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

optimize AVX2 SGEMM

tags/v0.3.8^2
wjc404 GitHub 5 years ago
parent
a0f0a802fc
commit
eb3c9f1db9
No known key found for this signature in database GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 490 additions and 0 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +490
    -0
      kernel/x86_64/sgemm_kernel_8x4_haswell.c

+ 490
- 0
kernel/x86_64/sgemm_kernel_8x4_haswell.c View File

@@ -0,0 +1,490 @@
/* %0 = "+r"(a_pointer), %1 = "+r"(b_pointer), %2 = "+r"(c_pointer), %3 = "+r"(ldc_in_bytes), %4 for k_count, %5 for c_store, %6 = &alpha, %7 = b_pref */
/* r11 = m_counter, r12 = k << 2(const), r13 = k_skip << 2, r14 = b_head_pos(const), r15 for assisting prefetch */

//recommended settings: GEMM_P = 320, GEMM_Q = 320.

#ifdef TRMMKERNEL
#define mult_alpha(acc,alpha,...) "vmulps "#acc","#alpha","#acc";"
#else
#define mult_alpha(acc,alpha,...) "vfmadd213ps ("#__VA_ARGS__"),"#alpha","#acc";"
#endif

#if defined TRMMKERNEL && !defined LEFT
#ifdef TRANSA
#define HEAD_SET_OFFSET(ndim) {}
#define TAIL_SET_OFFSET(ndim) {off+=ndim;}
#else
#define HEAD_SET_OFFSET(ndim) {off+=(ndim>4?4:ndim);}
#define TAIL_SET_OFFSET(ndim) {off+=(ndim>4?(ndim-4):0);}
#endif
#else
#define HEAD_SET_OFFSET(ndim) {}
#define TAIL_SET_OFFSET(ndim) {}
#endif

#if defined TRMMKERNEL && defined LEFT
#ifdef TRANSA
#define init_update_kskip(val) "subq $"#val",%%r13;"
#define save_update_kskip(val) ""
#else
#define init_update_kskip(val) ""
#define save_update_kskip(val) "addq $"#val",%%r13;"
#endif
#else
#define init_update_kskip(val) ""
#define save_update_kskip(val) ""
#endif

#ifdef TRMMKERNEL
#define init_set_k "movq %%r12,%4; subq %%r13,%4;"
#if LEFT != TRANSA
#define INIT_SET_KSKIP "movq %9,%%r13; salq $2,%%r13;"
#define init_set_pointers(a_copy,b_copy) "leaq (%0,%%r13,"#a_copy"),%0; leaq (%1,%%r13,"#b_copy"),%1;"
#define save_set_pointers(a_copy,b_copy) ""
#else
#define INIT_SET_KSKIP "movq %4,%%r13; subq %9,%%r13; salq $2,%%r13;"
#define init_set_pointers(a_copy,b_copy) ""
#define save_set_pointers(a_copy,b_copy) "leaq (%0,%%r13,"#a_copy"),%0; leaq (%1,%%r13,"#b_copy"),%1;"
#endif
#else
#define INIT_SET_KSKIP "xorq %%r13,%%r13;"
#define init_set_k "movq %%r12,%4;"
#define init_set_pointers(a_copy,b_copy) ""
#define save_set_pointers(a_copy,b_copy) ""
#endif
#define init_set_pa_pb_n12(mdim) init_set_pointers(mdim,4)
#define init_set_pa_pb_n8(mdim) init_set_pointers(mdim,4)
#define init_set_pa_pb_n4(mdim) init_set_pointers(mdim,4)
#define init_set_pa_pb_n2(mdim) init_set_pointers(mdim,2)
#define init_set_pa_pb_n1(mdim) init_set_pointers(mdim,1)
#define save_set_pa_pb_n12(mdim) save_set_pointers(mdim,4)
#define save_set_pa_pb_n8(mdim) save_set_pointers(mdim,4)
#define save_set_pa_pb_n4(mdim) save_set_pointers(mdim,4)
#define save_set_pa_pb_n2(mdim) save_set_pointers(mdim,2)
#define save_set_pa_pb_n1(mdim) save_set_pointers(mdim,1)

#if defined TRMMKERNEL && !defined LEFT && defined TRANSA
#define kernel_kstart_n8(mdim) \
KERNEL_k1m##mdim##n4 KERNEL_k1m##mdim##n4 KERNEL_k1m##mdim##n4 KERNEL_k1m##mdim##n4 "subq $16,%4;"
#define kernel_kstart_n12(mdim) \
KERNEL_k1m##mdim##n4 KERNEL_k1m##mdim##n4 KERNEL_k1m##mdim##n4 KERNEL_k1m##mdim##n4\
KERNEL_k1m##mdim##n8 KERNEL_k1m##mdim##n8 KERNEL_k1m##mdim##n8 KERNEL_k1m##mdim##n8 "subq $32,%4;"
#else
#define kernel_kstart_n8(mdim) ""
#define kernel_kstart_n12(mdim) ""
#endif
#define kernel_kstart_n4(mdim) ""
#define kernel_kstart_n2(mdim) ""
#define kernel_kstart_n1(mdim) ""

/* m = 8 *//* ymm0 for alpha, ymm1-ymm3 for temporary use, ymm4-ymm15 for accumulators */
#define KERNEL_k1m8n1 \
"vmovups (%0),%%ymm1; addq $32,%0;"\
"vbroadcastss (%1),%%ymm2; vfmadd231ps %%ymm1,%%ymm2,%%ymm4;"\
"addq $4,%1;"
#define KERNEL_h_k1m8n2 \
"vmovsldup (%0),%%ymm1; vmovshdup (%0),%%ymm2; addq $32,%0;"\
"vbroadcastsd (%1),%%ymm3; vfmadd231ps %%ymm1,%%ymm3,%%ymm4; vfmadd231ps %%ymm2,%%ymm3,%%ymm5;"
#define KERNEL_k1m8n2 KERNEL_h_k1m8n2 "addq $8,%1;"
#define KERNEL_h_k1m8n4 \
KERNEL_h_k1m8n2 "vbroadcastsd 8(%1),%%ymm3; vfmadd231ps %%ymm1,%%ymm3,%%ymm6; vfmadd231ps %%ymm2,%%ymm3,%%ymm7;"
#define KERNEL_k1m8n4 KERNEL_h_k1m8n4 "addq $16,%1;"
#define unit_kernel_k1m8n4(c1,c2,c3,c4,boff1,boff2,...) \
"vbroadcastsd "#boff1"("#__VA_ARGS__"),%%ymm3; vfmadd231ps %%ymm1,%%ymm3,"#c1"; vfmadd231ps %%ymm2,%%ymm3,"#c2";"\
"vbroadcastsd "#boff2"("#__VA_ARGS__"),%%ymm3; vfmadd231ps %%ymm1,%%ymm3,"#c3"; vfmadd231ps %%ymm2,%%ymm3,"#c4";"
#define KERNEL_h_k1m8n8 KERNEL_h_k1m8n4 unit_kernel_k1m8n4(%%ymm8,%%ymm9,%%ymm10,%%ymm11,0,8,%1,%%r12,4)
#define KERNEL_k1m8n8 KERNEL_h_k1m8n8 "addq $16,%1;"
#define KERNEL_h_k1m8n12 KERNEL_h_k1m8n8 unit_kernel_k1m8n4(%%ymm12,%%ymm13,%%ymm14,%%ymm15,0,8,%1,%%r12,8)
#define KERNEL_k1m8n12 KERNEL_h_k1m8n12 "addq $16,%1;"
#define KERNEL_k2m8n1 KERNEL_k1m8n1 KERNEL_k1m8n1
#define KERNEL_k2m8n2 KERNEL_k1m8n2 KERNEL_k1m8n2
#define KERNEL_k2m8n4 KERNEL_k1m8n4 KERNEL_k1m8n4
#define KERNEL_k2m8n8 KERNEL_k1m8n8 KERNEL_k1m8n8
#define KERNEL_k2m8n12 \
"vmovsldup (%0),%%ymm1; vmovshdup (%0),%%ymm2;"\
unit_kernel_k1m8n4(%%ymm4,%%ymm5,%%ymm6,%%ymm7,0,8,%1)\
unit_kernel_k1m8n4(%%ymm8,%%ymm9,%%ymm10,%%ymm11,0,8,%1,%%r12,4)\
unit_kernel_k1m8n4(%%ymm12,%%ymm13,%%ymm14,%%ymm15,0,8,%1,%%r12,8)\
"vmovsldup 32(%0),%%ymm1; vmovshdup 32(%0),%%ymm2; prefetcht0 512(%0); addq $64,%0;"\
unit_kernel_k1m8n4(%%ymm4,%%ymm5,%%ymm6,%%ymm7,16,24,%1)\
unit_kernel_k1m8n4(%%ymm8,%%ymm9,%%ymm10,%%ymm11,16,24,%1,%%r12,4)\
unit_kernel_k1m8n4(%%ymm12,%%ymm13,%%ymm14,%%ymm15,16,24,%1,%%r12,8) "addq $32,%1;"
#if defined TRMMKERNEL && !defined LEFT && !defined TRANSA
#define unit_kernel_endn4_k1m8n8(offa1,offb1,offb2) \
"vmovsldup "#offa1"(%0),%%ymm1; vmovshdup "#offa1"(%0),%%ymm2;"\
unit_kernel_k1m8n4(%%ymm8,%%ymm9,%%ymm10,%%ymm11,offb1,offb2,%1,%%r12,4)
#define unit_kernel_endn4_k1m8n12(offa1,offb1,offb2) \
"vmovsldup "#offa1"(%0),%%ymm1; vmovshdup "#offa1"(%0),%%ymm2;"\
unit_kernel_k1m8n4(%%ymm12,%%ymm13,%%ymm14,%%ymm15,offb1,offb2,%1,%%r12,8)
#define unit_kernel_endn8_k1m8n12(offa1,offb1,offb2) unit_kernel_endn4_k1m8n8(offa1,offb1,offb2)\
unit_kernel_k1m8n4(%%ymm12,%%ymm13,%%ymm14,%%ymm15,offb1,offb2,%1,%%r12,8)
#define kernel_kend_m8n8 \
unit_kernel_endn4_k1m8n8(0,0,8) unit_kernel_endn4_k1m8n8(32,16,24)\
unit_kernel_endn4_k1m8n8(64,32,40) unit_kernel_endn4_k1m8n8(96,48,56)
#define kernel_kend_m8n12 \
unit_kernel_endn8_k1m8n12(0,0,8) unit_kernel_endn8_k1m8n12(32,16,24)\
unit_kernel_endn8_k1m8n12(64,32,40) unit_kernel_endn8_k1m8n12(96,48,56)\
unit_kernel_endn4_k1m8n12(128,64,72) unit_kernel_endn4_k1m8n12(160,80,88)\
unit_kernel_endn4_k1m8n12(192,96,104) unit_kernel_endn4_k1m8n12(224,112,120)
#else
#define kernel_kend_m8n8 ""
#define kernel_kend_m8n12 ""
#endif
#define kernel_kend_m8n4 ""
#define kernel_kend_m8n2 ""
#define kernel_kend_m8n1 ""
#define INIT_m8n1 "vpxor %%ymm4,%%ymm4,%%ymm4;"
#define INIT_m8n2 INIT_m8n1 "vpxor %%ymm5,%%ymm5,%%ymm5;"
#define INIT_m8n4 INIT_m8n2 "vpxor %%ymm6,%%ymm6,%%ymm6;vpxor %%ymm7,%%ymm7,%%ymm7;"
#define unit_init_m8n4(c1,c2,c3,c4) \
"vpxor "#c1","#c1","#c1";vpxor "#c2","#c2","#c2";vpxor "#c3","#c3","#c3";vpxor "#c4","#c4","#c4";"
#define INIT_m8n8 INIT_m8n4 unit_init_m8n4(%%ymm8,%%ymm9,%%ymm10,%%ymm11)
#define INIT_m8n12 INIT_m8n8 unit_init_m8n4(%%ymm12,%%ymm13,%%ymm14,%%ymm15)
#define SAVE_m8n1 mult_alpha(%%ymm4,%%ymm0,%2) "vmovups %%ymm4,(%2);"
#define unit_save_m8n2(c1,c2) \
"vunpcklps "#c2","#c1",%%ymm2; vunpckhps "#c2","#c1",%%ymm3; vunpcklpd %%ymm3,%%ymm2,"#c1"; vunpckhpd %%ymm3,%%ymm2,"#c2";"\
mult_alpha(c1,%%ymm0,%5) "vmovups "#c1",(%5);"\
mult_alpha(c2,%%ymm0,%5,%3,1) "vmovups "#c2",(%5,%3,1);"\
"leaq (%5,%3,2),%5;"
#define SAVE_m8n2 "movq %2,%5;" unit_save_m8n2(%%ymm4,%%ymm5)
#define SAVE_m8n4 SAVE_m8n2 unit_save_m8n2(%%ymm6,%%ymm7)
#define SAVE_m8n8 SAVE_m8n4 unit_save_m8n2(%%ymm8,%%ymm9) unit_save_m8n2(%%ymm10,%%ymm11)
#define SAVE_m8n12 SAVE_m8n8 unit_save_m8n2(%%ymm12,%%ymm13) unit_save_m8n2(%%ymm14,%%ymm15)
#define COMPUTE_m8(ndim) \
init_update_kskip(32) INIT_m8n##ndim\
init_set_k "movq %%r14,%1;" init_set_pa_pb_n##ndim(8) "movq %2,%5; movq $0,%%r15;"\
kernel_kstart_n##ndim(8)\
"cmpq $64,%4; jb "#ndim"882f;"\
#ndim"881:\n\t"\
"cmpq $62,%%r15; movq $62,%%r15; cmoveq %3,%%r15;"\
KERNEL_k2m8n##ndim KERNEL_k2m8n##ndim\
"prefetcht1 (%5); subq $31,%5;"\
KERNEL_k2m8n##ndim KERNEL_k2m8n##ndim\
"addq %%r15,%5; prefetcht1 (%7); addq $16,%7;"\
"subq $32,%4; cmpq $64,%4; jnb "#ndim"881b;"\
"movq %2,%5;"\
#ndim"882:\n\t"\
"testq %4,%4; jz "#ndim"883f;"\
"prefetcht0 (%5); prefetcht0 31(%5);"\
KERNEL_k1m8n##ndim\
"prefetcht0 (%5,%3,4); prefetcht0 31(%5,%3,4); addq %3,%5;"\
"subq $4,%4; jmp "#ndim"882b;"\
#ndim"883:\n\t"\
kernel_kend_m8n##ndim "prefetcht0 (%%r14); prefetcht0 64(%%r14);"\
save_set_pa_pb_n##ndim(8) SAVE_m8n##ndim "addq $32,%2;" save_update_kskip(32)

/* m = 4 *//* xmm0 for alpha, xmm1-xmm3 for temporary use, xmm4-xmm15 for accumulators */
#define KERNEL_k1m4n1 \
"vmovups (%0),%%xmm1; addq $16,%0;"\
"vbroadcastss (%1),%%xmm2; vfmadd231ps %%xmm1,%%xmm2,%%xmm4;"\
"addq $4,%1;"
#define KERNEL_h_k1m4n2 \
"vmovsldup (%0),%%xmm1; vmovshdup (%0),%%xmm2; addq $16,%0;"\
"vmovddup (%1),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm4; vfmadd231ps %%xmm2,%%xmm3,%%xmm5;"
#define KERNEL_k1m4n2 KERNEL_h_k1m4n2 "addq $8,%1;"
#define KERNEL_h_k1m4n4 \
KERNEL_h_k1m4n2 "vmovddup 8(%1),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm6; vfmadd231ps %%xmm2,%%xmm3,%%xmm7;"
#define KERNEL_k1m4n4 KERNEL_h_k1m4n4 "addq $16,%1;"
#define unit_kernel_k1m4n4(c1,c2,c3,c4,offb1,offb2,...) \
"vmovddup "#offb1"("#__VA_ARGS__"),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,"#c1"; vfmadd231ps %%xmm2,%%xmm3,"#c2";"\
"vmovddup "#offb2"("#__VA_ARGS__"),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,"#c3"; vfmadd231ps %%xmm2,%%xmm3,"#c4";"
#define KERNEL_h_k1m4n8 KERNEL_h_k1m4n4 unit_kernel_k1m4n4(%%xmm8,%%xmm9,%%xmm10,%%xmm11,0,8,%1,%%r12,4)
#define KERNEL_k1m4n8 KERNEL_h_k1m4n8 "addq $16,%1;"
#define KERNEL_h_k1m4n12 KERNEL_h_k1m4n8 unit_kernel_k1m4n4(%%xmm12,%%xmm13,%%xmm14,%%xmm15,0,8,%1,%%r12,8)
#define KERNEL_k1m4n12 KERNEL_h_k1m4n12 "addq $16,%1;"
#if defined TRMMKERNEL && !defined LEFT && !defined TRANSA
#define unit_kernel_endn4_k1m4n8(offa1,offb1,offb2) \
"vmovsldup "#offa1"(%0),%%xmm1; vmovshdup "#offa1"(%0),%%xmm2;"\
unit_kernel_k1m4n4(%%xmm8,%%xmm9,%%xmm10,%%xmm11,offb1,offb2,%1,%%r12,4)
#define unit_kernel_endn4_k1m4n12(offa1,offb1,offb2) \
"vmovsldup "#offa1"(%0),%%xmm1; vmovshdup "#offa1"(%0),%%xmm2;"\
unit_kernel_k1m4n4(%%xmm12,%%xmm13,%%xmm14,%%xmm15,offb1,offb2,%1,%%r12,8)
#define unit_kernel_endn8_k1m4n12(offa1,offb1,offb2) unit_kernel_endn4_k1m4n8(offa1,offb1,offb2)\
unit_kernel_k1m4n4(%%xmm12,%%xmm13,%%xmm14,%%xmm15,offb1,offb2,%1,%%r12,8)
#define kernel_kend_m4n8 \
unit_kernel_endn4_k1m4n8(0,0,8) unit_kernel_endn4_k1m4n8(16,16,24)\
unit_kernel_endn4_k1m4n8(32,32,40) unit_kernel_endn4_k1m4n8(48,48,56)
#define kernel_kend_m4n12 \
unit_kernel_endn8_k1m4n12(0,0,8) unit_kernel_endn8_k1m4n12(16,16,24)\
unit_kernel_endn8_k1m4n12(32,32,40) unit_kernel_endn8_k1m4n12(48,48,56)\
unit_kernel_endn4_k1m4n12(64,64,72) unit_kernel_endn4_k1m4n12(80,80,88)\
unit_kernel_endn4_k1m4n12(96,96,104) unit_kernel_endn4_k1m4n12(112,112,120)
#else
#define kernel_kend_m4n8 ""
#define kernel_kend_m4n12 ""
#endif
#define kernel_kend_m4n4 ""
#define kernel_kend_m4n2 ""
#define kernel_kend_m4n1 ""
#define INIT_m4n1 "vpxor %%xmm4,%%xmm4,%%xmm4;"
#define INIT_m4n2 INIT_m4n1 "vpxor %%xmm5,%%xmm5,%%xmm5;"
#define INIT_m4n4 INIT_m4n2 "vpxor %%xmm6,%%xmm6,%%xmm6;vpxor %%xmm7,%%xmm7,%%xmm7;"
#define unit_init_m4n4(c1,c2,c3,c4) \
"vpxor "#c1","#c1","#c1";vpxor "#c2","#c2","#c2";vpxor "#c3","#c3","#c3";vpxor "#c4","#c4","#c4";"
#define INIT_m4n8 INIT_m4n4 unit_init_m4n4(%%xmm8,%%xmm9,%%xmm10,%%xmm11)
#define INIT_m4n12 INIT_m4n8 unit_init_m4n4(%%xmm12,%%xmm13,%%xmm14,%%xmm15)
#define SAVE_m4n1 \
mult_alpha(%%xmm4,%%xmm0,%2) "vmovups %%xmm4,(%2);"
#define unit_save_m4n2(c1,c2) \
"vunpcklps "#c2","#c1",%%xmm2; vunpckhps "#c2","#c1",%%xmm3; vunpcklpd %%xmm3,%%xmm2,"#c1"; vunpckhpd %%xmm3,%%xmm2,"#c2";"\
mult_alpha(c1,%%xmm0,%5) "vmovups "#c1",(%5);"\
mult_alpha(c2,%%xmm0,%5,%3,1) "vmovups "#c2",(%5,%3,1);"\
"leaq (%5,%3,2),%5;"
#define SAVE_m4n2 "movq %2,%5;" unit_save_m4n2(%%xmm4,%%xmm5)
#define SAVE_m4n4 SAVE_m4n2 unit_save_m4n2(%%xmm6,%%xmm7)
#define SAVE_m4n8 SAVE_m4n4 unit_save_m4n2(%%xmm8,%%xmm9) unit_save_m4n2(%%xmm10,%%xmm11)
#define SAVE_m4n12 SAVE_m4n8 unit_save_m4n2(%%xmm12,%%xmm13) unit_save_m4n2(%%xmm14,%%xmm15)
#define COMPUTE_m4(ndim) \
init_update_kskip(16) INIT_m4n##ndim\
init_set_k "movq %%r14,%1;" init_set_pa_pb_n##ndim(4)\
kernel_kstart_n##ndim(4)\
#ndim"442:\n\t"\
"testq %4,%4; jz "#ndim"443f;"\
KERNEL_k1m4n##ndim\
"subq $4,%4; jmp "#ndim"442b;"\
#ndim"443:\n\t"\
kernel_kend_m4n##ndim save_set_pa_pb_n##ndim(4) SAVE_m4n##ndim "addq $16,%2;" save_update_kskip(16)

/* m = 2 *//* xmm0 for alpha, xmm1-xmm3 and xmm10 for temporary use, xmm4-xmm9 for accumulators */
#define INIT_m2n1 "vpxor %%xmm4,%%xmm4,%%xmm4;"
#define KERNEL_k1m2n1 \
"vmovsd (%0),%%xmm1; addq $8,%0;"\
"vbroadcastss (%1),%%xmm2; vfmadd231ps %%xmm1,%%xmm2,%%xmm4;"\
"addq $4,%1;"
#ifdef TRMMKERNEL
#define SAVE_m2n1 "vmulps %%xmm4,%%xmm0,%%xmm4; vmovsd %%xmm4,(%2);"
#else
#define SAVE_m2n1 "vmovsd (%2),%%xmm1; vfmadd213ps %%xmm1,%%xmm0,%%xmm4; vmovsd %%xmm4,(%2);"
#endif
#define INIT_m2n2 INIT_m2n1 "vpxor %%xmm5,%%xmm5,%%xmm5;"
#define KERNEL_k1m2n2 \
"vmovsd (%0),%%xmm1; addq $8,%0;"\
"vbroadcastss (%1),%%xmm2; vfmadd231ps %%xmm1,%%xmm2,%%xmm4;"\
"vbroadcastss 4(%1),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm5;"\
"addq $8,%1;"
#ifdef TRMMKERNEL
#define SAVE_m2n2 SAVE_m2n1 "vmulps %%xmm5,%%xmm0,%%xmm5; vmovsd %%xmm5,(%2,%3,1);"
#else
#define SAVE_m2n2 SAVE_m2n1 "vmovsd (%2,%3,1),%%xmm1; vfmadd213ps %%xmm1,%%xmm0,%%xmm5; vmovsd %%xmm5,(%2,%3,1);"
#endif
#define INIT_m2n4 INIT_m2n2
#define INIT_m2n8 INIT_m2n4 "vpxor %%xmm6,%%xmm6,%%xmm6; vpxor %%xmm7,%%xmm7,%%xmm7;"
#define INIT_m2n12 INIT_m2n8 "vpxor %%xmm8,%%xmm8,%%xmm8; vpxor %%xmm9,%%xmm9,%%xmm9;"
#define KERNEL_k1m2n4 \
"vmovups (%1),%%xmm3; addq $16,%1;"\
"vbroadcastss (%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm4;"\
"vbroadcastss 4(%0),%%xmm2; vfmadd231ps %%xmm3,%%xmm2,%%xmm5;"\
"addq $8,%0;"
#define KERNEL_k1m2n8 \
"vmovups (%1),%%xmm3; vmovups (%1,%%r12,4),%%xmm2; addq $16,%1;"\
"vbroadcastss (%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm4; vfmadd231ps %%xmm2,%%xmm1,%%xmm6;"\
"vbroadcastss 4(%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm5; vfmadd231ps %%xmm2,%%xmm1,%%xmm7;"\
"addq $8,%0;"
#define KERNEL_k1m2n12 \
"vmovups (%1),%%xmm3; vmovups (%1,%%r12,4),%%xmm2; vmovups (%1,%%r12,8),%%xmm1; addq $16,%1;"\
"vbroadcastss (%0),%%xmm10; vfmadd231ps %%xmm3,%%xmm10,%%xmm4; vfmadd231ps %%xmm2,%%xmm10,%%xmm6; vfmadd231ps %%xmm1,%%xmm10,%%xmm8;"\
"vbroadcastss 4(%0),%%xmm10; vfmadd231ps %%xmm3,%%xmm10,%%xmm5; vfmadd231ps %%xmm2,%%xmm10,%%xmm7; vfmadd231ps %%xmm1,%%xmm10,%%xmm9;"\
"addq $8,%0;"
#if defined TRMMKERNEL && !defined LEFT && !defined TRANSA
#define unit_kernel_endn4_k1m2n8(aoff1,aoff2,boff) \
"vmovups "#boff"(%1,%%r12,4),%%xmm3;"\
"vbroadcastss "#aoff1"(%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm6;"\
"vbroadcastss "#aoff2"(%0),%%xmm2; vfmadd231ps %%xmm3,%%xmm2,%%xmm7;"
#define unit_kernel_endn4_k1m2n12(aoff1,aoff2,boff) \
"vmovups "#boff"(%1,%%r12,8),%%xmm3;"\
"vbroadcastss "#aoff1"(%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm8;"\
"vbroadcastss "#aoff2"(%0),%%xmm2; vfmadd231ps %%xmm3,%%xmm2,%%xmm9;"
#define unit_kernel_endn8_k1m2n12(aoff1,aoff2,boff) \
"vmovups "#boff"(%1,%%r12,4),%%xmm3; vmovups "#boff"(%1,%%r12,8),%%xmm2;"\
"vbroadcastss "#aoff1"(%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm6; vfmadd231ps %%xmm2,%%xmm1,%%xmm8;"\
"vbroadcastss "#aoff2"(%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm7; vfmadd231ps %%xmm2,%%xmm1,%%xmm9;"
#define kernel_kend_m2n8 \
unit_kernel_endn4_k1m2n8(0,4,0) unit_kernel_endn4_k1m2n8(8,12,16)\
unit_kernel_endn4_k1m2n8(16,20,32) unit_kernel_endn4_k1m2n8(24,28,48)
#define kernel_kend_m2n12 \
unit_kernel_endn8_k1m2n12(0,4,0) unit_kernel_endn8_k1m2n12(8,12,16)\
unit_kernel_endn8_k1m2n12(16,20,32) unit_kernel_endn8_k1m2n12(24,28,48)\
unit_kernel_endn4_k1m2n12(32,36,64) unit_kernel_endn4_k1m2n12(40,44,80)\
unit_kernel_endn4_k1m2n12(48,52,96) unit_kernel_endn4_k1m2n12(56,60,112)
#else
#define kernel_kend_m2n8 ""
#define kernel_kend_m2n12 ""
#endif
#define kernel_kend_m2n4 ""
#define kernel_kend_m2n2 ""
#define kernel_kend_m2n1 ""
#ifdef TRMMKERNEL
#define unit_save_m2n4(c1,c2) \
"vunpcklps "#c2","#c1",%%xmm1; vunpckhps "#c2","#c1",%%xmm2;"\
"vmulps %%xmm1,%%xmm0,%%xmm1; vmovsd %%xmm1,(%5); vmovhpd %%xmm1,(%5,%3,1); leaq (%5,%3,2),%5;"\
"vmulps %%xmm2,%%xmm0,%%xmm2; vmovsd %%xmm2,(%5); vmovhpd %%xmm2,(%5,%3,1); leaq (%5,%3,2),%5;"
#else
#define unit_save_m2n4(c1,c2) \
"vunpcklps "#c2","#c1",%%xmm1; vunpckhps "#c2","#c1",%%xmm2;"\
"vmovsd (%5),%%xmm3; vmovhpd (%5,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm1;"\
"vmovsd %%xmm1,(%5); vmovhpd %%xmm1,(%5,%3,1); leaq (%5,%3,2),%5;"\
"vmovsd (%5),%%xmm3; vmovhpd (%5,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm2;"\
"vmovsd %%xmm2,(%5); vmovhpd %%xmm2,(%5,%3,1); leaq (%5,%3,2),%5;"
#endif
#define SAVE_m2n4 "movq %2,%5;" unit_save_m2n4(%%xmm4,%%xmm5)
#define SAVE_m2n8 SAVE_m2n4 unit_save_m2n4(%%xmm6,%%xmm7)
#define SAVE_m2n12 SAVE_m2n8 unit_save_m2n4(%%xmm8,%%xmm9)
#define COMPUTE_m2(ndim) \
init_update_kskip(8) INIT_m2n##ndim\
init_set_k "movq %%r14,%1;" init_set_pa_pb_n##ndim(2)\
kernel_kstart_n##ndim(2)\
#ndim"222:\n\t"\
"testq %4,%4; jz "#ndim"223f;"\
KERNEL_k1m2n##ndim\
"subq $4,%4; jmp "#ndim"222b;"\
#ndim"223:\n\t"\
kernel_kend_m2n##ndim save_set_pa_pb_n##ndim(2) SAVE_m2n##ndim "addq $8,%2;" save_update_kskip(8)

/* m = 1 *//* xmm0 for alpha, xmm1-xmm3 and xmm10 for temporary use, xmm4-xmm6 for accumulators */
#define INIT_m1n1 "vpxor %%xmm4,%%xmm4,%%xmm4;"
#define KERNEL_k1m1n1 \
"vmovss (%1),%%xmm3; addq $4,%1;"\
"vmovss (%0),%%xmm1; vfmadd231ss %%xmm3,%%xmm1,%%xmm4;"\
"addq $4,%0;"
#ifdef TRMMKERNEL
#define SAVE_m1n1 "vmulss %%xmm4,%%xmm0,%%xmm4; vmovss %%xmm4,(%2);"
#else
#define SAVE_m1n1 "vfmadd213ss (%2),%%xmm0,%%xmm4; vmovss %%xmm4,(%2);"
#endif
#define INIT_m1n2 INIT_m1n1
#define KERNEL_k1m1n2 \
"vmovsd (%1),%%xmm3; addq $8,%1;"\
"vbroadcastss (%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm4;"\
"addq $4,%0;"
#ifdef TRMMKERNEL
#define SAVE_m1n2 \
"vmulps %%xmm4,%%xmm0,%%xmm4; vmovss %%xmm4,(%2); vextractps $1,%%xmm4,(%2,%3,1);"
#else
#define SAVE_m1n2 \
"vmovss (%2),%%xmm3; vinsertps $16,(%2,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm4;"\
"vmovss %%xmm4,(%2); vextractps $1,%%xmm4,(%2,%3,1);"
#endif
#define INIT_m1n4 INIT_m1n2
#define INIT_m1n8 INIT_m1n4 "vpxor %%xmm5,%%xmm5,%%xmm5;"
#define INIT_m1n12 INIT_m1n8 "vpxor %%xmm6,%%xmm6,%%xmm6;"
#define KERNEL_k1m1n4 \
"vmovups (%1),%%xmm3; addq $16,%1;"\
"vbroadcastss (%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm4;"\
"addq $4,%0;"
#define KERNEL_k1m1n8 \
"vmovups (%1),%%xmm3; vmovups (%1,%%r12,4),%%xmm2; addq $16,%1;"\
"vbroadcastss (%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm4; vfmadd231ps %%xmm2,%%xmm1,%%xmm5;"\
"addq $4,%0;"
#define KERNEL_k1m1n12 \
"vmovups (%1),%%xmm3; vmovups (%1,%%r12,4),%%xmm2; vmovups (%1,%%r12,8),%%xmm1; addq $16,%1;"\
"vbroadcastss (%0),%%xmm10; vfmadd231ps %%xmm3,%%xmm10,%%xmm4; vfmadd231ps %%xmm2,%%xmm10,%%xmm5; vfmadd231ps %%xmm1,%%xmm10,%%xmm6;"\
"addq $4,%0;"
#if defined TRMMKERNEL && !defined LEFT && !defined TRANSA
#define unit_kernel_endn4_k1m1n8(aoff,boff) \
"vmovups "#boff"(%1,%%r12,4),%%xmm3;"\
"vbroadcastss "#aoff"(%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm5;"
#define unit_kernel_endn4_k1m1n12(aoff,boff) \
"vmovups "#boff"(%1,%%r12,8),%%xmm3;"\
"vbroadcastss "#aoff"(%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm6;"
#define unit_kernel_endn8_k1m1n12(aoff,boff) \
"vmovups "#boff"(%1,%%r12,4),%%xmm3; vmovups "#boff"(%1,%%r12,8),%%xmm2;"\
"vbroadcastss "#aoff"(%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm5; vfmadd231ps %%xmm2,%%xmm1,%%xmm6;"
#define kernel_kend_m1n8 \
unit_kernel_endn4_k1m1n8(0,0) unit_kernel_endn4_k1m1n8(4,16)\
unit_kernel_endn4_k1m1n8(8,32) unit_kernel_endn4_k1m1n8(12,48)
#define kernel_kend_m1n12 \
unit_kernel_endn8_k1m1n12(0,0) unit_kernel_endn8_k1m1n12(4,16)\
unit_kernel_endn8_k1m1n12(8,32) unit_kernel_endn8_k1m1n12(12,48)\
unit_kernel_endn4_k1m1n12(16,64) unit_kernel_endn4_k1m1n12(20,80)\
unit_kernel_endn4_k1m1n12(24,96) unit_kernel_endn4_k1m1n12(28,112)
#else
#define kernel_kend_m1n8 ""
#define kernel_kend_m1n12 ""
#endif
#define kernel_kend_m1n4 ""
#define kernel_kend_m1n2 ""
#define kernel_kend_m1n1 ""
#ifdef TRMMKERNEL
#define unit_save_m1n4(c1) \
"vpxor %%xmm10,%%xmm10,%%xmm10; vmovsd "#c1",%%xmm10,%%xmm2; vmovhlps "#c1",%%xmm10,%%xmm1;"\
"vmulps %%xmm2,%%xmm0,%%xmm2; vmovss %%xmm2,(%5); vextractps $1,%%xmm2,(%5,%3,1); leaq (%5,%3,2),%5;"\
"vmulps %%xmm1,%%xmm0,%%xmm1; vmovss %%xmm1,(%5); vextractps $1,%%xmm1,(%5,%3,1); leaq (%5,%3,2),%5;"
#else
#define unit_save_m1n4(c1) \
"vpxor %%xmm10,%%xmm10,%%xmm10; vmovsd "#c1",%%xmm10,%%xmm2; vmovhlps "#c1",%%xmm10,%%xmm1;"\
"vmovss (%5),%%xmm3; vinsertps $16,(%5,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm2;"\
"vmovss %%xmm2,(%5); vextractps $1,%%xmm2,(%5,%3,1); leaq (%5,%3,2),%5;"\
"vmovss (%5),%%xmm3; vinsertps $16,(%5,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm1;"\
"vmovss %%xmm1,(%5); vextractps $1,%%xmm1,(%5,%3,1); leaq (%5,%3,2),%5;"
#endif
#define SAVE_m1n4 "movq %2,%5;" unit_save_m1n4(%%xmm4)
#define SAVE_m1n8 SAVE_m1n4 unit_save_m1n4(%%xmm5)
#define SAVE_m1n12 SAVE_m1n8 unit_save_m1n4(%%xmm6)
#define COMPUTE_m1(ndim) \
init_update_kskip(4) INIT_m1n##ndim\
init_set_k "movq %%r14,%1;" init_set_pa_pb_n##ndim(1)\
kernel_kstart_n##ndim(1)\
#ndim"112:\n\t"\
"testq %4,%4; jz "#ndim"113f;"\
KERNEL_k1m1n##ndim\
"subq $4,%4; jmp "#ndim"112b;"\
#ndim"113:\n\t"\
kernel_kend_m1n##ndim save_set_pa_pb_n##ndim(1) SAVE_m1n##ndim "addq $4,%2;" save_update_kskip(4)

#define COMPUTE(ndim) {\
HEAD_SET_OFFSET(ndim) next_b = b_pointer + ndim * K;\
__asm__ __volatile__(\
"vbroadcastss (%6),%%ymm0;"\
"movq %4,%%r12; salq $2,%%r12; movq %1,%%r14; movq %8,%%r11;" INIT_SET_KSKIP\
"cmpq $8,%%r11;jb 33101"#ndim"f;"\
"33109"#ndim":\n\t"\
COMPUTE_m8(ndim)\
"subq $8,%%r11;cmpq $8,%%r11;jnb 33109"#ndim"b;"\
"33101"#ndim":\n\t"\
"cmpq $4,%%r11;jb 33103"#ndim"f;"\
COMPUTE_m4(ndim)\
"subq $4,%%r11;"\
"33103"#ndim":\n\t"\
"cmpq $2,%%r11;jb 33104"#ndim"f;"\
COMPUTE_m2(ndim)\
"subq $2,%%r11;"\
"33104"#ndim":\n\t"\
"testq %%r11,%%r11;jz 33105"#ndim"f;"\
COMPUTE_m1(ndim)\
"33105"#ndim":\n\t"\
"movq %%r12,%4; sarq $2,%4; movq %%r14,%1; vzeroupper;"\
:"+r"(a_pointer),"+r"(b_pointer),"+r"(c_pointer),"+r"(ldc_in_bytes),"+r"(K),"+r"(ctemp),"+r"(const_val),"+r"(next_b)\
:"m"(M),"m"(off):"r11","r12","r13","r14","r15",\
"xmm0","xmm1","xmm2","xmm3","xmm4","xmm5","xmm6","xmm7","xmm8","xmm9","xmm10","xmm11","xmm12","xmm13","xmm14","xmm15","cc","memory");\
TAIL_SET_OFFSET(ndim) a_pointer -= M * K; b_pointer += ndim * K; c_pointer += (LDC * ndim - M);\
}

#include "common.h"
#include <stdint.h>
int __attribute__ ((noinline))
CNAME(BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float * __restrict__ A, float * __restrict__ B, float * __restrict__ C, BLASLONG LDC
#ifdef TRMMKERNEL
,BLASLONG offset
#endif
){
if(m==0||n==0||k==0||alpha==0.0) return 0;
int64_t ldc_in_bytes = (int64_t)LDC * sizeof(float);
float constval = alpha;
float *const_val=&constval;
int64_t M = (int64_t)m, K = (int64_t)k, off = 0;
#ifdef TRMMKERNEL
#ifdef LEFT
off = offset;
#else
off = -offset;
#endif
#endif
BLASLONG n_count = n;
float *a_pointer = A,*b_pointer = B,*c_pointer = C,*ctemp = C,*next_b = B;
for(;n_count>11;n_count-=12) COMPUTE(12)
for(;n_count>7;n_count-=8) COMPUTE(8)
for(;n_count>3;n_count-=4) COMPUTE(4)
for(;n_count>1;n_count-=2) COMPUTE(2)
if(n_count>0) COMPUTE(1)
return 0;
}

Write Preview
Loading…
Cancel
Save