[ESIMD] Fix saturation argument of DPAS #6647
Merged
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
The template argument for saturation was declared/used such a way that any type could be passed to it, which would cause enforcement of saturation when not intended. In even worse scenarios the DPAS call with 3 simd arguments was recognized as DPAS with 2 simd arguments + saturation argument: dpas(src0,src1,src2) was treated as dpas(src1,src2,sat), which caused totally incorrect behavior at runtime. Also, this patch fixes the incorrect detection of ops_per_channel for tfloat32 type on HOST. Signed-off-by: Vyacheslav N Klochkov <[email protected]>
kbobrovs
reviewed
Aug 26, 2022
| @@ -449,7 +449,10 @@ __esimd_dpas_inner(const __ESIMD_DNS::vector_type_t<T0, SZ> *src0, | |||
| __ESIMD_EMU_DNS::is_inttype<RT>::value>::set(); | |||
|
|
|||
| constexpr __ESIMD_NS::uint ops_per_chan = | |||
| src1_precision == __ESIMD_ENS::argument_type::BF16 || | |||
| src1_precision == __ESIMD_ENS::argument_type::TF32 || | |||
There was a problem hiding this comment.
Nit: seems this logic overlaps with __esimd_dpas_bits_precision. As a variant:
constexpr auto nbits1 = __esimd_dpas_bits_precision<src1_precision>();
constexpr auto nbits2 = __esimd_dpas_bits_precision<src2_precision>();
constexpr auto nbits = max(nbits1, nbits2);
constexpr __ESIMD_NS::uint ops_per_chan = std::min(32 / nbits, 8);
There was a problem hiding this comment.
Thank you. Fixed in c3b1981
Signed-off-by: Vyacheslav N Klochkov <[email protected]>
kbobrovs
approved these changes
Aug 26, 2022
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Outdated suggestions cannot be applied.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.
The template argument for saturation was declared/used such a way
that any type could be passed to it, which would cause enforcement of
saturation when not intended.
In even worse scenarios the DPAS call with 3 simd arguments was
recognized as DPAS with 2 simd arguments + saturation argument:
dpas(src0,src1,src2) was treated as dpas(src1,src2,sat),
which caused totally incorrect behavior at runtime.
Also, this patch fixes the incorrect detection of ops_per_channel for
tfloat32 type on HOST.
The corresponding LIT test: intel/llvm-test-suite#1180
Signed-off-by: Vyacheslav N Klochkov [email protected]