-
Notifications
You must be signed in to change notification settings - Fork 520
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Browse files
Browse the repository at this point in the history
Summary: Pull Request resolved: #3517 X-link: facebookresearch/FBGEMM#598 - Cherry-pick from CK PR ROCm/composable_kernel#1636 - Improve fp8 GEMM rowwise for 70B Prefill with seqlen = 1k/2k Reviewed By: xw285cornell, jwfromm Differential Revision: D67418190 fbshipit-source-id: b6d38715b26d91d6047d03941610fa7e20e54cb7
- Loading branch information
1 parent
7c26339
commit dd70f61
Showing
4 changed files
with
160 additions
and
4 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
70 changes: 70 additions & 0 deletions
70
...s/fp8_rowwise_256x256x128x128_32x32_4x2_8x32x1_8x32x1_1x32x1x8_8x8x1_1x1_intrawave_v3.hip
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,70 @@ | ||
| /* | ||
| * Copyright (c) Meta Platforms, Inc. and affiliates. | ||
| * All rights reserved. | ||
| * | ||
| * This source code is licensed under the BSD-style license found in the | ||
| * LICENSE file in the root directory of this source tree. | ||
| */ | ||
|
|
||
| #include "fp8_rowwise_common.h" | ||
|
|
||
| at::Tensor | ||
| fp8_rowwise_256x256x128x128_32x32_4x2_8x32x1_8x32x1_1x32x1x8_8x8x1_1x1_intrawave_v3( | ||
| at::Tensor XQ, | ||
| at::Tensor WQ, | ||
| at::Tensor x_scale, | ||
| at::Tensor w_scale, | ||
| at::Tensor Y) { | ||
| // Check if this input needs to be padded. | ||
| int M = size_to_dim_(XQ.dim() - 1, XQ.sizes()); | ||
| int N = WQ.size(0); | ||
| int K = WQ.size(1); | ||
| bool pad = (K % 128 != 0); | ||
|
|
||
| // Dispatch based on whether padding is needed or not. | ||
| if (pad) { | ||
| using DeviceGemmInstance = DeviceGemmHelper< | ||
| 256, | ||
| 256, | ||
| 128, | ||
| 128, | ||
| 32, | ||
| 32, | ||
| 4, | ||
| 2, | ||
| S<8, 32, 1>, | ||
| S<8, 32, 1>, | ||
| S<1, 32, 1, 8>, | ||
| S<8, 8, 1>, | ||
| 1, | ||
| 1, | ||
| ck::BlockGemmPipelineScheduler::Intrawave, | ||
| ck::BlockGemmPipelineVersion::v3, | ||
| ck::tensor_operation::device::GemmSpecialization::KPadding>; | ||
| // Run kernel instance. | ||
| return f8f8bf16_rowwise_impl<DeviceGemmInstance>( | ||
| XQ, WQ, x_scale, w_scale, Y); | ||
| } else { | ||
| using DeviceGemmInstance = DeviceGemmHelper< | ||
| 256, | ||
| 256, | ||
| 128, | ||
| 128, | ||
| 32, | ||
| 32, | ||
| 4, | ||
| 2, | ||
| S<8, 32, 1>, | ||
| S<8, 32, 1>, | ||
| S<1, 32, 1, 8>, | ||
| S<8, 8, 1>, | ||
| 1, | ||
| 1, | ||
| ck::BlockGemmPipelineScheduler::Intrawave, | ||
| ck::BlockGemmPipelineVersion::v3, | ||
| ck::tensor_operation::device::GemmSpecialization::Default>; | ||
| // Run kernel instance. | ||
| return f8f8bf16_rowwise_impl<DeviceGemmInstance>( | ||
| XQ, WQ, x_scale, w_scale, Y); | ||
| } | ||
| } |
70 changes: 70 additions & 0 deletions
70
...ls/fp8_rowwise_256x256x96x128_32x32_2x3_8x32x1_8x32x1_1x64x1x4_8x8x1_2x1_intrawave_v3.hip
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,70 @@ | ||
| /* | ||
| * Copyright (c) Meta Platforms, Inc. and affiliates. | ||
| * All rights reserved. | ||
| * | ||
| * This source code is licensed under the BSD-style license found in the | ||
| * LICENSE file in the root directory of this source tree. | ||
| */ | ||
|
|
||
| #include "fp8_rowwise_common.h" | ||
|
|
||
| at::Tensor | ||
| fp8_rowwise_256x256x96x128_32x32_2x3_8x32x1_8x32x1_1x64x1x4_8x8x1_2x1_intrawave_v3( | ||
| at::Tensor XQ, | ||
| at::Tensor WQ, | ||
| at::Tensor x_scale, | ||
| at::Tensor w_scale, | ||
| at::Tensor Y) { | ||
| // Check if this input needs to be padded. | ||
| int M = size_to_dim_(XQ.dim() - 1, XQ.sizes()); | ||
| int N = WQ.size(0); | ||
| int K = WQ.size(1); | ||
| bool kpad = (K % 128 != 0); | ||
|
|
||
| // Dispatch based on whether padding is needed or not. | ||
| if (kpad) { | ||
| using DeviceGemmInstance = DeviceGemmHelper< | ||
| 256, | ||
| 256, | ||
| 96, | ||
| 128, | ||
| 32, | ||
| 32, | ||
| 2, | ||
| 3, | ||
| S<8, 32, 1>, | ||
| S<8, 32, 1>, | ||
| S<1, 64, 1, 4>, | ||
| S<8, 8, 1>, | ||
| 2, | ||
| 1, | ||
| ck::BlockGemmPipelineScheduler::Intrawave, | ||
| ck::BlockGemmPipelineVersion::v3, | ||
| ck::tensor_operation::device::GemmSpecialization::KPadding>; | ||
| // Run kernel instance. | ||
| return f8f8bf16_rowwise_impl<DeviceGemmInstance>( | ||
| XQ, WQ, x_scale, w_scale, Y); | ||
| } else { | ||
| using DeviceGemmInstance = DeviceGemmHelper< | ||
| 256, | ||
| 256, | ||
| 96, | ||
| 128, | ||
| 32, | ||
| 32, | ||
| 2, | ||
| 3, | ||
| S<8, 32, 1>, | ||
| S<8, 32, 1>, | ||
| S<1, 64, 1, 4>, | ||
| S<8, 8, 1>, | ||
| 2, | ||
| 1, | ||
| ck::BlockGemmPipelineScheduler::Intrawave, | ||
| ck::BlockGemmPipelineVersion::v3, | ||
| ck::tensor_operation::device::GemmSpecialization::Default>; | ||
| // Run kernel instance. | ||
| return f8f8bf16_rowwise_impl<DeviceGemmInstance>( | ||
| XQ, WQ, x_scale, w_scale, Y); | ||
| } | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters