Skip to content

[fix] Now non-swizzled stride can be less than lda*A_col #979

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
solaslin merged 3 commits into from
Aug 5, 2025
Merged

[fix] Now non-swizzled stride can be less than lda*A_col #979

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
bb2527d
[fix] Now non-swizzled stride can be less than lda*A_col
Jul 30, 2025
ac3c2a4
Merge branch 'develop' into user/mengzcai/fix_nonSwizzle_stride
mengzcai Aug 3, 2025
ef34500
Merge branch 'develop' into user/mengzcai/fix_nonSwizzle_stride
mengzcai Aug 5, 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
1 change: 0 additions & 1 deletion projects/hipblaslt/clients/gtest/matmul_gtest.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -855,7 +855,6 @@ Tests:
K: [128, 129]
lda: [129, 2440]
batch_count: [1, 5]
stride_a: 0
Comment thread
solaslin marked this conversation as resolved.
Outdated
transA: T
transB: N
alpha: 1
Expand Down
18 changes: 12 additions & 6 deletions projects/hipblaslt/clients/include/testing_matmul.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1376,28 +1376,34 @@ void testing_matmul_with_bias(const Arguments& arg,
do_batched[i] = (arg.batch_count > 1);
num_batches[i] = (do_batched[i] ? arg.batch_count : 1);

stride_a[i] = (do_batched[i] && arg.stride_a[i] >= lda[i] * A_col[i]) ? arg.stride_a[i]
: lda[i] * A_col[i];
stride_a[i] = do_batched[i] ? arg.stride_a[i] : lda[i] * A_col[i];
stride_b[i] = do_batched[i] ? arg.stride_b[i] : ldb[i] * B_col[i];
stride_c[i] = do_batched[i] ? arg.stride_c[i] : ldc[i] * N[i];
stride_d[i] = do_batched[i] ? arg.stride_c[i] : ldd[i] * N[i];
stride_e[i] = do_batched[i] ? arg.stride_e[i] : lde[i] * N[i];

size_A[i] = stride_a[i] * num_batches[i];
size_A[i]
= stride_a[i] == 0 ? lda[i] * A_col[i] * num_batches[i] : stride_a[i] * num_batches[i];
size_dA[i] = size_A[i];
stride_da[i] = stride_a[i];
if(arg.swizzle_a && isSwizzleSupported(TiA))
{
//TODO:
// 1. support stride_a is 0 when swizzle
// 2. support --any_stride for hipblaslt-bench when swizzle
// 3. support arbitrary stride_a for both hipblaslt-bench and hipblaslt-test when swizzled
stride_a[i] = lda[i] * A_col[i];
stride_da[i] = 0;
size_t MiM = 16, MiK = 0, __ = 0, PackK = 0;
calculateKforSwizzling(TiA, arg, MiK, __, PackK);
size_t K_block = MiK * PackK;
size_t stride_swizzle
= ((M[i] + MiM - 1) / MiM) * MiM * ((K[i] + K_block - 1) / K_block) * K_block;
if(do_batched[i] && arg.stride_a[i] >= (int64_t)stride_swizzle)
if(do_batched[i] && stride_a[i] > 0 && stride_a[i] != lda[i] * A_col[i])
{
stride_da[i] = arg.stride_a[i];
stride_swizzle = (size_t)arg.stride_a[i];
hipblaslt_cerr << "Error stride_a: swizzle_a does not yet support arbitrary stride_a!" << std::endl;
stride_da[i] = stride_a[i];
stride_swizzle = (size_t)stride_a[i];
}
size_dA[i] = num_batches[i] * stride_swizzle;
}
Expand Down