Skip to content

Commit eeecd62

Browse files
authored
Update cgemm_kernel_8x2_haswell.c
1 parent 3ce6bcd commit eeecd62

File tree

1 file changed

+5
-0
lines changed

1 file changed

+5
-0
lines changed

kernel/x86_64/cgemm_kernel_8x2_haswell.c

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -104,6 +104,7 @@
104104
KERNEL_k1m8n##ndim "decq %5; jnz "#ndim"8882b;"\
105105
#ndim"8883:\n\t"\
106106
"prefetcht0 (%%r14); prefetcht0 64(%%r14);" SAVE_m8n##ndim
107+
107108
/* m=4, ymm 0-3 temp, ymm 4-15 acc, expanded accumulators */
108109
#define KERNEL_k1m4n1 \
109110
"vmovsldup (%0),%%ymm1; vmovshdup (%0),%%ymm2; addq $32,%0;"\
@@ -137,6 +138,7 @@
137138
"decq %5; jnz "#ndim"4441b;"\
138139
#ndim"4442:\n\t"\
139140
SAVE_m4n##ndim
141+
140142
/* m=2, xmm 0-3 temp, xmm 4-15 acc, expanded accumulators */
141143
#if A_CONJ == B_CONJ
142144
#define acc_m2n1_exp(ar,ai,b2,cl,cr) "vfmadd231ps %%xmm"#ar",%%xmm"#b2",%%xmm"#cl"; vfmadd231ps %%xmm"#ai",%%xmm"#b2",%%xmm"#cr";"
@@ -189,6 +191,7 @@
189191
"decq %5; jnz "#ndim"2221b;"\
190192
#ndim"2222:\n\t"\
191193
SAVE_m2n##ndim
194+
192195
/* m=1, xmm 0-3 temp, xmm 4-9 acc, expanded accumulators */
193196
#if A_CONJ == B_CONJ
194197
#define acc_m1n1_exp(ar,ai,b2,cl,cr) "vfmadd231ps %%xmm"#ar",%%xmm"#b2",%%xmm"#cl"; vfmadd231ps %%xmm"#ai",%%xmm"#b2",%%xmm"#cr";"
@@ -242,6 +245,7 @@
242245
"decq %5; jnz "#ndim"1111b;"\
243246
#ndim"1112:\n\t"\
244247
SAVE_m1n##ndim
248+
245249
#define COMPUTE(ndim) {\
246250
b_pref = b_ptr + ndim * K *2;\
247251
__asm__ __volatile__ (\
@@ -266,6 +270,7 @@
266270
"xmm6","xmm7","xmm8","xmm9","xmm10","xmm11","xmm12","xmm13","xmm14","xmm15");\
267271
a_ptr -= M * K *2; b_ptr += ndim * K *2; c_ptr += (ndim * LDC - M) * 2;\
268272
}
273+
269274
int __attribute__ ((noinline))
270275
CNAME(BLASLONG m, BLASLONG n, BLASLONG k, float alphar, float alphai, float * __restrict__ A, float * __restrict__ B, float * __restrict__ C, BLASLONG LDC)
271276
{

0 commit comments

Comments
 (0)