Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
Menu
Open sidebar
OpenDAS
TransformerEngine
Commits
27cb9950
Unverified
Commit
27cb9950
authored
Apr 04, 2024
by
Tim Moon
Committed by
GitHub
Apr 04, 2024
Browse files
Compile tuned RMSNorm kernels for hidden size 8192 (#747)
Signed-off-by:
Tim Moon
<
tmoon@nvidia.com
>
parent
ce328fac
Changes
2
Hide whitespace changes
Inline
Side-by-side
Showing
2 changed files
with
11 additions
and
0 deletions
+11
-0
transformer_engine/common/rmsnorm/rmsnorm_bwd_semi_cuda_kernel.cu
...mer_engine/common/rmsnorm/rmsnorm_bwd_semi_cuda_kernel.cu
+4
-0
transformer_engine/common/rmsnorm/rmsnorm_fwd_cuda_kernel.cu
transformer_engine/common/rmsnorm/rmsnorm_fwd_cuda_kernel.cu
+7
-0
No files found.
transformer_engine/common/rmsnorm/rmsnorm_bwd_semi_cuda_kernel.cu
View file @
27cb9950
...
@@ -201,6 +201,10 @@ REGISTER_BWD_TUNED_LAUNCHER(4096, fp32, fp32, fp32, fp32, 1, 1, 4, 16, 4);
...
@@ -201,6 +201,10 @@ REGISTER_BWD_TUNED_LAUNCHER(4096, fp32, fp32, fp32, fp32, 1, 1, 4, 16, 4);
REGISTER_BWD_TUNED_LAUNCHER
(
4096
,
fp16
,
fp16
,
fp16
,
fp32
,
1
,
1
,
4
,
16
,
4
);
REGISTER_BWD_TUNED_LAUNCHER
(
4096
,
fp16
,
fp16
,
fp16
,
fp32
,
1
,
1
,
4
,
16
,
4
);
REGISTER_BWD_TUNED_LAUNCHER
(
4096
,
bf16
,
bf16
,
bf16
,
fp32
,
1
,
1
,
4
,
16
,
4
);
REGISTER_BWD_TUNED_LAUNCHER
(
4096
,
bf16
,
bf16
,
bf16
,
fp32
,
1
,
1
,
4
,
16
,
4
);
REGISTER_BWD_TUNED_LAUNCHER
(
8192
,
fp32
,
fp32
,
fp32
,
fp32
,
1
,
1
,
4
,
16
,
4
);
REGISTER_BWD_TUNED_LAUNCHER
(
8192
,
fp16
,
fp16
,
fp16
,
fp32
,
1
,
1
,
4
,
16
,
4
);
REGISTER_BWD_TUNED_LAUNCHER
(
8192
,
bf16
,
bf16
,
bf16
,
fp32
,
1
,
1
,
4
,
16
,
4
);
// Create rmsnorm general launch function and register. Macro signature:
// Create rmsnorm general launch function and register. Macro signature:
// HIDDEN_SIZE, WTYPE, ITYPE, OTYPE, CTYPE, ...
// HIDDEN_SIZE, WTYPE, ITYPE, OTYPE, CTYPE, ...
// WARPS_M, WARPS_N, BYTES_PER_LDG, BYTES_PER_LDG_FINAL
// WARPS_M, WARPS_N, BYTES_PER_LDG, BYTES_PER_LDG_FINAL
...
...
transformer_engine/common/rmsnorm/rmsnorm_fwd_cuda_kernel.cu
View file @
27cb9950
...
@@ -170,6 +170,13 @@ REGISTER_FWD_TUNED_LAUNCHER(4096, fp32, fp32, fp32, fp32, 1, 1, 4, 16);
...
@@ -170,6 +170,13 @@ REGISTER_FWD_TUNED_LAUNCHER(4096, fp32, fp32, fp32, fp32, 1, 1, 4, 16);
REGISTER_FWD_TUNED_LAUNCHER
(
4096
,
fp16
,
fp16
,
fp16
,
fp32
,
1
,
1
,
4
,
16
);
REGISTER_FWD_TUNED_LAUNCHER
(
4096
,
fp16
,
fp16
,
fp16
,
fp32
,
1
,
1
,
4
,
16
);
REGISTER_FWD_TUNED_LAUNCHER
(
4096
,
bf16
,
bf16
,
bf16
,
fp32
,
1
,
1
,
4
,
16
);
REGISTER_FWD_TUNED_LAUNCHER
(
4096
,
bf16
,
bf16
,
bf16
,
fp32
,
1
,
1
,
4
,
16
);
REGISTER_FWD_TUNED_LAUNCHER
(
8192
,
bf16
,
bf16
,
fp8e4m3
,
fp32
,
1
,
1
,
4
,
16
);
REGISTER_FWD_TUNED_LAUNCHER
(
8192
,
fp16
,
fp16
,
fp8e4m3
,
fp32
,
1
,
1
,
4
,
16
);
REGISTER_FWD_TUNED_LAUNCHER
(
8192
,
fp32
,
fp32
,
fp8e4m3
,
fp32
,
1
,
1
,
4
,
16
);
REGISTER_FWD_TUNED_LAUNCHER
(
8192
,
fp32
,
fp32
,
fp32
,
fp32
,
1
,
1
,
4
,
16
);
REGISTER_FWD_TUNED_LAUNCHER
(
8192
,
fp16
,
fp16
,
fp16
,
fp32
,
1
,
1
,
4
,
16
);
REGISTER_FWD_TUNED_LAUNCHER
(
8192
,
bf16
,
bf16
,
bf16
,
fp32
,
1
,
1
,
4
,
16
);
// Create rmsnorm general launch function and register. Macro signature:
// Create rmsnorm general launch function and register. Macro signature:
// HIDDEN_SIZE, WTYPE, ITYPE, OTYPE, CTYPE, WARPS_M, WARPS_N, BYTES_PER_LDG
// HIDDEN_SIZE, WTYPE, ITYPE, OTYPE, CTYPE, WARPS_M, WARPS_N, BYTES_PER_LDG
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment