http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56253
Bug #: 56253
Summary: fp-contract does not work with SSE and AVX FMAs
(neither FMA4 nor FMA3)
Classification: Unclassified
Product: gcc
Version: 4.7.2
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: kr...@kde.org
Take the following testcase:
#include <immintrin.h>
__m256 foo(__m256 a, __m256 b, __m256 c)
{
return _mm256_add_ps(_mm256_mul_ps(a, b), c);
}
__m128 foo(__m128 a, __m128 b, __m128 c)
{
return _mm_add_ps(_mm_mul_ps(a, b), c);
}
float foo(float a, float b, float c)
{
return a * b + c;
}
compiled with 'g++ -O3 -mfma -ffp-contract=fast -fabi-version=0 -c' only the
third function uses fmas (same for -mfma4). The SSE and AVX variant should make
the same contraction as is implemented for scalar operations.
