinterpret_cast<Tensor *>(qkv), stream);
 flash_attention::prepare_flash_attn_bwd(*convertNVTETensorCheck(q), *convertNVTETensorCheck(k),
 *convertNVTETensorCheck(v), *convertNVTETensorCheck(qkv),
 stream);
}
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Highlight search terms from Google/DuckDuckGo/Bing referrer
(function() {
var ref = document.referrer;
var terms = [];
if (ref.includes('google.com') || ref.includes('duckduckgo.com') || ref.includes('bing.com')) {
var url = new URL(ref);
var q = url.searchParams.get('q') || url.searchParams.get('p');
if (q) {
terms = q.split(/\s+/).filter(function(t) { return t.length > 2; });
}
}
if (terms.length === 0) return;
var style = document.createElement('style');
style.textContent = '.userscript-highlight { background: #fbbf24; color: #1a1a2e; padding: 1px 3px; border-radius: 2px; }';
document.head.appendChild(style);
function highlight(node) {
if (node.nodeType === 3) { // text node
var text = node.textContent;
var found = false;
terms.forEach(function(term) {
var regex = new RegExp('(' + term.replace(/[.*+?^${}()|[\]\\]/g, '\\') + ')', 'gi');
if (regex.test(text)) {
found = true;
var frag = document.createDocumentFragment();
var parts = text.split(regex);
parts.forEach(function(part, i) {
if (i % 2 === 0) {
frag.appendChild(document.createTextNode(part));
} else {
var span = document.createElement('span');
span.className = 'userscript-highlight';
span.textContent = part;
frag.appendChild(span);
}
});
node.parentNode.replaceChild(frag, node);
}
});
} else if (node.nodeType === 1 && node.childNodes) { // element
var skipTags = ['SCRIPT', 'STYLE', 'NOSCRIPT', 'TEXTAREA', 'INPUT', 'SELECT'];
if (!skipTags.includes(node.tagName)) {
Array.from(node.childNodes).forEach(highlight);
}
}
}
highlight(document.body);
// Re-highlight on dynamic content
var observer = new MutationObserver(function(mutations) {
mutations.forEach(function(m) {
m.addedNodes.forEach(function(node) {
if (node.nodeType === 1 || node.nodeType === 3) highlight(node);
});
});
});
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Highlight Search Terms]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
Avoid memory allocations and deallocations when creating NVTETensor by ptrendx · Pull Request #1813 · NVIDIA/TransformerEngine · GitHub
Skip to content
Avoid memory allocations and deallocations when creating NVTETensor#1813

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

 Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and
privacy statement. We’ll occasionally send you account related emails.

Already on GitHub?
Sign in
to your account

 Merged

ptrendx
merged 19 commits into

from

May 29, 2025
 Merged


Avoid memory allocations and deallocations when creating NVTETensor
#1813
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


19 commits


Select commit
Hold shift + click to select a range
500533b

Changed the Tensor allocation strategy


ptrendx May 19, 2025
c410dc1

Fixes


ptrendx May 20, 2025
7a2b7c9

Disable debug flag


ptrendx May 20, 2025
795c8a7

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 21, 2025
7eb0000

Merge branch 'main' into pr_cpp_avoid_allocs


ptrendx May 27, 2025
cfbb136

Fix the double free error


ptrendx May 27, 2025
8fbe02a

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
56c32e6

Fix


ptrendx May 27, 2025
dd79cc6

Fixed pyTorch recipe extension


ptrendx May 27, 2025
bbfec4a

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
81a33fa

Fix


ptrendx May 27, 2025
13d9937

Fix


ptrendx May 27, 2025
5d6b374

Hide TensorAllocator and fix the usage in LayerNorm


ptrendx May 27, 2025
e7a70bd

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
da5c790

Cleaning


ptrendx May 27, 2025
6596719

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
7d4bc70

Fix


ptrendx May 27, 2025
db7621c

Fix permutation


ptrendx May 28, 2025
c6f4f40

Merge branch 'main' into pr_cpp_avoid_allocs


ptrendx May 28, 2025

File filter

Filter by extension
Filter by extension
Conversations

Failed to load comments. 
Loading
Jump to

Jump to file


Failed to load files. 
Loading
 Diff view

 Diff view


2 changes: 1 addition & 1 deletion
transformer_engine/common/common.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -200,7 +200,7 @@ std::vector<std::vector<Tensor *>> convert_tensor_array(NVTETensor **nvte_tensor
 for (size_t i = 0; i < outer_size; ++i) {
 ret.emplace_back();
 for (size_t j = 0; j < inner_size; ++j) {
 ret.back().push_back(reinterpret_cast<Tensor *>(nvte_tensors[i][j]));
 ret.back().push_back(convertNVTETensor(nvte_tensors[i][j]));
 }
 }
 return ret;
Expand Down

26 changes: 24 additions & 2 deletions
transformer_engine/common/common.h

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -89,18 +89,25 @@ struct SimpleTensor {
 }
 return acc;
 }

 void clear() {
 dptr = nullptr;
 shape.resize(0);
 dtype = DType::kFloat32;
 }
};

struct Tensor {
 public:
 SimpleTensor data;
 SimpleTensor columnwise_data;
 SimpleTensor amax;
 SimpleTensor scale;
 SimpleTensor scale_inv;
 SimpleTensor columnwise_scale_inv;

 public:
 NVTEScalingMode scaling_mode;
 NVTETensor nvte_tensor;

 Tensor()
 : data(),
Expand All@@ -109,7 +116,20 @@ struct Tensor {
 scale(nullptr, {1}, DType::kFloat32),
 scale_inv(nullptr, {1}, DType::kFloat32),
 columnwise_scale_inv(nullptr, {1}, DType::kFloat32),
 scaling_mode(NVTE_DELAYED_TENSOR_SCALING) {}
 scaling_mode(NVTE_DELAYED_TENSOR_SCALING),
 nvte_tensor(0) {}

 void clear() {
 data.clear();
 columnwise_data.clear();
 amax.clear();
 scale.clear();
 scale_inv.clear();
 columnwise_scale_inv.clear();
 scaling_mode = NVTE_DELAYED_TENSOR_SCALING;
 }

 explicit operator NVTETensor() const noexcept { return nvte_tensor; }

 size_t numel() const {
 size_t acc = 1;
Expand DownExpand Up@@ -620,6 +640,8 @@ bool is_supported_by_CC_100();
std::vector<std::vector<Tensor *>> convert_tensor_array(NVTETensor **nvte_tensors,
 size_t outer_size, size_t inner_size);

Tensor *convertNVTETensor(const NVTETensor tensor);
Tensor *convertNVTETensorCheck(const NVTETensor tensor);
} // namespace transformer_engine

#endif // TRANSFORMER_ENGINE_COMMON_COMMON_H_

28 changes: 14 additions & 14 deletions
transformer_engine/common/fused_attn/context_parallel.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -677,9 +677,9 @@ void nvte_cp_thd_read_half_tensor(const NVTETensor &tensor, const NVTETensor &cu
 NVTE_API_CALL(nvte_thd_read_half_tensor);
 using namespace transformer_engine;

 context_parallel::thd_read_half_tensor(*reinterpret_cast<Tensor *>(tensor),
 *reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(half), half_idx, stream);
 context_parallel::thd_read_half_tensor(*convertNVTETensorCheck(tensor),
 *convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(half), half_idx, stream);
}

void nvte_cp_thd_second_half_lse_correction(NVTETensor lse, const NVTETensor &lse_per_step,
Expand All@@ -689,8 +689,8 @@ void nvte_cp_thd_second_half_lse_correction(NVTETensor lse, const NVTETensor &ls
 using namespace transformer_engine;

 context_parallel::thd_second_half_lse_correction(
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(lse_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), lse_packed, stream);
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(lse_per_step),
 *convertNVTETensorCheck(cu_seqlens), lse_packed, stream);
}

void nvte_cp_thd_read_second_half_lse(const NVTETensor &lse, const NVTETensor &cu_seqlens,
Expand All@@ -700,8 +700,8 @@ void nvte_cp_thd_read_second_half_lse(const NVTETensor &lse, const NVTETensor &c
 using namespace transformer_engine;

 context_parallel::thd_read_second_half_lse(
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(half_lse), lse_packed, second_half_lse_seqlen, stream);
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(half_lse), lse_packed, second_half_lse_seqlen, stream);
}

void nvte_cp_thd_out_correction(NVTETensor out, const NVTETensor &out_per_step,
Expand All@@ -712,9 +712,9 @@ void nvte_cp_thd_out_correction(NVTETensor out, const NVTETensor &out_per_step,
 using namespace transformer_engine;

 context_parallel::thd_out_correction(
 *reinterpret_cast<Tensor *>(out), *reinterpret_cast<Tensor *>(out_per_step),
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(lse_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), only_second_half, lse_packed, stream);
 *convertNVTETensorCheck(out), *convertNVTETensorCheck(out_per_step),
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(lse_per_step),
 *convertNVTETensorCheck(cu_seqlens), only_second_half, lse_packed, stream);
}

void nvte_cp_thd_grad_correction(NVTETensor grad, const NVTETensor &grad_per_step,
Expand All@@ -727,8 +727,8 @@ void nvte_cp_thd_grad_correction(NVTETensor grad, const NVTETensor &grad_per_ste
 std::string second_half_str(second_half);

 context_parallel::thd_grad_correction(
 *reinterpret_cast<Tensor *>(grad), *reinterpret_cast<Tensor *>(grad_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), first_half_str, second_half_str, stream);
 *convertNVTETensorCheck(grad), *convertNVTETensorCheck(grad_per_step),
 *convertNVTETensorCheck(cu_seqlens), first_half_str, second_half_str, stream);
}

void nvte_cp_thd_get_partitioned_indices(const NVTETensor &cu_seqlens, NVTETensor output,
Expand All@@ -737,7 +737,7 @@ void nvte_cp_thd_get_partitioned_indices(const NVTETensor &cu_seqlens, NVTETenso
 NVTE_API_CALL(nvte_thd_get_partitioned_indices);
 using namespace transformer_engine;

 context_parallel::thd_get_partitioned_indices(*reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(output), total_tokens,
 context_parallel::thd_get_partitioned_indices(*convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(output), total_tokens,
 world_size, rank, stream);
}

10 changes: 5 additions & 5 deletions
transformer_engine/common/fused_attn/flash_attn.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -138,16 +138,16 @@ void nvte_prepare_flash_attn_fwd(NVTETensor qkvi, NVTETensor qkv, cudaStream_t s
 NVTE_API_CALL(nvte_prepare_flash_attn_fwd);
 using namespace transformer_engine;

 flash_attention::prepare_flash_attn_fwd(*reinterpret_cast<Tensor *>(qkvi),
 *reinterpret_cast<Tensor *>(qkv), stream);
 flash_attention::prepare_flash_attn_fwd(*convertNVTETensorCheck(qkvi),
 *convertNVTETensorCheck(qkv), stream);
}

void nvte_prepare_flash_attn_bwd(NVTETensor q, NVTETensor k, NVTETensor v, NVTETensor qkv,
 cudaStream_t stream) {
 NVTE_API_CALL(nvte_prepare_flash_attn_bwd);
 using namespace transformer_engine;

 flash_attention::prepare_flash_attn_bwd(
*reinterpret_cast<Tensor *>(q), *reinterpret_cast<Tensor *>(k),
*reinterpret_cast<Tensor *>(v), *reinterpret_cast<Tensor *>(qkv), stream);
 flash_attention::prepare_flash_attn_bwd(*convertNVTETensorCheck(q), *convertNVTETensorCheck(k),
 *convertNVTETensorCheck(v), *convertNVTETensorCheck(qkv),
 stream);
}
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Strip utm_, fbclid, gclid, etc. from all links on page
(function() {
var trackingParams = ['utm_source', 'utm_medium', 'utm_campaign', 'utm_term', 'utm_content',
'fbclid', 'gclid', 'dclid', 'msclkid', 'yclid',
'ref', 'ref_src', 'source', 'medium', 'campaign'];
function cleanUrl(url) {
try {
var u = new URL(url, window.location.origin);
var changed = false;
trackingParams.forEach(function(p) {
if (u.searchParams.has(p)) {
u.searchParams.delete(p);
changed = true;
}
});
return changed ? u.toString() : url;
} catch (e) {
return url;
}
}
function cleanLinks() {
document.querySelectorAll('a[href]').forEach(function(a) {
var clean = cleanUrl(a.href);
if (clean !== a.href) a.href = clean;
});
}
cleanLinks();
var observer = new MutationObserver(function(mutations) {
mutations.forEach(function(m) {
m.addedNodes.forEach(function(node) {
if (node.nodeType === 1) {
if (node.tagName === 'A') cleanLinks();
node.querySelectorAll('a[href]').forEach(function(a) {
var clean = cleanUrl(a.href);
if (clean !== a.href) a.href = clean;
});
}
});
});
});
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Remove Tracking Parameters from Links]', __e); }
})();
(function(){
try {
var __m = "youtube.com";
var __re = new RegExp('^' + "youtube\\.com" + '
Avoid memory allocations and deallocations when creating NVTETensor by ptrendx · Pull Request #1813 · NVIDIA/TransformerEngine · GitHub
Skip to content
Avoid memory allocations and deallocations when creating NVTETensor#1813

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

 Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and
privacy statement. We’ll occasionally send you account related emails.

Already on GitHub?
Sign in
to your account

 Merged

ptrendx
merged 19 commits into

from

May 29, 2025
 Merged


Avoid memory allocations and deallocations when creating NVTETensor
#1813
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


19 commits


Select commit
Hold shift + click to select a range
500533b

Changed the Tensor allocation strategy


ptrendx May 19, 2025
c410dc1

Fixes


ptrendx May 20, 2025
7a2b7c9

Disable debug flag


ptrendx May 20, 2025
795c8a7

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 21, 2025
7eb0000

Merge branch 'main' into pr_cpp_avoid_allocs


ptrendx May 27, 2025
cfbb136

Fix the double free error


ptrendx May 27, 2025
8fbe02a

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
56c32e6

Fix


ptrendx May 27, 2025
dd79cc6

Fixed pyTorch recipe extension


ptrendx May 27, 2025
bbfec4a

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
81a33fa

Fix


ptrendx May 27, 2025
13d9937

Fix


ptrendx May 27, 2025
5d6b374

Hide TensorAllocator and fix the usage in LayerNorm


ptrendx May 27, 2025
e7a70bd

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
da5c790

Cleaning


ptrendx May 27, 2025
6596719

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
7d4bc70

Fix


ptrendx May 27, 2025
db7621c

Fix permutation


ptrendx May 28, 2025
c6f4f40

Merge branch 'main' into pr_cpp_avoid_allocs


ptrendx May 28, 2025

File filter

Filter by extension
Filter by extension
Conversations

Failed to load comments. 
Loading
Jump to

Jump to file


Failed to load files. 
Loading
 Diff view

 Diff view


2 changes: 1 addition & 1 deletion
transformer_engine/common/common.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -200,7 +200,7 @@ std::vector<std::vector<Tensor *>> convert_tensor_array(NVTETensor **nvte_tensor
 for (size_t i = 0; i < outer_size; ++i) {
 ret.emplace_back();
 for (size_t j = 0; j < inner_size; ++j) {
 ret.back().push_back(reinterpret_cast<Tensor *>(nvte_tensors[i][j]));
 ret.back().push_back(convertNVTETensor(nvte_tensors[i][j]));
 }
 }
 return ret;
Expand Down

26 changes: 24 additions & 2 deletions
transformer_engine/common/common.h

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -89,18 +89,25 @@ struct SimpleTensor {
 }
 return acc;
 }

 void clear() {
 dptr = nullptr;
 shape.resize(0);
 dtype = DType::kFloat32;
 }
};

struct Tensor {
 public:
 SimpleTensor data;
 SimpleTensor columnwise_data;
 SimpleTensor amax;
 SimpleTensor scale;
 SimpleTensor scale_inv;
 SimpleTensor columnwise_scale_inv;

 public:
 NVTEScalingMode scaling_mode;
 NVTETensor nvte_tensor;

 Tensor()
 : data(),
Expand All@@ -109,7 +116,20 @@ struct Tensor {
 scale(nullptr, {1}, DType::kFloat32),
 scale_inv(nullptr, {1}, DType::kFloat32),
 columnwise_scale_inv(nullptr, {1}, DType::kFloat32),
 scaling_mode(NVTE_DELAYED_TENSOR_SCALING) {}
 scaling_mode(NVTE_DELAYED_TENSOR_SCALING),
 nvte_tensor(0) {}

 void clear() {
 data.clear();
 columnwise_data.clear();
 amax.clear();
 scale.clear();
 scale_inv.clear();
 columnwise_scale_inv.clear();
 scaling_mode = NVTE_DELAYED_TENSOR_SCALING;
 }

 explicit operator NVTETensor() const noexcept { return nvte_tensor; }

 size_t numel() const {
 size_t acc = 1;
Expand DownExpand Up@@ -620,6 +640,8 @@ bool is_supported_by_CC_100();
std::vector<std::vector<Tensor *>> convert_tensor_array(NVTETensor **nvte_tensors,
 size_t outer_size, size_t inner_size);

Tensor *convertNVTETensor(const NVTETensor tensor);
Tensor *convertNVTETensorCheck(const NVTETensor tensor);
} // namespace transformer_engine

#endif // TRANSFORMER_ENGINE_COMMON_COMMON_H_

28 changes: 14 additions & 14 deletions
transformer_engine/common/fused_attn/context_parallel.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -677,9 +677,9 @@ void nvte_cp_thd_read_half_tensor(const NVTETensor &tensor, const NVTETensor &cu
 NVTE_API_CALL(nvte_thd_read_half_tensor);
 using namespace transformer_engine;

 context_parallel::thd_read_half_tensor(*reinterpret_cast<Tensor *>(tensor),
 *reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(half), half_idx, stream);
 context_parallel::thd_read_half_tensor(*convertNVTETensorCheck(tensor),
 *convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(half), half_idx, stream);
}

void nvte_cp_thd_second_half_lse_correction(NVTETensor lse, const NVTETensor &lse_per_step,
Expand All@@ -689,8 +689,8 @@ void nvte_cp_thd_second_half_lse_correction(NVTETensor lse, const NVTETensor &ls
 using namespace transformer_engine;

 context_parallel::thd_second_half_lse_correction(
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(lse_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), lse_packed, stream);
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(lse_per_step),
 *convertNVTETensorCheck(cu_seqlens), lse_packed, stream);
}

void nvte_cp_thd_read_second_half_lse(const NVTETensor &lse, const NVTETensor &cu_seqlens,
Expand All@@ -700,8 +700,8 @@ void nvte_cp_thd_read_second_half_lse(const NVTETensor &lse, const NVTETensor &c
 using namespace transformer_engine;

 context_parallel::thd_read_second_half_lse(
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(half_lse), lse_packed, second_half_lse_seqlen, stream);
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(half_lse), lse_packed, second_half_lse_seqlen, stream);
}

void nvte_cp_thd_out_correction(NVTETensor out, const NVTETensor &out_per_step,
Expand All@@ -712,9 +712,9 @@ void nvte_cp_thd_out_correction(NVTETensor out, const NVTETensor &out_per_step,
 using namespace transformer_engine;

 context_parallel::thd_out_correction(
 *reinterpret_cast<Tensor *>(out), *reinterpret_cast<Tensor *>(out_per_step),
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(lse_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), only_second_half, lse_packed, stream);
 *convertNVTETensorCheck(out), *convertNVTETensorCheck(out_per_step),
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(lse_per_step),
 *convertNVTETensorCheck(cu_seqlens), only_second_half, lse_packed, stream);
}

void nvte_cp_thd_grad_correction(NVTETensor grad, const NVTETensor &grad_per_step,
Expand All@@ -727,8 +727,8 @@ void nvte_cp_thd_grad_correction(NVTETensor grad, const NVTETensor &grad_per_ste
 std::string second_half_str(second_half);

 context_parallel::thd_grad_correction(
 *reinterpret_cast<Tensor *>(grad), *reinterpret_cast<Tensor *>(grad_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), first_half_str, second_half_str, stream);
 *convertNVTETensorCheck(grad), *convertNVTETensorCheck(grad_per_step),
 *convertNVTETensorCheck(cu_seqlens), first_half_str, second_half_str, stream);
}

void nvte_cp_thd_get_partitioned_indices(const NVTETensor &cu_seqlens, NVTETensor output,
Expand All@@ -737,7 +737,7 @@ void nvte_cp_thd_get_partitioned_indices(const NVTETensor &cu_seqlens, NVTETenso
 NVTE_API_CALL(nvte_thd_get_partitioned_indices);
 using namespace transformer_engine;

 context_parallel::thd_get_partitioned_indices(*reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(output), total_tokens,
 context_parallel::thd_get_partitioned_indices(*convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(output), total_tokens,
 world_size, rank, stream);
}

10 changes: 5 additions & 5 deletions
transformer_engine/common/fused_attn/flash_attn.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -138,16 +138,16 @@ void nvte_prepare_flash_attn_fwd(NVTETensor qkvi, NVTETensor qkv, cudaStream_t s
 NVTE_API_CALL(nvte_prepare_flash_attn_fwd);
 using namespace transformer_engine;

 flash_attention::prepare_flash_attn_fwd(*reinterpret_cast<Tensor *>(qkvi),
 *reinterpret_cast<Tensor *>(qkv), stream);
 flash_attention::prepare_flash_attn_fwd(*convertNVTETensorCheck(qkvi),
 *convertNVTETensorCheck(qkv), stream);
}

void nvte_prepare_flash_attn_bwd(NVTETensor q, NVTETensor k, NVTETensor v, NVTETensor qkv,
 cudaStream_t stream) {
 NVTE_API_CALL(nvte_prepare_flash_attn_bwd);
 using namespace transformer_engine;

 flash_attention::prepare_flash_attn_bwd(
*reinterpret_cast<Tensor *>(q), *reinterpret_cast<Tensor *>(k),
*reinterpret_cast<Tensor *>(v), *reinterpret_cast<Tensor *>(qkv), stream);
 flash_attention::prepare_flash_attn_bwd(*convertNVTETensorCheck(q), *convertNVTETensorCheck(k),
 *convertNVTETensorCheck(v), *convertNVTETensorCheck(qkv),
 stream);
}
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Auto-enable theater mode on YouTube
(function() {
function tryTheater() {
var btn = document.querySelector('button[aria-label="Theater mode"], ytd-player #player button[title="Theater mode"]');
if (btn && !btn.classList.contains('activated')) {
btn.click();
}
}
// Try immediately
tryTheater();
// Try after navigation (SPA)
var lastUrl = location.href;
setInterval(function() {
if (location.href !== lastUrl) {
lastUrl = location.href;
setTimeout(tryTheater, 500);
}
}, 1000);
// Also try on player load
var observer = new MutationObserver(tryTheater);
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:YouTube Theater Mode Default]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
Avoid memory allocations and deallocations when creating NVTETensor by ptrendx · Pull Request #1813 · NVIDIA/TransformerEngine · GitHub
Skip to content
Avoid memory allocations and deallocations when creating NVTETensor#1813

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

 Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and
privacy statement. We’ll occasionally send you account related emails.

Already on GitHub?
Sign in
to your account

 Merged

ptrendx
merged 19 commits into

from

May 29, 2025
 Merged


Avoid memory allocations and deallocations when creating NVTETensor
#1813
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


19 commits


Select commit
Hold shift + click to select a range
500533b

Changed the Tensor allocation strategy


ptrendx May 19, 2025
c410dc1

Fixes


ptrendx May 20, 2025
7a2b7c9

Disable debug flag


ptrendx May 20, 2025
795c8a7

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 21, 2025
7eb0000

Merge branch 'main' into pr_cpp_avoid_allocs


ptrendx May 27, 2025
cfbb136

Fix the double free error


ptrendx May 27, 2025
8fbe02a

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
56c32e6

Fix


ptrendx May 27, 2025
dd79cc6

Fixed pyTorch recipe extension


ptrendx May 27, 2025
bbfec4a

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
81a33fa

Fix


ptrendx May 27, 2025
13d9937

Fix


ptrendx May 27, 2025
5d6b374

Hide TensorAllocator and fix the usage in LayerNorm


ptrendx May 27, 2025
e7a70bd

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
da5c790

Cleaning


ptrendx May 27, 2025
6596719

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
7d4bc70

Fix


ptrendx May 27, 2025
db7621c

Fix permutation


ptrendx May 28, 2025
c6f4f40

Merge branch 'main' into pr_cpp_avoid_allocs


ptrendx May 28, 2025

File filter

Filter by extension
Filter by extension
Conversations

Failed to load comments. 
Loading
Jump to

Jump to file


Failed to load files. 
Loading
 Diff view

 Diff view


2 changes: 1 addition & 1 deletion
transformer_engine/common/common.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -200,7 +200,7 @@ std::vector<std::vector<Tensor *>> convert_tensor_array(NVTETensor **nvte_tensor
 for (size_t i = 0; i < outer_size; ++i) {
 ret.emplace_back();
 for (size_t j = 0; j < inner_size; ++j) {
 ret.back().push_back(reinterpret_cast<Tensor *>(nvte_tensors[i][j]));
 ret.back().push_back(convertNVTETensor(nvte_tensors[i][j]));
 }
 }
 return ret;
Expand Down

26 changes: 24 additions & 2 deletions
transformer_engine/common/common.h

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -89,18 +89,25 @@ struct SimpleTensor {
 }
 return acc;
 }

 void clear() {
 dptr = nullptr;
 shape.resize(0);
 dtype = DType::kFloat32;
 }
};

struct Tensor {
 public:
 SimpleTensor data;
 SimpleTensor columnwise_data;
 SimpleTensor amax;
 SimpleTensor scale;
 SimpleTensor scale_inv;
 SimpleTensor columnwise_scale_inv;

 public:
 NVTEScalingMode scaling_mode;
 NVTETensor nvte_tensor;

 Tensor()
 : data(),
Expand All@@ -109,7 +116,20 @@ struct Tensor {
 scale(nullptr, {1}, DType::kFloat32),
 scale_inv(nullptr, {1}, DType::kFloat32),
 columnwise_scale_inv(nullptr, {1}, DType::kFloat32),
 scaling_mode(NVTE_DELAYED_TENSOR_SCALING) {}
 scaling_mode(NVTE_DELAYED_TENSOR_SCALING),
 nvte_tensor(0) {}

 void clear() {
 data.clear();
 columnwise_data.clear();
 amax.clear();
 scale.clear();
 scale_inv.clear();
 columnwise_scale_inv.clear();
 scaling_mode = NVTE_DELAYED_TENSOR_SCALING;
 }

 explicit operator NVTETensor() const noexcept { return nvte_tensor; }

 size_t numel() const {
 size_t acc = 1;
Expand DownExpand Up@@ -620,6 +640,8 @@ bool is_supported_by_CC_100();
std::vector<std::vector<Tensor *>> convert_tensor_array(NVTETensor **nvte_tensors,
 size_t outer_size, size_t inner_size);

Tensor *convertNVTETensor(const NVTETensor tensor);
Tensor *convertNVTETensorCheck(const NVTETensor tensor);
} // namespace transformer_engine

#endif // TRANSFORMER_ENGINE_COMMON_COMMON_H_

28 changes: 14 additions & 14 deletions
transformer_engine/common/fused_attn/context_parallel.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -677,9 +677,9 @@ void nvte_cp_thd_read_half_tensor(const NVTETensor &tensor, const NVTETensor &cu
 NVTE_API_CALL(nvte_thd_read_half_tensor);
 using namespace transformer_engine;

 context_parallel::thd_read_half_tensor(*reinterpret_cast<Tensor *>(tensor),
 *reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(half), half_idx, stream);
 context_parallel::thd_read_half_tensor(*convertNVTETensorCheck(tensor),
 *convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(half), half_idx, stream);
}

void nvte_cp_thd_second_half_lse_correction(NVTETensor lse, const NVTETensor &lse_per_step,
Expand All@@ -689,8 +689,8 @@ void nvte_cp_thd_second_half_lse_correction(NVTETensor lse, const NVTETensor &ls
 using namespace transformer_engine;

 context_parallel::thd_second_half_lse_correction(
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(lse_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), lse_packed, stream);
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(lse_per_step),
 *convertNVTETensorCheck(cu_seqlens), lse_packed, stream);
}

void nvte_cp_thd_read_second_half_lse(const NVTETensor &lse, const NVTETensor &cu_seqlens,
Expand All@@ -700,8 +700,8 @@ void nvte_cp_thd_read_second_half_lse(const NVTETensor &lse, const NVTETensor &c
 using namespace transformer_engine;

 context_parallel::thd_read_second_half_lse(
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(half_lse), lse_packed, second_half_lse_seqlen, stream);
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(half_lse), lse_packed, second_half_lse_seqlen, stream);
}

void nvte_cp_thd_out_correction(NVTETensor out, const NVTETensor &out_per_step,
Expand All@@ -712,9 +712,9 @@ void nvte_cp_thd_out_correction(NVTETensor out, const NVTETensor &out_per_step,
 using namespace transformer_engine;

 context_parallel::thd_out_correction(
 *reinterpret_cast<Tensor *>(out), *reinterpret_cast<Tensor *>(out_per_step),
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(lse_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), only_second_half, lse_packed, stream);
 *convertNVTETensorCheck(out), *convertNVTETensorCheck(out_per_step),
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(lse_per_step),
 *convertNVTETensorCheck(cu_seqlens), only_second_half, lse_packed, stream);
}

void nvte_cp_thd_grad_correction(NVTETensor grad, const NVTETensor &grad_per_step,
Expand All@@ -727,8 +727,8 @@ void nvte_cp_thd_grad_correction(NVTETensor grad, const NVTETensor &grad_per_ste
 std::string second_half_str(second_half);

 context_parallel::thd_grad_correction(
 *reinterpret_cast<Tensor *>(grad), *reinterpret_cast<Tensor *>(grad_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), first_half_str, second_half_str, stream);
 *convertNVTETensorCheck(grad), *convertNVTETensorCheck(grad_per_step),
 *convertNVTETensorCheck(cu_seqlens), first_half_str, second_half_str, stream);
}

void nvte_cp_thd_get_partitioned_indices(const NVTETensor &cu_seqlens, NVTETensor output,
Expand All@@ -737,7 +737,7 @@ void nvte_cp_thd_get_partitioned_indices(const NVTETensor &cu_seqlens, NVTETenso
 NVTE_API_CALL(nvte_thd_get_partitioned_indices);
 using namespace transformer_engine;

 context_parallel::thd_get_partitioned_indices(*reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(output), total_tokens,
 context_parallel::thd_get_partitioned_indices(*convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(output), total_tokens,
 world_size, rank, stream);
}

10 changes: 5 additions & 5 deletions
transformer_engine/common/fused_attn/flash_attn.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -138,16 +138,16 @@ void nvte_prepare_flash_attn_fwd(NVTETensor qkvi, NVTETensor qkv, cudaStream_t s
 NVTE_API_CALL(nvte_prepare_flash_attn_fwd);
 using namespace transformer_engine;

 flash_attention::prepare_flash_attn_fwd(*reinterpret_cast<Tensor *>(qkvi),
 *reinterpret_cast<Tensor *>(qkv), stream);
 flash_attention::prepare_flash_attn_fwd(*convertNVTETensorCheck(qkvi),
 *convertNVTETensorCheck(qkv), stream);
}

void nvte_prepare_flash_attn_bwd(NVTETensor q, NVTETensor k, NVTETensor v, NVTETensor qkv,
 cudaStream_t stream) {
 NVTE_API_CALL(nvte_prepare_flash_attn_bwd);
 using namespace transformer_engine;

 flash_attention::prepare_flash_attn_bwd(
*reinterpret_cast<Tensor *>(q), *reinterpret_cast<Tensor *>(k),
*reinterpret_cast<Tensor *>(v), *reinterpret_cast<Tensor *>(qkv), stream);
 flash_attention::prepare_flash_attn_bwd(*convertNVTETensorCheck(q), *convertNVTETensorCheck(k),
 *convertNVTETensorCheck(v), *convertNVTETensorCheck(qkv),
 stream);
}
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Remove or un-stick sticky/fixed headers that block content
(function() {
function unstick() {
document.querySelectorAll('header, nav, [role="banner"], .header, .navbar, .sticky, .fixed-top, [style*="position: fixed"], [style*="position:sticky"]').forEach(function(el) {
if (el.style.position === 'fixed' || el.style.position === 'sticky' || getComputedStyle(el).position === 'fixed' || getComputedStyle(el).position === 'sticky') {
el.style.position = 'static';
el.style.top = 'auto';
el.style.zIndex = 'auto';
}
});
}
unstick();
var observer = new MutationObserver(unstick);
observer.observe(document.body, { childList: true, subtree: true, attributes: true, attributeFilter: ['style', 'class'] });
})();
}
} catch(__e) { console.warn('[Userscript:Kill Sticky Headers]', __e); }
})();
})();
Avoid memory allocations and deallocations when creating NVTETensor by ptrendx · Pull Request #1813 · NVIDIA/TransformerEngine · GitHub
Skip to content
Avoid memory allocations and deallocations when creating NVTETensor#1813

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

 Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and
privacy statement. We’ll occasionally send you account related emails.

Already on GitHub?
Sign in
to your account

 Merged

ptrendx
merged 19 commits into

from

May 29, 2025
 Merged


Avoid memory allocations and deallocations when creating NVTETensor
#1813
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


19 commits


Select commit
Hold shift + click to select a range
500533b

Changed the Tensor allocation strategy


ptrendx May 19, 2025
c410dc1

Fixes


ptrendx May 20, 2025
7a2b7c9

Disable debug flag


ptrendx May 20, 2025
795c8a7

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 21, 2025
7eb0000

Merge branch 'main' into pr_cpp_avoid_allocs


ptrendx May 27, 2025
cfbb136

Fix the double free error


ptrendx May 27, 2025
8fbe02a

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
56c32e6

Fix


ptrendx May 27, 2025
dd79cc6

Fixed pyTorch recipe extension


ptrendx May 27, 2025
bbfec4a

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
81a33fa

Fix


ptrendx May 27, 2025
13d9937

Fix


ptrendx May 27, 2025
5d6b374

Hide TensorAllocator and fix the usage in LayerNorm


ptrendx May 27, 2025
e7a70bd

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
da5c790

Cleaning


ptrendx May 27, 2025
6596719

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] May 27, 2025
7d4bc70

Fix


ptrendx May 27, 2025
db7621c

Fix permutation


ptrendx May 28, 2025
c6f4f40

Merge branch 'main' into pr_cpp_avoid_allocs


ptrendx May 28, 2025

File filter

Filter by extension
Filter by extension
Conversations

Failed to load comments. 
Loading
Jump to

Jump to file


Failed to load files. 
Loading
 Diff view

 Diff view


2 changes: 1 addition & 1 deletion
transformer_engine/common/common.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -200,7 +200,7 @@ std::vector<std::vector<Tensor *>> convert_tensor_array(NVTETensor **nvte_tensor
 for (size_t i = 0; i < outer_size; ++i) {
 ret.emplace_back();
 for (size_t j = 0; j < inner_size; ++j) {
 ret.back().push_back(reinterpret_cast<Tensor *>(nvte_tensors[i][j]));
 ret.back().push_back(convertNVTETensor(nvte_tensors[i][j]));
 }
 }
 return ret;
Expand Down

26 changes: 24 additions & 2 deletions
transformer_engine/common/common.h

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -89,18 +89,25 @@ struct SimpleTensor {
 }
 return acc;
 }

 void clear() {
 dptr = nullptr;
 shape.resize(0);
 dtype = DType::kFloat32;
 }
};

struct Tensor {
 public:
 SimpleTensor data;
 SimpleTensor columnwise_data;
 SimpleTensor amax;
 SimpleTensor scale;
 SimpleTensor scale_inv;
 SimpleTensor columnwise_scale_inv;

 public:
 NVTEScalingMode scaling_mode;
 NVTETensor nvte_tensor;

 Tensor()
 : data(),
Expand All@@ -109,7 +116,20 @@ struct Tensor {
 scale(nullptr, {1}, DType::kFloat32),
 scale_inv(nullptr, {1}, DType::kFloat32),
 columnwise_scale_inv(nullptr, {1}, DType::kFloat32),
 scaling_mode(NVTE_DELAYED_TENSOR_SCALING) {}
 scaling_mode(NVTE_DELAYED_TENSOR_SCALING),
 nvte_tensor(0) {}

 void clear() {
 data.clear();
 columnwise_data.clear();
 amax.clear();
 scale.clear();
 scale_inv.clear();
 columnwise_scale_inv.clear();
 scaling_mode = NVTE_DELAYED_TENSOR_SCALING;
 }

 explicit operator NVTETensor() const noexcept { return nvte_tensor; }

 size_t numel() const {
 size_t acc = 1;
Expand DownExpand Up@@ -620,6 +640,8 @@ bool is_supported_by_CC_100();
std::vector<std::vector<Tensor *>> convert_tensor_array(NVTETensor **nvte_tensors,
 size_t outer_size, size_t inner_size);

Tensor *convertNVTETensor(const NVTETensor tensor);
Tensor *convertNVTETensorCheck(const NVTETensor tensor);
} // namespace transformer_engine

#endif // TRANSFORMER_ENGINE_COMMON_COMMON_H_

28 changes: 14 additions & 14 deletions
transformer_engine/common/fused_attn/context_parallel.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -677,9 +677,9 @@ void nvte_cp_thd_read_half_tensor(const NVTETensor &tensor, const NVTETensor &cu
 NVTE_API_CALL(nvte_thd_read_half_tensor);
 using namespace transformer_engine;

 context_parallel::thd_read_half_tensor(*reinterpret_cast<Tensor *>(tensor),
 *reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(half), half_idx, stream);
 context_parallel::thd_read_half_tensor(*convertNVTETensorCheck(tensor),
 *convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(half), half_idx, stream);
}

void nvte_cp_thd_second_half_lse_correction(NVTETensor lse, const NVTETensor &lse_per_step,
Expand All@@ -689,8 +689,8 @@ void nvte_cp_thd_second_half_lse_correction(NVTETensor lse, const NVTETensor &ls
 using namespace transformer_engine;

 context_parallel::thd_second_half_lse_correction(
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(lse_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), lse_packed, stream);
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(lse_per_step),
 *convertNVTETensorCheck(cu_seqlens), lse_packed, stream);
}

void nvte_cp_thd_read_second_half_lse(const NVTETensor &lse, const NVTETensor &cu_seqlens,
Expand All@@ -700,8 +700,8 @@ void nvte_cp_thd_read_second_half_lse(const NVTETensor &lse, const NVTETensor &c
 using namespace transformer_engine;

 context_parallel::thd_read_second_half_lse(
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(half_lse), lse_packed, second_half_lse_seqlen, stream);
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(half_lse), lse_packed, second_half_lse_seqlen, stream);
}

void nvte_cp_thd_out_correction(NVTETensor out, const NVTETensor &out_per_step,
Expand All@@ -712,9 +712,9 @@ void nvte_cp_thd_out_correction(NVTETensor out, const NVTETensor &out_per_step,
 using namespace transformer_engine;

 context_parallel::thd_out_correction(
 *reinterpret_cast<Tensor *>(out), *reinterpret_cast<Tensor *>(out_per_step),
 *reinterpret_cast<Tensor *>(lse), *reinterpret_cast<Tensor *>(lse_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), only_second_half, lse_packed, stream);
 *convertNVTETensorCheck(out), *convertNVTETensorCheck(out_per_step),
 *convertNVTETensorCheck(lse), *convertNVTETensorCheck(lse_per_step),
 *convertNVTETensorCheck(cu_seqlens), only_second_half, lse_packed, stream);
}

void nvte_cp_thd_grad_correction(NVTETensor grad, const NVTETensor &grad_per_step,
Expand All@@ -727,8 +727,8 @@ void nvte_cp_thd_grad_correction(NVTETensor grad, const NVTETensor &grad_per_ste
 std::string second_half_str(second_half);

 context_parallel::thd_grad_correction(
 *reinterpret_cast<Tensor *>(grad), *reinterpret_cast<Tensor *>(grad_per_step),
 *reinterpret_cast<Tensor *>(cu_seqlens), first_half_str, second_half_str, stream);
 *convertNVTETensorCheck(grad), *convertNVTETensorCheck(grad_per_step),
 *convertNVTETensorCheck(cu_seqlens), first_half_str, second_half_str, stream);
}

void nvte_cp_thd_get_partitioned_indices(const NVTETensor &cu_seqlens, NVTETensor output,
Expand All@@ -737,7 +737,7 @@ void nvte_cp_thd_get_partitioned_indices(const NVTETensor &cu_seqlens, NVTETenso
 NVTE_API_CALL(nvte_thd_get_partitioned_indices);
 using namespace transformer_engine;

 context_parallel::thd_get_partitioned_indices(*reinterpret_cast<Tensor *>(cu_seqlens),
 *reinterpret_cast<Tensor *>(output), total_tokens,
 context_parallel::thd_get_partitioned_indices(*convertNVTETensorCheck(cu_seqlens),
 *convertNVTETensorCheck(output), total_tokens,
 world_size, rank, stream);
}

10 changes: 5 additions & 5 deletions
transformer_engine/common/fused_attn/flash_attn.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -138,16 +138,16 @@ void nvte_prepare_flash_attn_fwd(NVTETensor qkvi, NVTETensor qkv, cudaStream_t s
 NVTE_API_CALL(nvte_prepare_flash_attn_fwd);
 using namespace transformer_engine;

 flash_attention::prepare_flash_attn_fwd(*reinterpret_cast<Tensor *>(qkvi),
 *reinterpret_cast<Tensor *>(qkv), stream);
 flash_attention::prepare_flash_attn_fwd(*convertNVTETensorCheck(qkvi),
 *convertNVTETensorCheck(qkv), stream);
}

void nvte_prepare_flash_attn_bwd(NVTETensor q, NVTETensor k, NVTETensor v, NVTETensor qkv,
 cudaStream_t stream) {
 NVTE_API_CALL(nvte_prepare_flash_attn_bwd);
 using namespace transformer_engine;

 flash_attention::prepare_flash_attn_bwd(
*reinterpret_cast<Tensor *>(q), *reinterpret_cast<Tensor *>(k),
*reinterpret_cast<Tensor *>(v), *reinterpret_cast<Tensor *>(qkv), stream);
 flash_attention::prepare_flash_attn_bwd(*convertNVTETensorCheck(q), *convertNVTETensorCheck(k),
 *convertNVTETensorCheck(v), *convertNVTETensorCheck(qkv),
 stream);
}
Loading