OSchip
/
OpenBLAS
Not watched
1
0
Fork 0
Code
Releases 66 Wiki evaluate Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain HPC
Browse Source

Fix inline assembly constraints

tags/v0.3.6^2
Martin Kroeker GitHub 7 years ago
parent
e976557d29
commit
efb9038f72
No known key found for this signature in database GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 97 additions and 97 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +97
    -97
      kernel/x86_64/sgemv_n_microk_bulldozer-4.c

+ 97
- 97
kernel/x86_64/sgemv_n_microk_bulldozer-4.c View File

@@ -37,14 +37,14 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO

__asm__ __volatile__
(
"vbroadcastss (%2), %%xmm12 \n\t" // x0
"vbroadcastss 4(%2), %%xmm13 \n\t" // x1
"vbroadcastss 8(%2), %%xmm14 \n\t" // x2
"vbroadcastss 12(%2), %%xmm15 \n\t" // x3
"vbroadcastss 16(%2), %%xmm0 \n\t" // x4
"vbroadcastss 20(%2), %%xmm1 \n\t" // x5
"vbroadcastss 24(%2), %%xmm2 \n\t" // x6
"vbroadcastss 28(%2), %%xmm3 \n\t" // x7
"vbroadcastss (%3), %%xmm12 \n\t" // x0
"vbroadcastss 4(%3), %%xmm13 \n\t" // x1
"vbroadcastss 8(%3), %%xmm14 \n\t" // x2
"vbroadcastss 12(%3), %%xmm15 \n\t" // x3
"vbroadcastss 16(%3), %%xmm0 \n\t" // x4
"vbroadcastss 20(%3), %%xmm1 \n\t" // x5
"vbroadcastss 24(%3), %%xmm2 \n\t" // x6
"vbroadcastss 28(%3), %%xmm3 \n\t" // x7

"vbroadcastss (%9), %%xmm8 \n\t" // alpha

@@ -54,22 +54,22 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
"vxorps %%xmm4, %%xmm4 , %%xmm4 \n\t"
"vxorps %%xmm5, %%xmm5 , %%xmm5 \n\t"

"vfmaddps %%xmm4, (%4,%0,4), %%xmm12, %%xmm4 \n\t"
"vfmaddps %%xmm5, (%5,%0,4), %%xmm13, %%xmm5 \n\t"
"vfmaddps %%xmm4, (%6,%0,4), %%xmm14, %%xmm4 \n\t"
"vfmaddps %%xmm5, (%7,%0,4), %%xmm15, %%xmm5 \n\t"
"vfmaddps %%xmm4, (%5,%0,4), %%xmm12, %%xmm4 \n\t"
"vfmaddps %%xmm5, (%6,%0,4), %%xmm13, %%xmm5 \n\t"
"vfmaddps %%xmm4, (%7,%0,4), %%xmm14, %%xmm4 \n\t"
"vfmaddps %%xmm5, (%8,%0,4), %%xmm15, %%xmm5 \n\t"
"addq $4 , %0 \n\t"

"vfmaddps %%xmm4, (%4,%8,4), %%xmm0 , %%xmm4 \n\t"
"vfmaddps %%xmm5, (%5,%8,4), %%xmm1 , %%xmm5 \n\t"
"vfmaddps %%xmm4, (%6,%8,4), %%xmm2 , %%xmm4 \n\t"
"vfmaddps %%xmm5, (%7,%8,4), %%xmm3 , %%xmm5 \n\t"
"addq $4 , %8 \n\t"
"vfmaddps %%xmm4, (%5,%2,4), %%xmm0 , %%xmm4 \n\t"
"vfmaddps %%xmm5, (%6,%2,4), %%xmm1 , %%xmm5 \n\t"
"vfmaddps %%xmm4, (%7,%2,4), %%xmm2 , %%xmm4 \n\t"
"vfmaddps %%xmm5, (%8,%2,4), %%xmm3 , %%xmm5 \n\t"
"addq $4 , %2 \n\t"
"vaddps %%xmm5 , %%xmm4, %%xmm4 \n\t"
"vfmaddps -16(%3,%0,4) , %%xmm4, %%xmm8,%%xmm6 \n\t"
"vfmaddps -16(%4,%0,4) , %%xmm4, %%xmm8,%%xmm6 \n\t"
"subq $4 , %1 \n\t"
"vmovups %%xmm6, -16(%3,%0,4) \n\t" // 4 * y
"vmovups %%xmm6, -16(%4,%0,4) \n\t" // 4 * y

"2: \n\t"

@@ -79,31 +79,31 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
"vxorps %%xmm4, %%xmm4 , %%xmm4 \n\t"
"vxorps %%xmm5, %%xmm5 , %%xmm5 \n\t"

"vfmaddps %%xmm4, (%4,%0,4), %%xmm12, %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%4,%0,4), %%xmm12, %%xmm5 \n\t"
"vfmaddps %%xmm4, (%5,%0,4), %%xmm13, %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%5,%0,4), %%xmm13, %%xmm5 \n\t"
"vfmaddps %%xmm4, (%6,%0,4), %%xmm14, %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%6,%0,4), %%xmm14, %%xmm5 \n\t"
"vfmaddps %%xmm4, (%7,%0,4), %%xmm15, %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%7,%0,4), %%xmm15, %%xmm5 \n\t"
"vfmaddps %%xmm4, (%4,%8,4), %%xmm0 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%4,%8,4), %%xmm0 , %%xmm5 \n\t"
"vfmaddps %%xmm4, (%5,%8,4), %%xmm1 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%5,%8,4), %%xmm1 , %%xmm5 \n\t"
"vfmaddps %%xmm4, (%6,%8,4), %%xmm2 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%6,%8,4), %%xmm2 , %%xmm5 \n\t"
"vfmaddps %%xmm4, (%7,%8,4), %%xmm3 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%7,%8,4), %%xmm3 , %%xmm5 \n\t"
"vfmaddps %%xmm4, (%5,%0,4), %%xmm12, %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%5,%0,4), %%xmm12, %%xmm5 \n\t"
"vfmaddps %%xmm4, (%6,%0,4), %%xmm13, %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%6,%0,4), %%xmm13, %%xmm5 \n\t"
"vfmaddps %%xmm4, (%7,%0,4), %%xmm14, %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%7,%0,4), %%xmm14, %%xmm5 \n\t"
"vfmaddps %%xmm4, (%8,%0,4), %%xmm15, %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%8,%0,4), %%xmm15, %%xmm5 \n\t"
"vfmaddps %%xmm4, (%5,%2,4), %%xmm0 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%5,%2,4), %%xmm0 , %%xmm5 \n\t"
"vfmaddps %%xmm4, (%6,%2,4), %%xmm1 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%6,%2,4), %%xmm1 , %%xmm5 \n\t"
"vfmaddps %%xmm4, (%7,%2,4), %%xmm2 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%7,%2,4), %%xmm2 , %%xmm5 \n\t"
"vfmaddps %%xmm4, (%8,%2,4), %%xmm3 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%8,%2,4), %%xmm3 , %%xmm5 \n\t"
"vfmaddps (%3,%0,4) , %%xmm4,%%xmm8,%%xmm4 \n\t"
"vfmaddps 16(%3,%0,4) , %%xmm5,%%xmm8,%%xmm5 \n\t"
"vmovups %%xmm4, (%3,%0,4) \n\t" // 4 * y
"vmovups %%xmm5, 16(%3,%0,4) \n\t" // 4 * y
"vfmaddps (%4,%0,4) , %%xmm4,%%xmm8,%%xmm4 \n\t"
"vfmaddps 16(%4,%0,4) , %%xmm5,%%xmm8,%%xmm5 \n\t"
"vmovups %%xmm4, (%4,%0,4) \n\t" // 4 * y
"vmovups %%xmm5, 16(%4,%0,4) \n\t" // 4 * y

"addq $8 , %0 \n\t"
"addq $8 , %8 \n\t"
"addq $8 , %2 \n\t"
"subq $8 , %1 \n\t"


@@ -120,62 +120,62 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
"vxorps %%xmm6, %%xmm6 , %%xmm6 \n\t"
"vxorps %%xmm7, %%xmm7 , %%xmm7 \n\t"

"prefetcht0 192(%4,%0,4) \n\t"
"vfmaddps %%xmm4, (%4,%0,4), %%xmm12, %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%4,%0,4), %%xmm12, %%xmm5 \n\t"
"prefetcht0 192(%5,%0,4) \n\t"
"vfmaddps %%xmm4, (%5,%0,4), %%xmm13, %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%5,%0,4), %%xmm13, %%xmm5 \n\t"
"vfmaddps %%xmm4, (%5,%0,4), %%xmm12, %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%5,%0,4), %%xmm12, %%xmm5 \n\t"
"prefetcht0 192(%6,%0,4) \n\t"
"vfmaddps %%xmm4, (%6,%0,4), %%xmm14, %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%6,%0,4), %%xmm14, %%xmm5 \n\t"
"vfmaddps %%xmm4, (%6,%0,4), %%xmm13, %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%6,%0,4), %%xmm13, %%xmm5 \n\t"
"prefetcht0 192(%7,%0,4) \n\t"
"vfmaddps %%xmm4, (%7,%0,4), %%xmm15, %%xmm4 \n\t"
"vfmaddps %%xmm4, (%7,%0,4), %%xmm14, %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%7,%0,4), %%xmm14, %%xmm5 \n\t"
"prefetcht0 192(%8,%0,4) \n\t"
"vfmaddps %%xmm4, (%8,%0,4), %%xmm15, %%xmm4 \n\t"
".align 2 \n\t"
"vfmaddps %%xmm5, 16(%7,%0,4), %%xmm15, %%xmm5 \n\t"

"vfmaddps %%xmm6, 32(%4,%0,4), %%xmm12, %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%4,%0,4), %%xmm12, %%xmm7 \n\t"
"vfmaddps %%xmm6, 32(%5,%0,4), %%xmm13, %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%5,%0,4), %%xmm13, %%xmm7 \n\t"
"vfmaddps %%xmm6, 32(%6,%0,4), %%xmm14, %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%6,%0,4), %%xmm14, %%xmm7 \n\t"
"vfmaddps %%xmm6, 32(%7,%0,4), %%xmm15, %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%7,%0,4), %%xmm15, %%xmm7 \n\t"
"prefetcht0 192(%4,%8,4) \n\t"
"vfmaddps %%xmm4, (%4,%8,4), %%xmm0 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%4,%8,4), %%xmm0 , %%xmm5 \n\t"
"prefetcht0 192(%5,%8,4) \n\t"
"vfmaddps %%xmm4, (%5,%8,4), %%xmm1 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%5,%8,4), %%xmm1 , %%xmm5 \n\t"
"prefetcht0 192(%6,%8,4) \n\t"
"vfmaddps %%xmm4, (%6,%8,4), %%xmm2 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%6,%8,4), %%xmm2 , %%xmm5 \n\t"
"prefetcht0 192(%7,%8,4) \n\t"
"vfmaddps %%xmm4, (%7,%8,4), %%xmm3 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%7,%8,4), %%xmm3 , %%xmm5 \n\t"
"vfmaddps %%xmm5, 16(%8,%0,4), %%xmm15, %%xmm5 \n\t"
"vfmaddps %%xmm6, 32(%5,%0,4), %%xmm12, %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%5,%0,4), %%xmm12, %%xmm7 \n\t"
"vfmaddps %%xmm6, 32(%6,%0,4), %%xmm13, %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%6,%0,4), %%xmm13, %%xmm7 \n\t"
"vfmaddps %%xmm6, 32(%7,%0,4), %%xmm14, %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%7,%0,4), %%xmm14, %%xmm7 \n\t"
"vfmaddps %%xmm6, 32(%8,%0,4), %%xmm15, %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%8,%0,4), %%xmm15, %%xmm7 \n\t"
"prefetcht0 192(%5,%2,4) \n\t"
"vfmaddps %%xmm4, (%5,%2,4), %%xmm0 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%5,%2,4), %%xmm0 , %%xmm5 \n\t"
"prefetcht0 192(%6,%2,4) \n\t"
"vfmaddps %%xmm4, (%6,%2,4), %%xmm1 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%6,%2,4), %%xmm1 , %%xmm5 \n\t"
"prefetcht0 192(%7,%2,4) \n\t"
"vfmaddps %%xmm4, (%7,%2,4), %%xmm2 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%7,%2,4), %%xmm2 , %%xmm5 \n\t"
"prefetcht0 192(%8,%2,4) \n\t"
"vfmaddps %%xmm4, (%8,%2,4), %%xmm3 , %%xmm4 \n\t"
"vfmaddps %%xmm5, 16(%8,%2,4), %%xmm3 , %%xmm5 \n\t"
"vfmaddps %%xmm6, 32(%4,%8,4), %%xmm0 , %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%4,%8,4), %%xmm0 , %%xmm7 \n\t"
"vfmaddps %%xmm6, 32(%5,%8,4), %%xmm1 , %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%5,%8,4), %%xmm1 , %%xmm7 \n\t"
"vfmaddps %%xmm6, 32(%6,%8,4), %%xmm2 , %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%6,%8,4), %%xmm2 , %%xmm7 \n\t"
"vfmaddps %%xmm6, 32(%7,%8,4), %%xmm3 , %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%7,%8,4), %%xmm3 , %%xmm7 \n\t"
"vfmaddps %%xmm6, 32(%5,%2,4), %%xmm0 , %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%5,%2,4), %%xmm0 , %%xmm7 \n\t"
"vfmaddps %%xmm6, 32(%6,%2,4), %%xmm1 , %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%6,%2,4), %%xmm1 , %%xmm7 \n\t"
"vfmaddps %%xmm6, 32(%7,%2,4), %%xmm2 , %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%7,%2,4), %%xmm2 , %%xmm7 \n\t"
"vfmaddps %%xmm6, 32(%8,%2,4), %%xmm3 , %%xmm6 \n\t"
"vfmaddps %%xmm7, 48(%8,%2,4), %%xmm3 , %%xmm7 \n\t"
"vfmaddps (%3,%0,4) , %%xmm4,%%xmm8,%%xmm4 \n\t"
"vfmaddps 16(%3,%0,4) , %%xmm5,%%xmm8,%%xmm5 \n\t"
"vfmaddps 32(%3,%0,4) , %%xmm6,%%xmm8,%%xmm6 \n\t"
"vfmaddps 48(%3,%0,4) , %%xmm7,%%xmm8,%%xmm7 \n\t"
"vfmaddps (%4,%0,4) , %%xmm4,%%xmm8,%%xmm4 \n\t"
"vfmaddps 16(%4,%0,4) , %%xmm5,%%xmm8,%%xmm5 \n\t"
"vfmaddps 32(%4,%0,4) , %%xmm6,%%xmm8,%%xmm6 \n\t"
"vfmaddps 48(%4,%0,4) , %%xmm7,%%xmm8,%%xmm7 \n\t"

"addq $16, %0 \n\t"
"vmovups %%xmm4,-64(%3,%0,4) \n\t" // 4 * y
"vmovups %%xmm5,-48(%3,%0,4) \n\t" // 4 * y
"addq $16, %8 \n\t"
"vmovups %%xmm6,-32(%3,%0,4) \n\t" // 4 * y
"vmovups %%xmm7,-16(%3,%0,4) \n\t" // 4 * y
"vmovups %%xmm4,-64(%4,%0,4) \n\t" // 4 * y
"vmovups %%xmm5,-48(%4,%0,4) \n\t" // 4 * y
"addq $16, %2 \n\t"
"vmovups %%xmm6,-32(%4,%0,4) \n\t" // 4 * y
"vmovups %%xmm7,-16(%4,%0,4) \n\t" // 4 * y

"subq $16, %1 \n\t"
"jnz 1b \n\t"
@@ -184,15 +184,15 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO

:
"+r" (i), // 0
"+r" (n) // 1
"+r" (n), // 1
"+r" (lda4) // 2
:
"r" (x), // 2
"r" (y), // 3
"r" (ap[0]), // 4
"r" (ap[1]), // 5
"r" (ap[2]), // 6
"r" (ap[3]), // 7
"r" (lda4), // 8
"r" (x), // 3
"r" (y), // 4
"r" (ap[0]), // 5
"r" (ap[1]), // 6
"r" (ap[2]), // 7
"r" (ap[3]), // 8
"r" (alpha) // 9
: "cc",
"%xmm0", "%xmm1",


Write Preview
Loading…
Cancel
Save