CANN: optimize the rope ops #15335
Merged
CANN: optimize the rope ops #15335
Conversation
This file contains hidden or 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
github-actions
bot
added
ggml
noemotiovon
reviewed
Aug 15, 2025
|
|
||
| if(ctx.init_ptr == nullptr || !is_attention) { |
There was a problem hiding this comment.
Add a comment indicating that is_attention is a flag used for accuracy testing.
There was a problem hiding this comment.
Thank you for your suggestion; it has been revised.
ggml/src/ggml-cann/aclnn_ops.cpp
Outdated
| if(ctx.init_ptr != nullptr){ | ||
| ACL_CHECK(aclrtFree(ctx.init_ptr)); | ||
| } | ||
| ACL_CHECK(aclrtMalloc(&ctx.init_ptr,theta_scale_length * sizeof(float_t), ACL_MEM_MALLOC_HUGE_FIRST)); |
There was a problem hiding this comment.
There is a missing space after &ctx.init_ptr,.
There was a problem hiding this comment.
Thank you for your suggestion; it has been revised.
ggml/src/ggml-cann/common.h
Outdated
| void* init_ptr = nullptr; | ||
| void* sin_ptr = nullptr; | ||
| void* cos_ptr = nullptr; | ||
| int64_t max_position_length = 200000; |
There was a problem hiding this comment.
A maximum prompt length of 200,000 is a bit excessive; let's initialize it to 65,536 here. And rename it to max_prompt_length.
There was a problem hiding this comment.
Thank you for your suggestion; it has been revised.
hipudding
approved these changes
Aug 19, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Optimize the performance of the rope operator by reusing sin_tensor and cos_tensor across different layers for each token.
Before Optimization
root@worker-33-138:/home/y00939322/rope_test/llama.cpp-master# ./build/bin/llama-bench -m /home/y00939322/qwen2.5-0.5b-instruct-fp16.gguf -p 5 -n 5 -b 1 -sm none -mg 0 -t 8 -fa 1
Optimized
root@worker-33-138:/home/y00939322/rope_test/llama.cpp-rope_ops# ./build/bin/llama-bench -m /home/y00939322/qwen2.5-0.5b-instruct-fp16.gguf -p 5 -n 5 -b 1 -sm none -mg 0 -t 8 -fa 1
Verifying the Operator Precision：
ROPE(type=f32,ne_a=[128,32,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0): OK
ROPE(type=f32,ne_a=[128,40,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0): OK
ROPE(type=f32,ne_a=[128,52,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0): OK
ROPE(type=f32,ne_a=[128,64,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0): OK
ROPE(type=f32,ne_a=[64,1,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0): OK
ROPE(type=f32,ne_a=[64,71,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0): OK
ROPE(type=f32,ne_a=[64,8,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0): OK