feat(bench): add vectorized + unrolled optimal version#63
Open
dssgabriel wants to merge 2 commits into
Open
Conversation
dssgabriel
added
C-enhancement
Cpp-Linter Report
|
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
| vacc = _mm256_fmadd_pd(va7, vb7, vacc); | ||
| } | ||
|
|
||
| __m128d hsum = _mm_add_pd(_mm256_castpd256_pd128(vacc), _mm256_extractf128_pd(vacc, 1)); |
There was a problem hiding this comment.
clang-tidy diagnostic
benchmarks/gemm.cpp:294:30: warning: [portability-simd-intrinsics]
'_mm_add_pd' is a non-portable x86_64 intrinsic function
294 | __m128d hsum = _mm_add_pd(_mm256_castpd256_pd128(vacc), _mm256_extractf128_pd(vacc, 1));
| ^
/home/runner/work/layout-tiled/layout-tiled/benchmarks/gemm.cpp:539:10: warning: do not declare C-style arrays, use 'std::array' instead [cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays]
539 | static struct option long_options[] = {
| ^
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:1683:5: error: 5 uninitialized fields at the end of the constructor call [clang-analyzer-optin.cplusplus.UninitializedObject,-warnings-as-errors]
1683 | ignore_unused(arg_index, named_arg_index);
| ^
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:1213:7: note: uninitialized field 'this->context_.num_args_'
1213 | int num_args_;
| ^~~~~~~~~
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:1214:15: note: uninitialized pointer 'this->context_.types_'
1214 | const type* types_;
| ^~~~~~
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:520:15: note: uninitialized pointer 'this->context_.parse_context::fmt_.data_'
520 | const Char* data_;
| ^~~~~
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:521:10: note: uninitialized field 'this->context_.parse_context::fmt_.size_'
521 | size_t size_;
| ^~~~~
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:870:7: note: uninitialized field 'this->context_.parse_context::next_arg_id_'
870 | int next_arg_id_;
| ^~~~~~~~~~~~
/home/runner/work/layout-tiled/layout-tiled/benchmarks/gemm.cpp:549:10: note: Assuming the condition is true
549 | while ((opt = getopt_long(argc, argv, "n:t:o:h", static_cast<const struct option*>(long_options), nullptr)) != -1) {
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/runner/work/layout-tiled/layout-tiled/benchmarks/gemm.cpp:549:3: note: Loop condition is true. Entering loop body
549 | while ((opt = getopt_long(argc, argv, "n:t:o:h", static_cast<const struct option*>(long_options), nullptr)) != -1) {
| ^
/home/runner/work/layout-tiled/layout-tiled/benchmarks/gemm.cpp:550:5: note: Control jumps to the 'default' case at line 571
550 | switch (opt) {
| ^
/home/runner/work/layout-tiled/layout-tiled/benchmarks/gemm.cpp:572:18: note: Calling constructor for 'fstring<std::basic_string<char> &>'
572 | fmt::print("{}", HELP_MSG);
| ^~~~
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:2709:5: note: Taking true branch
2709 | if (FMT_USE_CONSTEVAL) parse_format_string<char>(s, checker(s, arg_pack()));
| ^
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:2709:57: note: Calling constructor for 'format_string_checker<char, 1, 0, false>'
2709 | if (FMT_USE_CONSTEVAL) parse_format_string<char>(s, checker(s, arg_pack()));
| ^~~~~~~~~~~~~~~~~~~~~~
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:1683:5: note: 5 uninitialized fields at the end of the constructor call
1683 | ignore_unused(arg_index, named_arg_index);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~Broadcast 4 elems of A and multiply against cols of B, accumulating into 8 vals of C. This beats the simple vectorization strategy used in `*_vec_unroll_k`. Signed-off-by: Gabriel Dos Santos <gabriel.dossantos@cea.fr>
| __m256d vb7 = _mm256_loadu_pd(&tB[k + 28, j]); | ||
| vacc = _mm256_fmadd_pd(va7, vb7, vacc); | ||
| } | ||
| __m128d hsum = _mm_add_pd(_mm256_castpd256_pd128(vacc), _mm256_extractf128_pd(vacc, 1)); |
There was a problem hiding this comment.
clang-tidy diagnostic
benchmarks/gemm.cpp:293:30: warning: [portability-simd-intrinsics]
'_mm_add_pd' is a non-portable x86_64 intrinsic function
293 | __m128d hsum = _mm_add_pd(_mm256_castpd256_pd128(vacc), _mm256_extractf128_pd(vacc, 1));
| ^
/home/runner/work/layout-tiled/layout-tiled/benchmarks/gemm.cpp:627:10: warning: do not declare C-style arrays, use 'std::array' instead [cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays]
627 | static struct option long_options[] = {
| ^
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:1683:5: error: 5 uninitialized fields at the end of the constructor call [clang-analyzer-optin.cplusplus.UninitializedObject,-warnings-as-errors]
1683 | ignore_unused(arg_index, named_arg_index);
| ^
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:520:15: note: uninitialized pointer 'this->context_.parse_context::fmt_.data_'
520 | const Char* data_;
| ^~~~~
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:521:10: note: uninitialized field 'this->context_.parse_context::fmt_.size_'
521 | size_t size_;
| ^~~~~
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:870:7: note: uninitialized field 'this->context_.parse_context::next_arg_id_'
870 | int next_arg_id_;
| ^~~~~~~~~~~~
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:1213:7: note: uninitialized field 'this->context_.num_args_'
1213 | int num_args_;
| ^~~~~~~~~
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:1214:15: note: uninitialized pointer 'this->context_.types_'
1214 | const type* types_;
| ^~~~~~
/home/runner/work/layout-tiled/layout-tiled/benchmarks/gemm.cpp:637:10: note: Assuming the condition is true
637 | while ((opt = getopt_long(argc, argv, "n:t:o:h", static_cast<const struct option*>(long_options), nullptr)) != -1) {
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/runner/work/layout-tiled/layout-tiled/benchmarks/gemm.cpp:637:3: note: Loop condition is true. Entering loop body
637 | while ((opt = getopt_long(argc, argv, "n:t:o:h", static_cast<const struct option*>(long_options), nullptr)) != -1) {
| ^
/home/runner/work/layout-tiled/layout-tiled/benchmarks/gemm.cpp:638:5: note: Control jumps to the 'default' case at line 659
638 | switch (opt) {
| ^
/home/runner/work/layout-tiled/layout-tiled/benchmarks/gemm.cpp:660:18: note: Calling constructor for 'fstring<std::basic_string<char> &>'
660 | fmt::print("{}", HELP_MSG);
| ^~~~
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:2709:5: note: Taking true branch
2709 | if (FMT_USE_CONSTEVAL) parse_format_string<char>(s, checker(s, arg_pack()));
| ^
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:2709:57: note: Calling constructor for 'format_string_checker<char, 1, 0, false>'
2709 | if (FMT_USE_CONSTEVAL) parse_format_string<char>(s, checker(s, arg_pack()));
| ^~~~~~~~~~~~~~~~~~~~~~
/home/runner/work/layout-tiled/layout-tiled/build/_deps/fmt-src/include/fmt/base.h:1683:5: note: 5 uninitialized fields at the end of the constructor call
1683 | ignore_unused(arg_index, named_arg_index);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Description
As discussed with @hmt23, add a "more optimal" version of the GEMM benchmark where we unroll and vectorize our most efficient implementation ($B$ in LTL layout).
mbtsvw/Related issue(s):
Scope: benchmarks
Type of change: feature
Checklist