floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Save and restore input argument 8 (lda4) 

Fixes miscompilation with gcc9 -ftree-vectorize (related to issue #2009)
This commit is contained in:
Martin Kroeker 2019-02-14 22:43:18 +01:00 committed by GitHub
parent 69edc5bbe7
commit 46e415b140
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 5 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
kernel/x86_64/sgemv_n_microk_haswell-4.c
View File

@ -26,7 +26,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*****************************************************************************/ *****************************************************************************/
#define HAVE_KERNEL_4x8 1 #define HAVE_KERNEL_4x8 1
static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLONG lda4, FLOAT *alpha) __attribute__ ((noinline)); static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLONG lda4, FLOAT *alpha) __attribute__ ((noinline));
@ -49,6 +48,8 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
"vbroadcastss (%9), %%ymm6 \n\t" // alpha "vbroadcastss (%9), %%ymm6 \n\t" // alpha
"movq %8, %%xmm10 \n\t" //save lda
"testq $0x04, %1 \n\t" "testq $0x04, %1 \n\t"
"jz 2f \n\t" "jz 2f \n\t"
@ -151,6 +152,7 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
"4: \n\t" "4: \n\t"
"vzeroupper \n\t" "vzeroupper \n\t"
"movq %%xmm10, %8 \n\t" //restore lda
: :
"+r" (i), // 0 "+r" (i), // 0
@ -170,6 +172,7 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
"%xmm4", "%xmm5", "%xmm4", "%xmm5",
"%xmm6", "%xmm7", "%xmm6", "%xmm7",
"%xmm8", "%xmm9", "%xmm8", "%xmm9",
"%xmm10",
"%xmm12", "%xmm13", "%xmm14", "%xmm15", "%xmm12", "%xmm13", "%xmm14", "%xmm15",
"memory" "memory"
); );
@ -177,7 +180,6 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
} }
#define HAVE_KERNEL_4x4 1 #define HAVE_KERNEL_4x4 1
static void sgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha) __attribute__ ((noinline)); static void sgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha) __attribute__ ((noinline));
@ -196,6 +198,7 @@ static void sgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
"vbroadcastss (%8), %%ymm6 \n\t" // alpha "vbroadcastss (%8), %%ymm6 \n\t" // alpha
"testq $0x04, %1 \n\t" "testq $0x04, %1 \n\t"
"jz 2f \n\t" "jz 2f \n\t"