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

don't use this sgemv_n on Windows

This commit is contained in:
wernsaar 2014-07-19 07:15:34 +02:00
parent c8a4a56177
commit b3938fe371
4 changed files with 53 additions and 41 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
kernel/x86_64/KERNEL.BULLDOZER
View File

@ -1,4 +1,9 @@
ifdef OS_WINDOWS
SGEMVNKERNEL = ../arm/gemv_n.c
else
SGEMVNKERNEL = sgemv_n_avx.c
endif
ZGEMVNKERNEL = zgemv_n_dup.S
ZGEMVTKERNEL = zgemv_t.S

5
kernel/x86_64/KERNEL.PILEDRIVER
View File

@ -1,4 +1,9 @@
ifdef OS_WINDOWS
SGEMVNKERNEL = ../arm/gemv_n.c
else
SGEMVNKERNEL = sgemv_n_avx.c
endif
ZGEMVNKERNEL = zgemv_n_dup.S
ZGEMVTKERNEL = zgemv_t.S

6
kernel/x86_64/sgemv_n_avx.c
View File

@ -61,8 +61,10 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
FLOAT *a_ptr;
FLOAT *x_ptr;
FLOAT *y_ptr;
BLASLONG n1,n2;
BLASLONG m1,m2;
BLASLONG n1;
BLASLONG m1;
BLASLONG register m2;
BLASLONG register n2;
FLOAT *xbuffer,*ybuffer;
xbuffer = buffer;
ybuffer = xbuffer + 2048 + 256;

64
kernel/x86_64/sgemv_n_microk_bulldozer.c
View File

@ -31,7 +31,7 @@ static void sgemv_kernel_64( long n, float alpha, float *a, long lda, float *x,
float *pre = a + lda*3;
__asm __volatile
__asm__ __volatile__
(
"movq %0, %%rax\n\t" // n -> rax
"vbroadcastss %1, %%ymm1\n\t" // alpha -> ymm1
@ -103,10 +103,10 @@ static void sgemv_kernel_64( long n, float alpha, float *a, long lda, float *x,
"m" (x), // 4
"m" (y), // 5
"m" (pre) // 6
: "rax", "rcx", "rdx", "rsi", "rdi", "r8", "r9", "r10", "r11",
"xmm0" , "xmm1",
"xmm8", "xmm9", "xmm10", "xmm11",
"xmm12", "xmm13", "xmm14", "xmm15",
: "%rax", "%rcx", "%rdx", "%rsi", "%rdi", "%r8",
"%xmm0", "%xmm1",
"%xmm8", "%xmm9", "%xmm10", "%xmm11",
"%xmm12", "%xmm13", "%xmm14", "%xmm15",
"memory"
);
@ -120,7 +120,7 @@ static void sgemv_kernel_32( long n, float alpha, float *a, long lda, float *x,
float *pre = a + lda*3;
__asm __volatile
__asm__ __volatile__
(
"movq %0, %%rax\n\t" // n -> rax
"vbroadcastss %1, %%xmm1\n\t" // alpha -> xmm1
@ -190,11 +190,6 @@ static void sgemv_kernel_32( long n, float alpha, float *a, long lda, float *x,
"m" (x), // 4
"m" (y), // 5
"m" (pre) // 6
: "rax", "rcx", "rdx", "rsi", "rdi", "r8", "r9", "r10", "r11",
"xmm0" , "xmm1",
"xmm8", "xmm9", "xmm10", "xmm11",
"xmm12", "xmm13", "xmm14", "xmm15",
"memory"
);
}
@ -202,9 +197,9 @@ static void sgemv_kernel_32( long n, float alpha, float *a, long lda, float *x,
static void sgemv_kernel_16( long n, float alpha, float *a, long lda, float *x, float *y)
{
float *pre = a + lda*1;
float *pre = a + lda*3;
__asm __volatile
__asm__ __volatile__
(
"movq %0, %%rax\n\t" // n -> rax
"vbroadcastss %1, %%ymm1\n\t" // alpha -> ymm1
@ -248,9 +243,10 @@ static void sgemv_kernel_16( long n, float alpha, float *a, long lda, float *x,
"m" (x), // 4
"m" (y), // 5
"m" (pre) // 6
: "rax", "rcx", "rdx", "rsi", "rdi", "r8", "r9", "r10", "r11",
"xmm0" , "xmm1",
"xmm12", "xmm13", "xmm14", "xmm15",
: "%rax", "%rcx", "%rdx", "%rsi", "%rdi", "%r8",
"%xmm0", "%xmm1",
"%xmm8", "%xmm9", "%xmm10", "%xmm11",
"%xmm12", "%xmm13", "%xmm14", "%xmm15",
"memory"
);
@ -261,7 +257,7 @@ static void sgemv_kernel_8( long n, float alpha, float *a, long lda, float *x, f
{
__asm __volatile
__asm__ __volatile__
(
"movq %0, %%rax\n\t" // n -> rax
"vbroadcastss %1, %%ymm1\n\t" // alpha -> ymm1
@ -295,9 +291,10 @@ static void sgemv_kernel_8( long n, float alpha, float *a, long lda, float *x, f
"m" (lda), // 3
"m" (x), // 4
"m" (y) // 5
: "rax", "rcx", "rdx", "rsi", "rdi", "r8", "r9", "r10", "r11",
"xmm0" , "xmm1",
"xmm12", "xmm13", "xmm14", "xmm15",
: "%rax", "%rcx", "%rdx", "%rsi", "%rdi", "%r8",
"%xmm0", "%xmm1",
"%xmm8", "%xmm9", "%xmm10", "%xmm11",
"%xmm12", "%xmm13", "%xmm14", "%xmm15",
"memory"
);
@ -308,7 +305,7 @@ static void sgemv_kernel_4( long n, float alpha, float *a, long lda, float *x, f
{
__asm __volatile
__asm__ __volatile__
(
"movq %0, %%rax\n\t" // n -> rax
"vbroadcastss %1, %%xmm1\n\t" // alpha -> xmm1
@ -342,9 +339,10 @@ static void sgemv_kernel_4( long n, float alpha, float *a, long lda, float *x, f
"m" (lda), // 3
"m" (x), // 4
"m" (y) // 5
: "rax", "rcx", "rdx", "rsi", "rdi", "r8", "r9", "r10", "r11",
"xmm0" , "xmm1",
"xmm12", "xmm13", "xmm14", "xmm15",
: "%rax", "%rcx", "%rdx", "%rsi", "%rdi", "%r8",
"%xmm0", "%xmm1",
"%xmm8", "%xmm9", "%xmm10", "%xmm11",
"%xmm12", "%xmm13", "%xmm14", "%xmm15",
"memory"
);
@ -354,7 +352,7 @@ static void sgemv_kernel_2( long n, float alpha, float *a, long lda, float *x, f
{
__asm __volatile
__asm__ __volatile__
(
"movq %0, %%rax\n\t" // n -> rax
"vmovss %1, %%xmm1\n\t" // alpha -> xmm1
@ -392,9 +390,10 @@ static void sgemv_kernel_2( long n, float alpha, float *a, long lda, float *x, f
"m" (lda), // 3
"m" (x), // 4
"m" (y) // 5
: "rax", "rcx", "rdx", "rsi", "rdi", "r8", "r9", "r10", "r11",
"xmm0" , "xmm1",
"xmm12", "xmm13", "xmm14", "xmm15",
: "%rax", "%rcx", "%rdx", "%rsi", "%rdi", "%r8",
"%xmm0", "%xmm1",
"%xmm8", "%xmm9", "%xmm10", "%xmm11",
"%xmm12", "%xmm13", "%xmm14", "%xmm15",
"memory"
);
@ -406,7 +405,7 @@ static void sgemv_kernel_1( long n, float alpha, float *a, long lda, float *x, f
{
__asm __volatile
__asm__ __volatile__
(
"movq %0, %%rax\n\t" // n -> rax
"vmovss %1, %%xmm1\n\t" // alpha -> xmm1
@ -440,9 +439,10 @@ static void sgemv_kernel_1( long n, float alpha, float *a, long lda, float *x, f
"m" (lda), // 3
"m" (x), // 4
"m" (y) // 5
: "rax", "rcx", "rdx", "rsi", "rdi", "r8", "r9", "r10", "r11",
"xmm0" , "xmm1",
"xmm12", "xmm13", "xmm14", "xmm15",
: "%rax", "%rcx", "%rdx", "%rsi", "%rdi", "%r8",
"%xmm0", "%xmm1",
"%xmm8", "%xmm9", "%xmm10", "%xmm11",
"%xmm12", "%xmm13", "%xmm14", "%xmm15",
"memory"
);