diff --git a/kernel/x86_64/daxpy.c b/kernel/x86_64/daxpy.c index 9207e209f..56d323cbe 100644 --- a/kernel/x86_64/daxpy.c +++ b/kernel/x86_64/daxpy.c @@ -79,11 +79,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS if ( (inc_x == 1) && (inc_y == 1) ) { -#if defined(SANDYBRIDGE) - BLASLONG n1 = n & -32; -#else BLASLONG n1 = n & -16; -#endif if ( n1 ) daxpy_kernel_8(n1, x, y , &da ); diff --git a/kernel/x86_64/daxpy_microk_sandy-2.c b/kernel/x86_64/daxpy_microk_sandy-2.c index 963ad322d..522e084dc 100644 --- a/kernel/x86_64/daxpy_microk_sandy-2.c +++ b/kernel/x86_64/daxpy_microk_sandy-2.c @@ -37,48 +37,67 @@ static void daxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha) __asm__ __volatile__ ( "vbroadcastsd (%4), %%ymm0 \n\t" // alpha - - ".align 16 \n\t" - "1: \n\t" "vmovups (%3,%0,8), %%ymm8 \n\t" "vmovups 32(%3,%0,8), %%ymm9 \n\t" "vmovups 64(%3,%0,8), %%ymm10 \n\t" "vmovups 96(%3,%0,8), %%ymm11 \n\t" - "vmovups 128(%3,%0,8), %%ymm12 \n\t" - "vmovups 160(%3,%0,8), %%ymm13 \n\t" - "vmovups 192(%3,%0,8), %%ymm14 \n\t" - "vmovups 224(%3,%0,8), %%ymm15 \n\t" + "vmovups (%2,%0,8), %%ymm4 \n\t" + "vmovups 32(%2,%0,8), %%ymm5 \n\t" + "vmovups 64(%2,%0,8), %%ymm6 \n\t" + "vmovups 96(%2,%0,8), %%ymm7 \n\t" - "vmulpd (%2,%0,8), %%ymm0, %%ymm1 \n\t" - "vmulpd 32(%2,%0,8), %%ymm0, %%ymm2 \n\t" - "vaddpd %%ymm8 , %%ymm1, %%ymm8 \n\t" - "vmulpd 64(%2,%0,8), %%ymm0, %%ymm3 \n\t" - "vaddpd %%ymm9 , %%ymm2, %%ymm9 \n\t" - "vmulpd 96(%2,%0,8), %%ymm0, %%ymm4 \n\t" - "vaddpd %%ymm10, %%ymm3, %%ymm10 \n\t" - "vmulpd 128(%2,%0,8), %%ymm0, %%ymm5 \n\t" - "vaddpd %%ymm11, %%ymm4, %%ymm11 \n\t" - "vmulpd 160(%2,%0,8), %%ymm0, %%ymm6 \n\t" - "vaddpd %%ymm12, %%ymm5, %%ymm12 \n\t" - "vmulpd 192(%2,%0,8), %%ymm0, %%ymm7 \n\t" - "vmulpd 224(%2,%0,8), %%ymm0, %%ymm1 \n\t" + "addq $16, %0 \n\t" + "subq $16, %1 \n\t" + "jz 2f \n\t" - "vaddpd %%ymm13, %%ymm6, %%ymm13 \n\t" - "vmovups %%ymm8 , (%3,%0,8) \n\t" - "vaddpd %%ymm14, %%ymm7, %%ymm14 \n\t" - "vmovups %%ymm9 , 32(%3,%0,8) \n\t" - "vaddpd %%ymm15, %%ymm1, %%ymm15 \n\t" - "vmovups %%ymm10, 64(%3,%0,8) \n\t" - "vmovups %%ymm11, 96(%3,%0,8) \n\t" - "vmovups %%ymm12,128(%3,%0,8) \n\t" - "vmovups %%ymm13,160(%3,%0,8) \n\t" - "vmovups %%ymm14,192(%3,%0,8) \n\t" - "vmovups %%ymm15,224(%3,%0,8) \n\t" + ".align 16 \n\t" + "1: \n\t" - "addq $32, %0 \n\t" - "subq $32, %1 \n\t" + "vmulpd %%ymm4, %%ymm0, %%ymm4 \n\t" + "vaddpd %%ymm8 , %%ymm4, %%ymm12 \n\t" + "vmulpd %%ymm5, %%ymm0, %%ymm5 \n\t" + "vaddpd %%ymm9 , %%ymm5, %%ymm13 \n\t" + "vmulpd %%ymm6, %%ymm0, %%ymm6 \n\t" + "vaddpd %%ymm10, %%ymm6, %%ymm14 \n\t" + "vmulpd %%ymm7, %%ymm0, %%ymm7 \n\t" + "vaddpd %%ymm11, %%ymm7, %%ymm15 \n\t" + + "vmovups (%3,%0,8), %%ymm8 \n\t" + "vmovups 32(%3,%0,8), %%ymm9 \n\t" + "vmovups 64(%3,%0,8), %%ymm10 \n\t" + "vmovups 96(%3,%0,8), %%ymm11 \n\t" + + "vmovups (%2,%0,8), %%ymm4 \n\t" + "vmovups 32(%2,%0,8), %%ymm5 \n\t" + "vmovups 64(%2,%0,8), %%ymm6 \n\t" + "vmovups 96(%2,%0,8), %%ymm7 \n\t" + + "vmovups %%ymm12, -128(%3,%0,8) \n\t" + "vmovups %%ymm13, -96(%3,%0,8) \n\t" + "vmovups %%ymm14, -64(%3,%0,8) \n\t" + "vmovups %%ymm15, -32(%3,%0,8) \n\t" + + "addq $16, %0 \n\t" + "subq $16, %1 \n\t" "jnz 1b \n\t" - "vzeroupper \n\t" + + "2: \n\t" + "vmulpd %%ymm4, %%ymm0, %%ymm4 \n\t" + "vmulpd %%ymm5, %%ymm0, %%ymm5 \n\t" + "vmulpd %%ymm6, %%ymm0, %%ymm6 \n\t" + "vmulpd %%ymm7, %%ymm0, %%ymm7 \n\t" + + "vaddpd %%ymm8 , %%ymm4, %%ymm12 \n\t" + "vaddpd %%ymm9 , %%ymm5, %%ymm13 \n\t" + "vaddpd %%ymm10, %%ymm6, %%ymm14 \n\t" + "vaddpd %%ymm11, %%ymm7, %%ymm15 \n\t" + + "vmovups %%ymm12, -128(%3,%0,8) \n\t" + "vmovups %%ymm13, -96(%3,%0,8) \n\t" + "vmovups %%ymm14, -64(%3,%0,8) \n\t" + "vmovups %%ymm15, -32(%3,%0,8) \n\t" + + "vzeroupper \n\t" : : diff --git a/kernel/x86_64/saxpy.c b/kernel/x86_64/saxpy.c index b37e24d9b..0b76c42f7 100644 --- a/kernel/x86_64/saxpy.c +++ b/kernel/x86_64/saxpy.c @@ -75,11 +75,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS if ( (inc_x == 1) && (inc_y == 1) ) { -#if defined(SANDYBRIDGE) - BLASLONG n1 = n & -64; -#else BLASLONG n1 = n & -32; -#endif if ( n1 ) saxpy_kernel_16(n1, x, y , &da ); diff --git a/kernel/x86_64/saxpy_microk_sandy-2.c b/kernel/x86_64/saxpy_microk_sandy-2.c index 8a4392d37..159a23175 100644 --- a/kernel/x86_64/saxpy_microk_sandy-2.c +++ b/kernel/x86_64/saxpy_microk_sandy-2.c @@ -37,48 +37,67 @@ static void saxpy_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha) __asm__ __volatile__ ( "vbroadcastss (%4), %%ymm0 \n\t" // alpha - - ".align 16 \n\t" - "1: \n\t" "vmovups (%3,%0,4), %%ymm8 \n\t" "vmovups 32(%3,%0,4), %%ymm9 \n\t" "vmovups 64(%3,%0,4), %%ymm10 \n\t" "vmovups 96(%3,%0,4), %%ymm11 \n\t" - "vmovups 128(%3,%0,4), %%ymm12 \n\t" - "vmovups 160(%3,%0,4), %%ymm13 \n\t" - "vmovups 192(%3,%0,4), %%ymm14 \n\t" - "vmovups 224(%3,%0,4), %%ymm15 \n\t" + "vmovups (%2,%0,4), %%ymm4 \n\t" + "vmovups 32(%2,%0,4), %%ymm5 \n\t" + "vmovups 64(%2,%0,4), %%ymm6 \n\t" + "vmovups 96(%2,%0,4), %%ymm7 \n\t" - "vmulps (%2,%0,4), %%ymm0, %%ymm1 \n\t" - "vmulps 32(%2,%0,4), %%ymm0, %%ymm2 \n\t" - "vaddps %%ymm8 , %%ymm1, %%ymm8 \n\t" - "vmulps 64(%2,%0,4), %%ymm0, %%ymm3 \n\t" - "vaddps %%ymm9 , %%ymm2, %%ymm9 \n\t" - "vmulps 96(%2,%0,4), %%ymm0, %%ymm4 \n\t" - "vaddps %%ymm10, %%ymm3, %%ymm10 \n\t" - "vmulps 128(%2,%0,4), %%ymm0, %%ymm5 \n\t" - "vaddps %%ymm11, %%ymm4, %%ymm11 \n\t" - "vmulps 160(%2,%0,4), %%ymm0, %%ymm6 \n\t" - "vaddps %%ymm12, %%ymm5, %%ymm12 \n\t" - "vmulps 192(%2,%0,4), %%ymm0, %%ymm7 \n\t" - "vmulps 224(%2,%0,4), %%ymm0, %%ymm1 \n\t" + "addq $32, %0 \n\t" + "subq $32, %1 \n\t" + "jz 2f \n\t" - "vaddps %%ymm13, %%ymm6, %%ymm13 \n\t" - "vmovups %%ymm8 , (%3,%0,4) \n\t" - "vaddps %%ymm14, %%ymm7, %%ymm14 \n\t" - "vmovups %%ymm9 , 32(%3,%0,4) \n\t" - "vaddps %%ymm15, %%ymm1, %%ymm15 \n\t" - "vmovups %%ymm10, 64(%3,%0,4) \n\t" - "vmovups %%ymm11, 96(%3,%0,4) \n\t" - "vmovups %%ymm12,128(%3,%0,4) \n\t" - "vmovups %%ymm13,160(%3,%0,4) \n\t" - "vmovups %%ymm14,192(%3,%0,4) \n\t" - "vmovups %%ymm15,224(%3,%0,4) \n\t" + ".align 16 \n\t" + "1: \n\t" - "addq $64, %0 \n\t" - "subq $64, %1 \n\t" + "vmulps %%ymm4, %%ymm0, %%ymm4 \n\t" + "vaddps %%ymm8 , %%ymm4, %%ymm12 \n\t" + "vmulps %%ymm5, %%ymm0, %%ymm5 \n\t" + "vaddps %%ymm9 , %%ymm5, %%ymm13 \n\t" + "vmulps %%ymm6, %%ymm0, %%ymm6 \n\t" + "vaddps %%ymm10, %%ymm6, %%ymm14 \n\t" + "vmulps %%ymm7, %%ymm0, %%ymm7 \n\t" + "vaddps %%ymm11, %%ymm7, %%ymm15 \n\t" + + "vmovups (%3,%0,4), %%ymm8 \n\t" + "vmovups 32(%3,%0,4), %%ymm9 \n\t" + "vmovups 64(%3,%0,4), %%ymm10 \n\t" + "vmovups 96(%3,%0,4), %%ymm11 \n\t" + + "vmovups (%2,%0,4), %%ymm4 \n\t" + "vmovups 32(%2,%0,4), %%ymm5 \n\t" + "vmovups 64(%2,%0,4), %%ymm6 \n\t" + "vmovups 96(%2,%0,4), %%ymm7 \n\t" + + "vmovups %%ymm12, -128(%3,%0,4) \n\t" + "vmovups %%ymm13, -96(%3,%0,4) \n\t" + "vmovups %%ymm14, -64(%3,%0,4) \n\t" + "vmovups %%ymm15, -32(%3,%0,4) \n\t" + + "addq $32, %0 \n\t" + "subq $32, %1 \n\t" "jnz 1b \n\t" - "vzeroupper \n\t" + + "2: \n\t" + "vmulps %%ymm4, %%ymm0, %%ymm4 \n\t" + "vmulps %%ymm5, %%ymm0, %%ymm5 \n\t" + "vmulps %%ymm6, %%ymm0, %%ymm6 \n\t" + "vmulps %%ymm7, %%ymm0, %%ymm7 \n\t" + + "vaddps %%ymm8 , %%ymm4, %%ymm12 \n\t" + "vaddps %%ymm9 , %%ymm5, %%ymm13 \n\t" + "vaddps %%ymm10, %%ymm6, %%ymm14 \n\t" + "vaddps %%ymm11, %%ymm7, %%ymm15 \n\t" + + "vmovups %%ymm12, -128(%3,%0,4) \n\t" + "vmovups %%ymm13, -96(%3,%0,4) \n\t" + "vmovups %%ymm14, -64(%3,%0,4) \n\t" + "vmovups %%ymm15, -32(%3,%0,4) \n\t" + + "vzeroupper \n\t" : : @@ -90,7 +109,7 @@ static void saxpy_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha) : "cc", "%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm4", "%xmm5", "%xmm6", "%xmm7", - "%xmm8", "%xmm9", "%xmm10", "%xmm11", + "%xmm8", "%xmm9", "%xmm10", "%xmm11", "%xmm12", "%xmm13", "%xmm14", "%xmm15", "memory" );