/*********************************************************************/ /* Copyright 2009, 2010 The University of Texas at Austin. */ /* All rights reserved. */ /* */ /* Redistribution and use in source and binary forms, with or */ /* without modification, are permitted provided that the following */ /* conditions are met: */ /* */ /* 1. Redistributions of source code must retain the above */ /* copyright notice, this list of conditions and the following */ /* disclaimer. */ /* */ /* 2. Redistributions in binary form must reproduce the above */ /* copyright notice, this list of conditions and the following */ /* disclaimer in the documentation and/or other materials */ /* provided with the distribution. */ /* */ /* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */ /* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */ /* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */ /* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */ /* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */ /* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */ /* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */ /* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */ /* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */ /* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */ /* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */ /* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */ /* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */ /* POSSIBILITY OF SUCH DAMAGE. */ /* */ /* The views and conclusions contained in the software and */ /* documentation are those of the authors and should not be */ /* interpreted as representing official policies, either expressed */ /* or implied, of The University of Texas at Austin. */ /*********************************************************************/ #define ASSEMBLER #include "common.h" #define OLD_M %rdi #define OLD_N %rsi #define OLD_K %rdx #define M %r13 #define N %r14 #define K %r15 #define A %rcx #define B %r8 #define C %r9 #define LDC %r10 #define I %r11 #define AO %rdi #define BO %rsi #define CO1 %rbx #define CO2 %rbp #define BB %r12 #define PREA %rdx #ifndef WINDOWS_ABI #define STACKSIZE 128 #define OLD_LDC 8 + STACKSIZE(%rsp) #define OLD_OFFSET 16 + STACKSIZE(%rsp) #define ALPHA_R 48(%rsp) #define ALPHA_I 56(%rsp) #define J 64(%rsp) #define OFFSET 72(%rsp) #define KK 80(%rsp) #define KKK 88(%rsp) #else #define STACKSIZE 512 #define OLD_ALPHA_I 40 + STACKSIZE(%rsp) #define OLD_A 48 + STACKSIZE(%rsp) #define OLD_B 56 + STACKSIZE(%rsp) #define OLD_C 64 + STACKSIZE(%rsp) #define OLD_LDC 72 + STACKSIZE(%rsp) #define OLD_OFFSET 80 + STACKSIZE(%rsp) #define ALPHA_R 224(%rsp) #define ALPHA_I 232(%rsp) #define J 240(%rsp) #define OFFSET 248(%rsp) #define KK 256(%rsp) #define KKK 264(%rsp) #endif #define PREFETCHSIZE (8 * 17 + 4) #define PREFETCH prefetcht0 #if defined(OS_LINUX) && defined(CORE_BARCELONA) .align 32768 #endif PROLOGUE PROFCODE subq $STACKSIZE, %rsp movq %rbx, 0(%rsp) movq %rbp, 8(%rsp) movq %r12, 16(%rsp) movq %r13, 24(%rsp) movq %r14, 32(%rsp) movq %r15, 40(%rsp) #ifdef WINDOWS_ABI movq %rdi, 48(%rsp) movq %rsi, 56(%rsp) movups %xmm6, 64(%rsp) movups %xmm7, 80(%rsp) movups %xmm8, 96(%rsp) movups %xmm9, 112(%rsp) movups %xmm10, 128(%rsp) movups %xmm11, 144(%rsp) movups %xmm12, 160(%rsp) movups %xmm13, 176(%rsp) movups %xmm14, 192(%rsp) movups %xmm15, 208(%rsp) movq ARG1, OLD_M movq ARG2, OLD_N movq ARG3, OLD_K movq OLD_A, A movq OLD_B, B movq OLD_C, C movq OLD_LDC, LDC #ifdef TRMMKERNEL movq OLD_OFFSET, %r11 #endif movaps %xmm3, %xmm0 movss OLD_ALPHA_I, %xmm1 #else movq OLD_LDC, LDC #ifdef TRMMKERNEL movq OLD_OFFSET, %r11 #endif #endif unpcklps %xmm1, %xmm0 movlps %xmm0, ALPHA_R movlps %xmm0, ALPHA_I subq $-32 * SIZE, A subq $-32 * SIZE, B movq OLD_M, M movq OLD_N, N movq OLD_K, K salq $ZBASE_SHIFT, LDC movq N, J sarq $2, J NOBRANCH jle .L50 ALIGN_4 .L10: #if defined(TRMMKERNEL) && defined(LEFT) movq OFFSET, %rax movq %rax, KK #endif movq C, CO1 leaq (C, LDC, 1), CO2 movq A, AO movq K, %rax salq $BASE_SHIFT + 2, %rax leaq (B, %rax), BB movq M, I sarq $3, I NOBRANCH jle .L20 ALIGN_4 .L11: #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movq B, BO #else movq B, BO movq KK, %rax leaq (, %rax, SIZE), %rax leaq (AO, %rax, 8), AO leaq (BO, %rax, 4), BO #endif movaps -32 * SIZE(AO), %xmm0 xorpd %xmm3, %xmm3 movaps -28 * SIZE(AO), %xmm1 xorpd %xmm4, %xmm4 movaps -32 * SIZE(BO), %xmm2 xorpd %xmm5, %xmm5 prefetcht0 -32 * SIZE(BB) xorpd %xmm6, %xmm6 prefetcht2 7 * SIZE(CO1) movapd %xmm4, %xmm8 movapd %xmm4, %xmm9 prefetcht2 7 * SIZE(CO2) movapd %xmm4, %xmm10 movapd %xmm4, %xmm11 prefetcht2 7 * SIZE(CO1, LDC, 2) movapd %xmm4, %xmm12 movaps %xmm4, %xmm13 prefetcht2 7 * SIZE(CO2, LDC, 2) movaps %xmm4, %xmm14 movaps %xmm4, %xmm15 subq $-24 * SIZE, BB leaq (PREFETCHSIZE + 0) * SIZE(AO), PREA #ifndef TRMMKERNEL movq K, %rax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movq K, %rax subq KK, %rax movq %rax, KKK #else movq KK, %rax #ifdef LEFT addq $8, %rax #else addq $4, %rax #endif movq %rax, KKK #endif sarq $3, %rax NOBRANCH jle .L15 ALIGN_3 .L12: PREFETCH -32 * SIZE(PREA) addps %xmm6, %xmm10 addps %xmm3, %xmm14 movaps %xmm2, %xmm3 pshufd $0x39, %xmm2, %xmm7 mulps %xmm0, %xmm2 mulps %xmm1, %xmm3 addps %xmm4, %xmm11 addps %xmm5, %xmm15 movaps %xmm7, %xmm5 pshufd $0x39, %xmm7, %xmm6 mulps %xmm0, %xmm7 mulps %xmm1, %xmm5 addps %xmm2, %xmm8 movaps -28 * SIZE(BO), %xmm2 addps %xmm3, %xmm12 movaps %xmm6, %xmm3 pshufd $0x39, %xmm6, %xmm4 mulps %xmm0, %xmm6 mulps %xmm1, %xmm3 addps %xmm7, %xmm9 addps %xmm5, %xmm13 movaps %xmm4, %xmm5 mulps %xmm0, %xmm4 movaps -24 * SIZE(AO), %xmm0 mulps %xmm1, %xmm5 movaps -20 * SIZE(AO), %xmm1 addps %xmm6, %xmm10 addps %xmm3, %xmm14 movaps %xmm2, %xmm3 pshufd $0x39, %xmm2, %xmm7 mulps %xmm0, %xmm2 mulps %xmm1, %xmm3 addps %xmm4, %xmm11 addps %xmm5, %xmm15 movaps %xmm7, %xmm5 pshufd $0x39, %xmm7, %xmm6 mulps %xmm0, %xmm7 mulps %xmm1, %xmm5 addps %xmm2, %xmm8 movaps -24 * SIZE(BO), %xmm2 addps %xmm3, %xmm12 movaps %xmm6, %xmm3 pshufd $0x39, %xmm6, %xmm4 mulps %xmm0, %xmm6 mulps %xmm1, %xmm3 addps %xmm7, %xmm9 addps %xmm5, %xmm13 movaps %xmm4, %xmm5 mulps %xmm0, %xmm4 movaps -16 * SIZE(AO), %xmm0 mulps %xmm1, %xmm5 movaps -12 * SIZE(AO), %xmm1 addps %xmm6, %xmm10 addps %xmm3, %xmm14 movaps %xmm2, %xmm3 pshufd $0x39, %xmm2, %xmm7 mulps %xmm0, %xmm2 mulps %xmm1, %xmm3 addps %xmm4, %xmm11 addps %xmm5, %xmm15 PREFETCH -16 * SIZE(PREA) movaps %xmm7, %xmm5 pshufd $0x39, %xmm7, %xmm6 mulps %xmm0, %xmm7 mulps %xmm1, %xmm5 addps %xmm2, %xmm8 movaps -20 * SIZE(BO), %xmm2 addps %xmm3, %xmm12 movaps %xmm6, %xmm3 pshufd $0x39, %xmm6, %xmm4 mulps %xmm0, %xmm6 mulps %xmm1, %xmm3 addps %xmm7, %xmm9 addps %xmm5, %xmm13 movaps %xmm4, %xmm5 mulps %xmm0, %xmm4 movaps -8 * SIZE(AO), %xmm0 mulps %xmm1, %xmm5 movaps -4 * SIZE(AO), %xmm1 addps %xmm6, %xmm10 addps %xmm3, %xmm14 movaps %xmm2, %xmm3 pshufd $0x39, %xmm2, %xmm7 mulps %xmm0, %xmm2 mulps %xmm1, %xmm3 addps %xmm4, %xmm11 addps %xmm5, %xmm15 movaps %xmm7, %xmm5 pshufd $0x39, %xmm7, %xmm6 mulps %xmm0, %xmm7 mulps %xmm1, %xmm5 addps %xmm2, %xmm8 movaps -16 * SIZE(BO), %xmm2 addps %xmm3, %xmm12 movaps %xmm6, %xmm3 pshufd $0x39, %xmm6, %xmm4 mulps %xmm0, %xmm6 mulps %xmm1, %xmm3 addps %xmm7, %xmm9 addps %xmm5, %xmm13 movaps %xmm4, %xmm5 mulps %xmm0, %xmm4 movaps 0 * SIZE(AO), %xmm0 mulps %xmm1, %xmm5 movaps 4 * SIZE(AO), %xmm1 addps %xmm6, %xmm10 addps %xmm3, %xmm14 PREFETCH 0 * SIZE(PREA) movaps %xmm2, %xmm3 pshufd $0x39, %xmm2, %xmm7 mulps %xmm0, %xmm2 mulps %xmm1, %xmm3 addps %xmm4, %xmm11 addps %xmm5, %xmm15 movaps %xmm7, %xmm5 pshufd $0x39, %xmm7, %xmm6 mulps %xmm0, %xmm7 mulps %xmm1, %xmm5 addps %xmm2, %xmm8 movaps -12 * SIZE(BO), %xmm2 addps %xmm3, %xmm12 movaps %xmm6, %xmm3 pshufd $0x39, %xmm6, %xmm4 mulps %xmm0, %xmm6 mulps %xmm1, %xmm3 addps %xmm7, %xmm9 addps %xmm5, %xmm13 movaps %xmm4, %xmm5 mulps %xmm0, %xmm4 movaps 8 * SIZE(AO), %xmm0 mulps %xmm1, %xmm5 movaps 12 * SIZE(AO), %xmm1 addps %xmm6, %xmm10 addps %xmm3, %xmm14 movaps %xmm2, %xmm3 pshufd $0x39, %xmm2, %xmm7 mulps %xmm0, %xmm2 mulps %xmm1, %xmm3 addps %xmm4, %xmm11 addps %xmm5, %xmm15 movaps %xmm7, %xmm5 pshufd $0x39, %xmm7, %xmm6 mulps %xmm0, %xmm7 mulps %xmm1, %xmm5 addps %xmm2, %xmm8 movaps -8 * SIZE(BO), %xmm2 addps %xmm3, %xmm12 movaps %xmm6, %xmm3 pshufd $0x39, %xmm6, %xmm4 mulps %xmm0, %xmm6 mulps %xmm1, %xmm3 addps %xmm7, %xmm9 addps %xmm5, %xmm13 movaps %xmm4, %xmm5 mulps %xmm0, %xmm4 movaps 16 * SIZE(AO), %xmm0 mulps %xmm1, %xmm5 movaps 20 * SIZE(AO), %xmm1 addps %xmm6, %xmm10 addps %xmm3, %xmm14 PREFETCH 16 * SIZE(PREA) movaps %xmm2, %xmm3 pshufd $0x39, %xmm2, %xmm7 mulps %xmm0, %xmm2 mulps %xmm1, %xmm3 addps %xmm4, %xmm11 addps %xmm5, %xmm15 movaps %xmm7, %xmm5 pshufd $0x39, %xmm7, %xmm6 mulps %xmm0, %xmm7 mulps %xmm1, %xmm5 addps %xmm2, %xmm8 movaps -4 * SIZE(BO), %xmm2 addps %xmm3, %xmm12 movaps %xmm6, %xmm3 pshufd $0x39, %xmm6, %xmm4 mulps %xmm0, %xmm6 mulps %xmm1, %xmm3 addps %xmm7, %xmm9 addps %xmm5, %xmm13 movaps %xmm4, %xmm5 mulps %xmm0, %xmm4 movaps 24 * SIZE(AO), %xmm0 mulps %xmm1, %xmm5 movaps 28 * SIZE(AO), %xmm1 addps %xmm6, %xmm10 addps %xmm3, %xmm14 movaps %xmm2, %xmm3 pshufd $0x39, %xmm2, %xmm7 mulps %xmm0, %xmm2 mulps %xmm1, %xmm3 addps %xmm4, %xmm11 addps %xmm5, %xmm15 movaps %xmm7, %xmm5 subq $-64 * SIZE, AO pshufd $0x39, %xmm7, %xmm6 mulps %xmm0, %xmm7 mulps %xmm1, %xmm5 addps %xmm2, %xmm8 movaps 0 * SIZE(BO), %xmm2 addps %xmm3, %xmm12 movaps %xmm6, %xmm3 subq $-32 * SIZE, BO pshufd $0x39, %xmm6, %xmm4 mulps %xmm0, %xmm6 mulps %xmm1, %xmm3 addps %xmm7, %xmm9 addps %xmm5, %xmm13 movaps %xmm4, %xmm5 mulps %xmm0, %xmm4 movaps -32 * SIZE(AO), %xmm0 mulps %xmm1, %xmm5 movaps -28 * SIZE(AO), %xmm1 subq $-64 * SIZE, PREA subq $1, %rax BRANCH jg .L12 ALIGN_3 .L15: prefetcht0 -16 * SIZE(BB) #ifndef TRMMKERNEL movq K, %rax #else movq KKK, %rax #endif andq $7, %rax # if (k & 1) BRANCH je .L18 ALIGN_3 .L16: addps %xmm6, %xmm10 addps %xmm3, %xmm14 movaps %xmm2, %xmm3 pshufd $0x39, %xmm2, %xmm7 mulps %xmm0, %xmm2 mulps %xmm1, %xmm3 addps %xmm4, %xmm11 addps %xmm5, %xmm15 movaps %xmm7, %xmm5 pshufd $0x39, %xmm7, %xmm6 mulps %xmm0, %xmm7 mulps %xmm1, %xmm5 addps %xmm2, %xmm8 movaps -28 * SIZE(BO), %xmm2 addps %xmm3, %xmm12 movaps %xmm6, %xmm3 pshufd $0x39, %xmm6, %xmm4 mulps %xmm0, %xmm6 mulps %xmm1, %xmm3 addps %xmm7, %xmm9 addps %xmm5, %xmm13 movaps %xmm4, %xmm5 mulps %xmm0, %xmm4 movaps -24 * SIZE(AO), %xmm0 mulps %xmm1, %xmm5 movaps -20 * SIZE(AO), %xmm1 addq $8 * SIZE, AO addq $4 * SIZE, BO subq $1, %rax BRANCH jg .L16 ALIGN_3 .L18: movups ALPHA_R, %xmm7 addps %xmm6, %xmm10 addps %xmm3, %xmm14 addps %xmm4, %xmm11 addps %xmm5, %xmm15 movaps %xmm9, %xmm4 shufps $0xd8, %xmm8, %xmm9 shufps $0xd8, %xmm11, %xmm8 shufps $0xd8, %xmm10, %xmm11 shufps $0xd8, %xmm4, %xmm10 movaps %xmm8, %xmm4 shufps $0xd8, %xmm10, %xmm8 shufps $0xd8, %xmm4, %xmm10 movaps %xmm9, %xmm5 shufps $0xd8, %xmm11, %xmm9 shufps $0xd8, %xmm5, %xmm11 movaps %xmm13, %xmm4 shufps $0xd8, %xmm12, %xmm13 shufps $0xd8, %xmm15, %xmm12 shufps $0xd8, %xmm14, %xmm15 shufps $0xd8, %xmm4, %xmm14 movaps %xmm12, %xmm4 shufps $0xd8, %xmm14, %xmm12 shufps $0xd8, %xmm4, %xmm14 movaps %xmm13, %xmm5 shufps $0xd8, %xmm15, %xmm13 shufps $0xd8, %xmm5, %xmm15 movsd 0 * SIZE(CO1), %xmm0 movhps 2 * SIZE(CO1), %xmm0 movsd 4 * SIZE(CO1), %xmm1 movhps 6 * SIZE(CO1), %xmm1 movsd 8 * SIZE(CO1), %xmm2 movhps 10 * SIZE(CO1), %xmm2 movsd 12 * SIZE(CO1), %xmm3 movhps 14 * SIZE(CO1), %xmm3 pshufd $0x50, %xmm8, %xmm4 pshufd $0xfa, %xmm8, %xmm8 pshufd $0x50, %xmm12, %xmm5 pshufd $0xfa, %xmm12, %xmm12 mulps %xmm7, %xmm4 mulps %xmm7, %xmm8 mulps %xmm7, %xmm5 mulps %xmm7, %xmm12 addps %xmm0, %xmm4 addps %xmm1, %xmm8 addps %xmm2, %xmm5 addps %xmm3, %xmm12 movlps %xmm4, 0 * SIZE(CO1) movhps %xmm4, 2 * SIZE(CO1) movlps %xmm8, 4 * SIZE(CO1) movhps %xmm8, 6 * SIZE(CO1) movlps %xmm5, 8 * SIZE(CO1) movhps %xmm5, 10 * SIZE(CO1) movlps %xmm12, 12 * SIZE(CO1) movhps %xmm12, 14 * SIZE(CO1) movsd 0 * SIZE(CO2), %xmm0 movhps 2 * SIZE(CO2), %xmm0 movsd 4 * SIZE(CO2), %xmm1 movhps 6 * SIZE(CO2), %xmm1 movsd 8 * SIZE(CO2), %xmm2 movhps 10 * SIZE(CO2), %xmm2 movsd 12 * SIZE(CO2), %xmm3 movhps 14 * SIZE(CO2), %xmm3 pshufd $0x50, %xmm9, %xmm4 pshufd $0xfa, %xmm9, %xmm9 pshufd $0x50, %xmm13, %xmm5 pshufd $0xfa, %xmm13, %xmm13 mulps %xmm7, %xmm4 mulps %xmm7, %xmm9 mulps %xmm7, %xmm5 mulps %xmm7, %xmm13 addps %xmm0, %xmm4 addps %xmm1, %xmm9 addps %xmm2, %xmm5 addps %xmm3, %xmm13 movlps %xmm4, 0 * SIZE(CO2) movhps %xmm4, 2 * SIZE(CO2) movlps %xmm9, 4 * SIZE(CO2) movhps %xmm9, 6 * SIZE(CO2) movlps %xmm5, 8 * SIZE(CO2) movhps %xmm5, 10 * SIZE(CO2) movlps %xmm13, 12 * SIZE(CO2) movhps %xmm13, 14 * SIZE(CO2) movsd 0 * SIZE(CO1, LDC, 2), %xmm0 movhps 2 * SIZE(CO1, LDC, 2), %xmm0 movsd 4 * SIZE(CO1, LDC, 2), %xmm1 movhps 6 * SIZE(CO1, LDC, 2), %xmm1 movsd 8 * SIZE(CO1, LDC, 2), %xmm2 movhps 10 * SIZE(CO1, LDC, 2), %xmm2 movsd 12 * SIZE(CO1, LDC, 2), %xmm3 movhps 14 * SIZE(CO1, LDC, 2), %xmm3 pshufd $0x50, %xmm10, %xmm4 pshufd $0xfa, %xmm10, %xmm10 pshufd $0x50, %xmm14, %xmm5 pshufd $0xfa, %xmm14, %xmm14 mulps %xmm7, %xmm4 mulps %xmm7, %xmm10 mulps %xmm7, %xmm5 mulps %xmm7, %xmm14 addps %xmm0, %xmm4 addps %xmm1, %xmm10 addps %xmm2, %xmm5 addps %xmm3, %xmm14 movlps %xmm4, 0 * SIZE(CO1, LDC, 2) movhps %xmm4, 2 * SIZE(CO1, LDC, 2) movlps %xmm10, 4 * SIZE(CO1, LDC, 2) movhps %xmm10, 6 * SIZE(CO1, LDC, 2) movlps %xmm5, 8 * SIZE(CO1, LDC, 2) movhps %xmm5, 10 * SIZE(CO1, LDC, 2) movlps %xmm14, 12 * SIZE(CO1, LDC, 2) movhps %xmm14, 14 * SIZE(CO1, LDC, 2) movsd 0 * SIZE(CO2, LDC, 2), %xmm0 movhps 2 * SIZE(CO2, LDC, 2), %xmm0 movsd 4 * SIZE(CO2, LDC, 2), %xmm1 movhps 6 * SIZE(CO2, LDC, 2), %xmm1 movsd 8 * SIZE(CO2, LDC, 2), %xmm2 movhps 10 * SIZE(CO2, LDC, 2), %xmm2 movsd 12 * SIZE(CO2, LDC, 2), %xmm3 movhps 14 * SIZE(CO2, LDC, 2), %xmm3 pshufd $0x50, %xmm11, %xmm4 pshufd $0xfa, %xmm11, %xmm11 pshufd $0x50, %xmm15, %xmm5 pshufd $0xfa, %xmm15, %xmm15 mulps %xmm7, %xmm4 mulps %xmm7, %xmm11 mulps %xmm7, %xmm5 mulps %xmm7, %xmm15 addps %xmm0, %xmm4 addps %xmm1, %xmm11 addps %xmm2, %xmm5 addps %xmm3, %xmm15 movlps %xmm4, 0 * SIZE(CO2, LDC, 2) movhps %xmm4, 2 * SIZE(CO2, LDC, 2) movlps %xmm11, 4 * SIZE(CO2, LDC, 2) movhps %xmm11, 6 * SIZE(CO2, LDC, 2) movlps %xmm5, 8 * SIZE(CO2, LDC, 2) movhps %xmm5, 10 * SIZE(CO2, LDC, 2) movlps %xmm15, 12 * SIZE(CO2, LDC, 2) movhps %xmm15, 14 * SIZE(CO2, LDC, 2) addq $16 * SIZE, CO1 addq $16 * SIZE, CO2 decq I BRANCH jg .L11 ALIGN_4 .L20: testq $4, M BRANCH jle .L30 #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movq B, BO #else movq B, BO movq KK, %rax leaq (, %rax, SIZE), %rax leaq (AO, %rax, 4), AO leaq (BO, %rax, 4), BO #endif movaps -32 * SIZE(AO), %xmm0 xorps %xmm3, %xmm3 xorps %xmm4, %xmm4 movaps -32 * SIZE(BO), %xmm2 xorps %xmm5, %xmm5 xorps %xmm6, %xmm6 movaps %xmm4, %xmm8 movaps %xmm4, %xmm9 movaps %xmm4, %xmm10 movaps %xmm4, %xmm11 #ifndef TRMMKERNEL movq K, %rax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movq K, %rax subq KK, %rax movq %rax, KKK #else movq KK, %rax #ifdef LEFT addq $4, %rax #else addq $4, %rax #endif movq %rax, KKK #endif sarq $2, %rax NOBRANCH jle .L25 ALIGN_3 .L22: addps %xmm6, %xmm10 PREFETCH (PREFETCHSIZE + 0) * SIZE(AO) pshufd $0x39, %xmm2, %xmm7 mulps %xmm0, %xmm2 addps %xmm4, %xmm11 pshufd $0x39, %xmm7, %xmm6 mulps %xmm0, %xmm7 addps %xmm2, %xmm8 movaps -28 * SIZE(BO), %xmm2 pshufd $0x39, %xmm6, %xmm4 mulps %xmm0, %xmm6 addps %xmm7, %xmm9 mulps %xmm0, %xmm4 movaps -28 * SIZE(AO), %xmm0 addps %xmm6, %xmm10 pshufd $0x39, %xmm2, %xmm7 mulps %xmm0, %xmm2 addps %xmm4, %xmm11 pshufd $0x39, %xmm7, %xmm6 mulps %xmm0, %xmm7 addps %xmm2, %xmm8 movaps -24 * SIZE(BO), %xmm2 pshufd $0x39, %xmm6, %xmm4 mulps %xmm0, %xmm6 addps %xmm7, %xmm9 mulps %xmm0, %xmm4 movaps -24 * SIZE(AO), %xmm0 addps %xmm6, %xmm10 pshufd $0x39, %xmm2, %xmm7 mulps %xmm0, %xmm2 addps %xmm4, %xmm11 pshufd $0x39, %xmm7, %xmm6 mulps %xmm0, %xmm7 addps %xmm2, %xmm8 movaps -20 * SIZE(BO), %xmm2 pshufd $0x39, %xmm6, %xmm4 mulps %xmm0, %xmm6 addps %xmm7, %xmm9 mulps %xmm0, %xmm4 movaps -20 * SIZE(AO), %xmm0 addps %xmm6, %xmm10 pshufd $0x39, %xmm2, %xmm7 mulps %xmm0, %xmm2 addps %xmm4, %xmm11 pshufd $0x39, %xmm7, %xmm6 mulps %xmm0, %xmm7 addps %xmm2, %xmm8 movaps -16 * SIZE(BO), %xmm2 pshufd $0x39, %xmm6, %xmm4 mulps %xmm0, %xmm6 subq $-16 * SIZE, AO addps %xmm7, %xmm9 mulps %xmm0, %xmm4 movaps -32 * SIZE(AO), %xmm0 subq $-16 * SIZE, BO subq $1, %rax BRANCH jg .L22 ALIGN_3 .L25: #ifndef TRMMKERNEL movq K, %rax #else movq KKK, %rax #endif andq $3, %rax BRANCH je .L28 ALIGN_3 .L26: addps %xmm6, %xmm10 pshufd $0x39, %xmm2, %xmm7 mulps %xmm0, %xmm2 addps %xmm4, %xmm11 pshufd $0x39, %xmm7, %xmm6 mulps %xmm0, %xmm7 addps %xmm2, %xmm8 movaps -28 * SIZE(BO), %xmm2 pshufd $0x39, %xmm6, %xmm4 mulps %xmm0, %xmm6 addps %xmm7, %xmm9 mulps %xmm0, %xmm4 movaps -28 * SIZE(AO), %xmm0 addq $4 * SIZE, AO addq $4 * SIZE, BO subq $1, %rax BRANCH jg .L26 ALIGN_3 .L28: movups ALPHA_R, %xmm7 addps %xmm6, %xmm10 addps %xmm4, %xmm11 movaps %xmm9, %xmm4 shufps $0xd8, %xmm8, %xmm9 shufps $0xd8, %xmm11, %xmm8 shufps $0xd8, %xmm10, %xmm11 shufps $0xd8, %xmm4, %xmm10 movaps %xmm8, %xmm4 shufps $0xd8, %xmm10, %xmm8 shufps $0xd8, %xmm4, %xmm10 movaps %xmm9, %xmm5 shufps $0xd8, %xmm11, %xmm9 shufps $0xd8, %xmm5, %xmm11 movsd 0 * SIZE(CO1), %xmm0 movhps 2 * SIZE(CO1), %xmm0 movsd 4 * SIZE(CO1), %xmm1 movhps 6 * SIZE(CO1), %xmm1 pshufd $0x50, %xmm8, %xmm4 pshufd $0xfa, %xmm8, %xmm8 mulps %xmm7, %xmm4 mulps %xmm7, %xmm8 addps %xmm0, %xmm4 addps %xmm1, %xmm8 movlps %xmm4, 0 * SIZE(CO1) movhps %xmm4, 2 * SIZE(CO1) movlps %xmm8, 4 * SIZE(CO1) movhps %xmm8, 6 * SIZE(CO1) movsd 0 * SIZE(CO2), %xmm0 movhps 2 * SIZE(CO2), %xmm0 movsd 4 * SIZE(CO2), %xmm1 movhps 6 * SIZE(CO2), %xmm1 pshufd $0x50, %xmm9, %xmm4 pshufd $0xfa, %xmm9, %xmm9 mulps %xmm7, %xmm4 mulps %xmm7, %xmm9 addps %xmm0, %xmm4 addps %xmm1, %xmm9 movlps %xmm4, 0 * SIZE(CO2) movhps %xmm4, 2 * SIZE(CO2) movlps %xmm9, 4 * SIZE(CO2) movhps %xmm9, 6 * SIZE(CO2) movsd 0 * SIZE(CO1, LDC, 2), %xmm0 movhps 2 * SIZE(CO1, LDC, 2), %xmm0 movsd 4 * SIZE(CO1, LDC, 2), %xmm1 movhps 6 * SIZE(CO1, LDC, 2), %xmm1 pshufd $0x50, %xmm10, %xmm4 pshufd $0xfa, %xmm10, %xmm10 mulps %xmm7, %xmm4 mulps %xmm7, %xmm10 addps %xmm0, %xmm4 addps %xmm1, %xmm10 movlps %xmm4, 0 * SIZE(CO1, LDC, 2) movhps %xmm4, 2 * SIZE(CO1, LDC, 2) movlps %xmm10, 4 * SIZE(CO1, LDC, 2) movhps %xmm10, 6 * SIZE(CO1, LDC, 2) movsd 0 * SIZE(CO2, LDC, 2), %xmm0 movhps 2 * SIZE(CO2, LDC, 2), %xmm0 movsd 4 * SIZE(CO2, LDC, 2), %xmm1 movhps 6 * SIZE(CO2, LDC, 2), %xmm1 pshufd $0x50, %xmm11, %xmm4 pshufd $0xfa, %xmm11, %xmm11 mulps %xmm7, %xmm4 mulps %xmm7, %xmm11 addps %xmm0, %xmm4 addps %xmm1, %xmm11 movlps %xmm4, 0 * SIZE(CO2, LDC, 2) movhps %xmm4, 2 * SIZE(CO2, LDC, 2) movlps %xmm11, 4 * SIZE(CO2, LDC, 2) movhps %xmm11, 6 * SIZE(CO2, LDC, 2) addq $8 * SIZE, CO1 addq $8 * SIZE, CO2 ALIGN_4 .L30: testq $2, M BRANCH jle .L40 #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movq B, BO #else movq B, BO movq KK, %rax leaq (, %rax, SIZE), %rax leaq (AO, %rax, 2), AO leaq (BO, %rax, 4), BO #endif movaps -32 * SIZE(AO), %xmm0 xorps %xmm3, %xmm3 xorps %xmm4, %xmm4 movaps -32 * SIZE(BO), %xmm2 xorps %xmm5, %xmm5 xorps %xmm6, %xmm6 movaps %xmm4, %xmm8 movaps %xmm4, %xmm9 movaps %xmm4, %xmm10 movaps %xmm4, %xmm11 #ifndef TRMMKERNEL movq K, %rax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movq K, %rax subq KK, %rax movq %rax, KKK #else movq KK, %rax #ifdef LEFT addq $2, %rax #else addq $4, %rax #endif movq %rax, KKK #endif sarq $2, %rax NOBRANCH jle .L35 ALIGN_3 .L32: PREFETCH (PREFETCHSIZE + 0) * SIZE(AO) pshufd $0x44, %xmm0, %xmm1 addps %xmm3, %xmm8 pshufd $0x50, %xmm2, %xmm3 mulps %xmm1, %xmm3 addps %xmm4, %xmm9 pshufd $0xfa, %xmm2, %xmm4 movaps -28 * SIZE(BO), %xmm2 mulps %xmm1, %xmm4 pshufd $0xee, %xmm0, %xmm1 movaps -28 * SIZE(AO), %xmm0 addps %xmm3, %xmm10 pshufd $0x50, %xmm2, %xmm3 mulps %xmm1, %xmm3 addps %xmm4, %xmm11 pshufd $0xfa, %xmm2, %xmm4 movaps -24 * SIZE(BO), %xmm2 mulps %xmm1, %xmm4 pshufd $0x44, %xmm0, %xmm1 addps %xmm3, %xmm8 pshufd $0x50, %xmm2, %xmm3 mulps %xmm1, %xmm3 addps %xmm4, %xmm9 pshufd $0xfa, %xmm2, %xmm4 movaps -20 * SIZE(BO), %xmm2 mulps %xmm1, %xmm4 pshufd $0xee, %xmm0, %xmm1 movaps -24 * SIZE(AO), %xmm0 addps %xmm3, %xmm10 pshufd $0x50, %xmm2, %xmm3 mulps %xmm1, %xmm3 addps %xmm4, %xmm11 pshufd $0xfa, %xmm2, %xmm4 movaps -16 * SIZE(BO), %xmm2 mulps %xmm1, %xmm4 subq $-8 * SIZE, AO subq $-16 * SIZE, BO subq $1, %rax BRANCH jg .L32 ALIGN_3 .L35: #ifndef TRMMKERNEL movq K, %rax #else movq KKK, %rax #endif andq $3, %rax BRANCH je .L38 ALIGN_3 .L36: pshufd $0x44, %xmm0, %xmm1 movsd -30 * SIZE(AO), %xmm0 addps %xmm3, %xmm8 pshufd $0x50, %xmm2, %xmm3 mulps %xmm1, %xmm3 addps %xmm4, %xmm9 pshufd $0xfa, %xmm2, %xmm4 movaps -28 * SIZE(BO), %xmm2 mulps %xmm1, %xmm4 addq $2 * SIZE, AO addq $4 * SIZE, BO subq $1, %rax BRANCH jg .L36 ALIGN_3 .L38: movups ALPHA_R, %xmm7 addps %xmm10, %xmm8 addps %xmm11, %xmm9 addps %xmm3, %xmm8 addps %xmm4, %xmm9 movsd 0 * SIZE(CO1), %xmm0 movhps 2 * SIZE(CO1), %xmm0 movsd 0 * SIZE(CO2), %xmm1 movhps 2 * SIZE(CO2), %xmm1 pshufd $0x50, %xmm8, %xmm4 pshufd $0xfa, %xmm8, %xmm8 mulps %xmm7, %xmm4 mulps %xmm7, %xmm8 addps %xmm0, %xmm4 addps %xmm1, %xmm8 movlps %xmm4, 0 * SIZE(CO1) movhps %xmm4, 2 * SIZE(CO1) movlps %xmm8, 0 * SIZE(CO2) movhps %xmm8, 2 * SIZE(CO2) movsd 0 * SIZE(CO1, LDC, 2), %xmm0 movhps 2 * SIZE(CO1, LDC, 2), %xmm0 movsd 0 * SIZE(CO2, LDC, 2), %xmm1 movhps 2 * SIZE(CO2, LDC, 2), %xmm1 pshufd $0x50, %xmm9, %xmm4 pshufd $0xfa, %xmm9, %xmm9 mulps %xmm7, %xmm4 mulps %xmm7, %xmm9 addps %xmm0, %xmm4 addps %xmm1, %xmm9 movlps %xmm4, 0 * SIZE(CO1, LDC, 2) movhps %xmm4, 2 * SIZE(CO1, LDC, 2) movlps %xmm9, 0 * SIZE(CO2, LDC, 2) movhps %xmm9, 2 * SIZE(CO2, LDC, 2) addq $4 * SIZE, CO1 addq $4 * SIZE, CO2 ALIGN_4 .L40: testq $1, M BRANCH jle .L49 #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movq B, BO #else movq B, BO movq KK, %rax leaq (, %rax, SIZE), %rax leaq (AO, %rax, 1), AO leaq (BO, %rax, 4), BO #endif movsd -32 * SIZE(AO), %xmm0 xorps %xmm8, %xmm8 movaps -32 * SIZE(BO), %xmm2 xorps %xmm9, %xmm9 #ifndef TRMMKERNEL movq K, %rax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movq K, %rax subq KK, %rax movq %rax, KKK #else movq KK, %rax #ifdef LEFT addq $1, %rax #else addq $4, %rax #endif movq %rax, KKK #endif sarq $2, %rax NOBRANCH jle .L45 ALIGN_3 .L42: PREFETCH (PREFETCHSIZE + 0) * SIZE(AO) pshufd $0x00, %xmm0, %xmm1 movss -31 * SIZE(AO), %xmm0 mulps %xmm1, %xmm2 addps %xmm2, %xmm8 movaps -28 * SIZE(BO), %xmm2 pshufd $0x00, %xmm0, %xmm1 movss -30 * SIZE(AO), %xmm0 mulps %xmm1, %xmm2 addps %xmm2, %xmm9 movaps -24 * SIZE(BO), %xmm2 pshufd $0x00, %xmm0, %xmm1 movss -29 * SIZE(AO), %xmm0 mulps %xmm1, %xmm2 addps %xmm2, %xmm8 movaps -20 * SIZE(BO), %xmm2 pshufd $0x00, %xmm0, %xmm1 movss -28 * SIZE(AO), %xmm0 mulps %xmm1, %xmm2 addps %xmm2, %xmm9 movaps -16 * SIZE(BO), %xmm2 subq $ -4 * SIZE, AO subq $-16 * SIZE, BO subq $1, %rax BRANCH jg .L42 ALIGN_3 .L45: #ifndef TRMMKERNEL movq K, %rax #else movq KKK, %rax #endif andq $3, %rax BRANCH je .L48 ALIGN_3 .L46: pshufd $0x00, %xmm0, %xmm1 movss -31 * SIZE(AO), %xmm0 mulps %xmm1, %xmm2 addps %xmm2, %xmm8 movaps -28 * SIZE(BO), %xmm2 addq $1 * SIZE, AO addq $4 * SIZE, BO subq $1, %rax BRANCH jg .L46 ALIGN_3 .L48: movups ALPHA_R, %xmm7 addps %xmm9, %xmm8 movsd 0 * SIZE(CO1), %xmm0 movhps 0 * SIZE(CO2), %xmm0 movsd 0 * SIZE(CO1, LDC, 2), %xmm1 movhps 0 * SIZE(CO2, LDC, 2), %xmm1 pshufd $0x50, %xmm8, %xmm4 pshufd $0xfa, %xmm8, %xmm8 mulps %xmm7, %xmm4 mulps %xmm7, %xmm8 addps %xmm0, %xmm4 addps %xmm1, %xmm8 movlps %xmm4, 0 * SIZE(CO1) movhps %xmm4, 0 * SIZE(CO2) movlps %xmm8, 0 * SIZE(CO1, LDC, 2) movhps %xmm8, 0 * SIZE(CO2, LDC, 2) ALIGN_4 .L49: #if defined(TRMMKERNEL) && !defined(LEFT) addq $4, KK #endif movq BO, B leaq (C, LDC, 4), C subq $1, J BRANCH jg .L10 ALIGN_4 .L50: testq $2, N jle .L90 #if defined(TRMMKERNEL) && defined(LEFT) movq OFFSET, %rax movq %rax, KK #endif movq C, CO1 leaq (C, LDC, 1), CO2 movq A, AO movq K, %rax salq $BASE_SHIFT + 1, %rax leaq (B, %rax), BB movq M, I sarq $3, I NOBRANCH jle .L60 ALIGN_4 .L51: #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movq B, BO #else movq B, BO movq KK, %rax leaq (, %rax, SIZE), %rax leaq (AO, %rax, 8), AO leaq (BO, %rax, 2), BO #endif prefetcht2 -32 * SIZE(BB) subq $-8 * SIZE, BB movaps -32 * SIZE(AO), %xmm0 xorps %xmm3, %xmm3 movaps -28 * SIZE(AO), %xmm1 xorps %xmm4, %xmm4 movaps -32 * SIZE(BO), %xmm2 xorps %xmm5, %xmm5 xorps %xmm6, %xmm6 prefetcht0 7 * SIZE(CO1) movaps %xmm4, %xmm8 movaps %xmm4, %xmm9 prefetcht0 7 * SIZE(CO2) movaps %xmm4, %xmm10 movaps %xmm4, %xmm11 #ifndef TRMMKERNEL movq K, %rax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movq K, %rax subq KK, %rax movq %rax, KKK #else movq KK, %rax #ifdef LEFT addq $8, %rax #else addq $2, %rax #endif movq %rax, KKK #endif sarq $2, %rax NOBRANCH jle .L55 ALIGN_3 .L52: PREFETCH (PREFETCHSIZE + 0) * SIZE(AO) addps %xmm3, %xmm8 pshufd $0x00, %xmm2, %xmm3 mulps %xmm0, %xmm3 addps %xmm4, %xmm9 pshufd $0x55, %xmm2, %xmm4 mulps %xmm0, %xmm4 movaps -24 * SIZE(AO), %xmm0 addps %xmm5, %xmm10 pshufd $0x00, %xmm2, %xmm5 mulps %xmm1, %xmm5 addps %xmm6, %xmm11 pshufd $0x55, %xmm2, %xmm6 mulps %xmm1, %xmm6 movaps -20 * SIZE(AO), %xmm1 addps %xmm3, %xmm8 pshufd $0xaa, %xmm2, %xmm3 mulps %xmm0, %xmm3 addps %xmm4, %xmm9 pshufd $0xff, %xmm2, %xmm4 mulps %xmm0, %xmm4 movaps -16 * SIZE(AO), %xmm0 PREFETCH (PREFETCHSIZE + 16) * SIZE(AO) addps %xmm5, %xmm10 pshufd $0xaa, %xmm2, %xmm5 mulps %xmm1, %xmm5 addps %xmm6, %xmm11 pshufd $0xff, %xmm2, %xmm6 movaps -28 * SIZE(BO), %xmm2 mulps %xmm1, %xmm6 movaps -12 * SIZE(AO), %xmm1 addps %xmm3, %xmm8 pshufd $0x00, %xmm2, %xmm3 mulps %xmm0, %xmm3 addps %xmm4, %xmm9 pshufd $0x55, %xmm2, %xmm4 mulps %xmm0, %xmm4 movaps -8 * SIZE(AO), %xmm0 addps %xmm5, %xmm10 pshufd $0x00, %xmm2, %xmm5 mulps %xmm1, %xmm5 addps %xmm6, %xmm11 pshufd $0x55, %xmm2, %xmm6 mulps %xmm1, %xmm6 movaps -4 * SIZE(AO), %xmm1 addps %xmm3, %xmm8 pshufd $0xaa, %xmm2, %xmm3 mulps %xmm0, %xmm3 addps %xmm4, %xmm9 pshufd $0xff, %xmm2, %xmm4 mulps %xmm0, %xmm4 movaps 0 * SIZE(AO), %xmm0 addps %xmm5, %xmm10 pshufd $0xaa, %xmm2, %xmm5 mulps %xmm1, %xmm5 addps %xmm6, %xmm11 pshufd $0xff, %xmm2, %xmm6 movaps -24 * SIZE(BO), %xmm2 mulps %xmm1, %xmm6 movaps 4 * SIZE(AO), %xmm1 subq $-32 * SIZE, AO subq $ -8 * SIZE, BO subq $1, %rax BRANCH jg .L52 ALIGN_3 .L55: #ifndef TRMMKERNEL movq K, %rax #else movq KKK, %rax #endif andq $3, %rax # if (k & 1) BRANCH je .L58 ALIGN_3 .L56: addps %xmm3, %xmm8 pshufd $0x00, %xmm2, %xmm3 mulps %xmm0, %xmm3 addps %xmm4, %xmm9 pshufd $0x55, %xmm2, %xmm4 mulps %xmm0, %xmm4 movaps -24 * SIZE(AO), %xmm0 addps %xmm5, %xmm10 pshufd $0x00, %xmm2, %xmm5 mulps %xmm1, %xmm5 addps %xmm6, %xmm11 pshufd $0x55, %xmm2, %xmm6 movsd -30 * SIZE(BO), %xmm2 mulps %xmm1, %xmm6 movaps -20 * SIZE(AO), %xmm1 addq $8 * SIZE, AO addq $2 * SIZE, BO subq $1, %rax BRANCH jg .L56 ALIGN_3 .L58: movups ALPHA_R, %xmm7 addps %xmm3, %xmm8 addps %xmm4, %xmm9 addps %xmm5, %xmm10 addps %xmm6, %xmm11 movsd 0 * SIZE(CO1), %xmm0 movhps 2 * SIZE(CO1), %xmm0 movsd 4 * SIZE(CO1), %xmm1 movhps 6 * SIZE(CO1), %xmm1 movsd 8 * SIZE(CO1), %xmm2 movhps 10 * SIZE(CO1), %xmm2 movsd 12 * SIZE(CO1), %xmm3 movhps 14 * SIZE(CO1), %xmm3 pshufd $0x50, %xmm8, %xmm4 pshufd $0xfa, %xmm8, %xmm8 pshufd $0x50, %xmm10, %xmm5 pshufd $0xfa, %xmm10, %xmm10 mulps %xmm7, %xmm4 mulps %xmm7, %xmm8 mulps %xmm7, %xmm5 mulps %xmm7, %xmm10 addps %xmm0, %xmm4 addps %xmm1, %xmm8 addps %xmm2, %xmm5 addps %xmm3, %xmm10 movlps %xmm4, 0 * SIZE(CO1) movhps %xmm4, 2 * SIZE(CO1) movlps %xmm8, 4 * SIZE(CO1) movhps %xmm8, 6 * SIZE(CO1) movlps %xmm5, 8 * SIZE(CO1) movhps %xmm5, 10 * SIZE(CO1) movlps %xmm10, 12 * SIZE(CO1) movhps %xmm10, 14 * SIZE(CO1) movsd 0 * SIZE(CO2), %xmm0 movhps 2 * SIZE(CO2), %xmm0 movsd 4 * SIZE(CO2), %xmm1 movhps 6 * SIZE(CO2), %xmm1 movsd 8 * SIZE(CO2), %xmm2 movhps 10 * SIZE(CO2), %xmm2 movsd 12 * SIZE(CO2), %xmm3 movhps 14 * SIZE(CO2), %xmm3 pshufd $0x50, %xmm9, %xmm4 pshufd $0xfa, %xmm9, %xmm9 pshufd $0x50, %xmm11, %xmm5 pshufd $0xfa, %xmm11, %xmm11 mulps %xmm7, %xmm4 mulps %xmm7, %xmm9 mulps %xmm7, %xmm5 mulps %xmm7, %xmm11 addps %xmm0, %xmm4 addps %xmm1, %xmm9 addps %xmm2, %xmm5 addps %xmm3, %xmm11 movlps %xmm4, 0 * SIZE(CO2) movhps %xmm4, 2 * SIZE(CO2) movlps %xmm9, 4 * SIZE(CO2) movhps %xmm9, 6 * SIZE(CO2) movlps %xmm5, 8 * SIZE(CO2) movhps %xmm5, 10 * SIZE(CO2) movlps %xmm11, 12 * SIZE(CO2) movhps %xmm11, 14 * SIZE(CO2) addq $16 * SIZE, CO1 addq $16 * SIZE, CO2 decq I BRANCH jg .L51 ALIGN_4 .L60: testq $4, M BRANCH jle .L70 #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movq B, BO #else movq B, BO movq KK, %rax leaq (, %rax, SIZE), %rax leaq (AO, %rax, 4), AO leaq (BO, %rax, 2), BO #endif movaps -32 * SIZE(AO), %xmm0 xorps %xmm3, %xmm3 xorps %xmm4, %xmm4 movaps -32 * SIZE(BO), %xmm2 xorps %xmm8, %xmm8 xorps %xmm9, %xmm9 xorps %xmm10, %xmm10 xorps %xmm11, %xmm11 #ifndef TRMMKERNEL movq K, %rax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movq K, %rax subq KK, %rax movq %rax, KKK #else movq KK, %rax #ifdef LEFT addq $4, %rax #else addq $2, %rax #endif movq %rax, KKK #endif sarq $2, %rax NOBRANCH jle .L65 ALIGN_3 .L62: PREFETCH (PREFETCHSIZE + 0) * SIZE(AO) addps %xmm3, %xmm8 pshufd $0x00, %xmm2, %xmm3 mulps %xmm0, %xmm3 addps %xmm4, %xmm9 pshufd $0x55, %xmm2, %xmm4 mulps %xmm0, %xmm4 movaps -28 * SIZE(AO), %xmm0 addps %xmm3, %xmm10 pshufd $0xaa, %xmm2, %xmm3 mulps %xmm0, %xmm3 addps %xmm4, %xmm11 pshufd $0xff, %xmm2, %xmm4 movaps -28 * SIZE(BO), %xmm2 mulps %xmm0, %xmm4 movaps -24 * SIZE(AO), %xmm0 addps %xmm3, %xmm8 pshufd $0x00, %xmm2, %xmm3 mulps %xmm0, %xmm3 addps %xmm4, %xmm9 pshufd $0x55, %xmm2, %xmm4 mulps %xmm0, %xmm4 movaps -20 * SIZE(AO), %xmm0 addps %xmm3, %xmm10 pshufd $0xaa, %xmm2, %xmm3 mulps %xmm0, %xmm3 addps %xmm4, %xmm11 pshufd $0xff, %xmm2, %xmm4 movaps -24 * SIZE(BO), %xmm2 mulps %xmm0, %xmm4 movaps -16 * SIZE(AO), %xmm0 subq $-16 * SIZE, AO subq $ -8 * SIZE, BO subq $1, %rax BRANCH jg .L62 ALIGN_3 .L65: #ifndef TRMMKERNEL movq K, %rax #else movq KKK, %rax #endif andq $3, %rax BRANCH je .L68 ALIGN_3 .L66: addps %xmm3, %xmm8 pshufd $0x00, %xmm2, %xmm3 mulps %xmm0, %xmm3 addps %xmm4, %xmm9 pshufd $0x55, %xmm2, %xmm4 movsd -30 * SIZE(BO), %xmm2 mulps %xmm0, %xmm4 movaps -28 * SIZE(AO), %xmm0 addq $4 * SIZE, AO addq $2 * SIZE, BO subq $1, %rax BRANCH jg .L66 ALIGN_3 .L68: movups ALPHA_R, %xmm7 addps %xmm10, %xmm8 addps %xmm11, %xmm9 addps %xmm3, %xmm8 addps %xmm4, %xmm9 movsd 0 * SIZE(CO1), %xmm0 movhps 2 * SIZE(CO1), %xmm0 movsd 4 * SIZE(CO1), %xmm1 movhps 6 * SIZE(CO1), %xmm1 pshufd $0x50, %xmm8, %xmm4 pshufd $0xfa, %xmm8, %xmm8 mulps %xmm7, %xmm4 mulps %xmm7, %xmm8 addps %xmm0, %xmm4 addps %xmm1, %xmm8 movlps %xmm4, 0 * SIZE(CO1) movhps %xmm4, 2 * SIZE(CO1) movlps %xmm8, 4 * SIZE(CO1) movhps %xmm8, 6 * SIZE(CO1) movsd 0 * SIZE(CO2), %xmm0 movhps 2 * SIZE(CO2), %xmm0 movsd 4 * SIZE(CO2), %xmm1 movhps 6 * SIZE(CO2), %xmm1 pshufd $0x50, %xmm9, %xmm4 pshufd $0xfa, %xmm9, %xmm9 mulps %xmm7, %xmm4 mulps %xmm7, %xmm9 addps %xmm0, %xmm4 addps %xmm1, %xmm9 movlps %xmm4, 0 * SIZE(CO2) movhps %xmm4, 2 * SIZE(CO2) movlps %xmm9, 4 * SIZE(CO2) movhps %xmm9, 6 * SIZE(CO2) addq $8 * SIZE, CO1 addq $8 * SIZE, CO2 ALIGN_4 .L70: testq $2, M BRANCH jle .L80 #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movq B, BO #else movq B, BO movq KK, %rax leaq (, %rax, SIZE), %rax leaq (AO, %rax, 2), AO leaq (BO, %rax, 2), BO #endif movaps -32 * SIZE(AO), %xmm0 xorps %xmm3, %xmm3 movaps -32 * SIZE(BO), %xmm2 xorps %xmm8, %xmm8 xorps %xmm9, %xmm9 #ifndef TRMMKERNEL movq K, %rax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movq K, %rax subq KK, %rax movq %rax, KKK #else movq KK, %rax #ifdef LEFT addq $2, %rax #else addq $2, %rax #endif movq %rax, KKK #endif sarq $2, %rax NOBRANCH jle .L75 ALIGN_3 .L72: PREFETCH (PREFETCHSIZE + 0) * SIZE(AO) addps %xmm3, %xmm8 pshufd $0x44, %xmm0, %xmm1 pshufd $0x50, %xmm2, %xmm3 mulps %xmm1, %xmm3 addps %xmm3, %xmm9 pshufd $0xee, %xmm0, %xmm1 movaps -28 * SIZE(AO), %xmm0 pshufd $0xfa, %xmm2, %xmm3 movaps -28 * SIZE(BO), %xmm2 mulps %xmm1, %xmm3 addps %xmm3, %xmm8 pshufd $0x44, %xmm0, %xmm1 pshufd $0x50, %xmm2, %xmm3 mulps %xmm1, %xmm3 addps %xmm3, %xmm9 pshufd $0xee, %xmm0, %xmm1 movaps -24 * SIZE(AO), %xmm0 pshufd $0xfa, %xmm2, %xmm3 movaps -24 * SIZE(BO), %xmm2 mulps %xmm1, %xmm3 subq $-8 * SIZE, AO subq $-8 * SIZE, BO subq $1, %rax BRANCH jg .L72 ALIGN_3 .L75: #ifndef TRMMKERNEL movq K, %rax #else movq KKK, %rax #endif andq $3, %rax BRANCH je .L78 ALIGN_3 .L76: addps %xmm3, %xmm8 pshufd $0x44, %xmm0, %xmm1 movsd -30 * SIZE(AO), %xmm0 pshufd $0x50, %xmm2, %xmm3 movsd -30 * SIZE(BO), %xmm2 mulps %xmm1, %xmm3 addq $2 * SIZE, AO addq $2 * SIZE, BO subq $1, %rax BRANCH jg .L76 ALIGN_3 .L78: movups ALPHA_R, %xmm7 addps %xmm9, %xmm8 addps %xmm3, %xmm8 movsd 0 * SIZE(CO1), %xmm0 movhps 2 * SIZE(CO1), %xmm0 movsd 0 * SIZE(CO2), %xmm1 movhps 2 * SIZE(CO2), %xmm1 pshufd $0x50, %xmm8, %xmm4 pshufd $0xfa, %xmm8, %xmm8 mulps %xmm7, %xmm4 mulps %xmm7, %xmm8 addps %xmm0, %xmm4 addps %xmm1, %xmm8 movlps %xmm4, 0 * SIZE(CO1) movhps %xmm4, 2 * SIZE(CO1) movlps %xmm8, 0 * SIZE(CO2) movhps %xmm8, 2 * SIZE(CO2) addq $4 * SIZE, CO1 addq $4 * SIZE, CO2 ALIGN_4 .L80: testq $1, M BRANCH jle .L89 #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movq B, BO #else movq B, BO movq KK, %rax leaq (, %rax, SIZE), %rax leaq (AO, %rax, 1), AO leaq (BO, %rax, 2), BO #endif movsd -32 * SIZE(AO), %xmm0 xorps %xmm8, %xmm8 movsd -32 * SIZE(BO), %xmm2 xorps %xmm9, %xmm9 #ifndef TRMMKERNEL movq K, %rax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movq K, %rax subq KK, %rax movq %rax, KKK #else movq KK, %rax #ifdef LEFT addq $1, %rax #else addq $2, %rax #endif movq %rax, KKK #endif sarq $2, %rax NOBRANCH jle .L85 ALIGN_3 .L82: PREFETCH (PREFETCHSIZE + 0) * SIZE(AO) pshufd $0x00, %xmm0, %xmm1 movss -31 * SIZE(AO), %xmm0 mulps %xmm1, %xmm2 addps %xmm2, %xmm8 movsd -30 * SIZE(BO), %xmm2 pshufd $0x00, %xmm0, %xmm1 movss -30 * SIZE(AO), %xmm0 mulps %xmm1, %xmm2 addps %xmm2, %xmm9 movsd -28 * SIZE(BO), %xmm2 pshufd $0x00, %xmm0, %xmm1 movss -29 * SIZE(AO), %xmm0 mulps %xmm1, %xmm2 addps %xmm2, %xmm8 movsd -26 * SIZE(BO), %xmm2 pshufd $0x00, %xmm0, %xmm1 movss -28 * SIZE(AO), %xmm0 mulps %xmm1, %xmm2 addps %xmm2, %xmm9 movsd -24 * SIZE(BO), %xmm2 subq $-4 * SIZE, AO subq $-8 * SIZE, BO subq $1, %rax BRANCH jg .L82 ALIGN_3 .L85: #ifndef TRMMKERNEL movq K, %rax #else movq KKK, %rax #endif andq $3, %rax BRANCH je .L88 ALIGN_3 .L86: pshufd $0x00, %xmm0, %xmm1 movss -31 * SIZE(AO), %xmm0 mulps %xmm1, %xmm2 addps %xmm2, %xmm8 movsd -30 * SIZE(BO), %xmm2 addq $1 * SIZE, AO addq $2 * SIZE, BO subq $1, %rax BRANCH jg .L86 ALIGN_3 .L88: movups ALPHA_R, %xmm7 addps %xmm9, %xmm8 movsd 0 * SIZE(CO1), %xmm0 movhps 0 * SIZE(CO2), %xmm0 pshufd $0x50, %xmm8, %xmm4 mulps %xmm7, %xmm4 addps %xmm0, %xmm4 movlps %xmm4, 0 * SIZE(CO1) movhps %xmm4, 0 * SIZE(CO2) ALIGN_4 .L89: #if defined(TRMMKERNEL) && !defined(LEFT) addq $2, KK #endif movq BO, B leaq (C, LDC, 2), C ALIGN_4 .L90: testq $1, N jle .L999 #if defined(TRMMKERNEL) && defined(LEFT) movq OFFSET, %rax movq %rax, KK #endif movq C, CO1 movq A, AO movq M, I sarq $3, I NOBRANCH jle .L100 ALIGN_4 .L91: #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movq B, BO #else movq B, BO movq KK, %rax leaq (, %rax, SIZE), %rax leaq (AO, %rax, 8), AO leaq (BO, %rax, 1), BO #endif movaps -32 * SIZE(AO), %xmm0 xorps %xmm8, %xmm8 movaps -28 * SIZE(AO), %xmm1 xorps %xmm9, %xmm9 movsd -32 * SIZE(BO), %xmm2 xorps %xmm10, %xmm10 prefetcht0 7 * SIZE(CO1) xorps %xmm11, %xmm11 #ifndef TRMMKERNEL movq K, %rax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movq K, %rax subq KK, %rax movq %rax, KKK #else movq KK, %rax #ifdef LEFT addq $8, %rax #else addq $1, %rax #endif movq %rax, KKK #endif sarq $2, %rax NOBRANCH jle .L95 ALIGN_3 .L92: PREFETCH (PREFETCHSIZE + 0) * SIZE(AO) pshufd $0x00, %xmm2, %xmm3 mulps %xmm3, %xmm0 addps %xmm0, %xmm8 movaps -24 * SIZE(AO), %xmm0 mulps %xmm3, %xmm1 addps %xmm1, %xmm9 movaps -20 * SIZE(AO), %xmm1 pshufd $0x55, %xmm2, %xmm3 movsd -30 * SIZE(BO), %xmm2 mulps %xmm3, %xmm0 addps %xmm0, %xmm10 movaps -16 * SIZE(AO), %xmm0 mulps %xmm3, %xmm1 addps %xmm1, %xmm11 movaps -12 * SIZE(AO), %xmm1 PREFETCH (PREFETCHSIZE + 16) * SIZE(AO) pshufd $0x00, %xmm2, %xmm3 mulps %xmm3, %xmm0 addps %xmm0, %xmm8 movaps -8 * SIZE(AO), %xmm0 mulps %xmm3, %xmm1 addps %xmm1, %xmm9 movaps -4 * SIZE(AO), %xmm1 pshufd $0x55, %xmm2, %xmm3 movsd -28 * SIZE(BO), %xmm2 mulps %xmm3, %xmm0 addps %xmm0, %xmm10 movaps 0 * SIZE(AO), %xmm0 mulps %xmm3, %xmm1 addps %xmm1, %xmm11 movaps 4 * SIZE(AO), %xmm1 subq $-32 * SIZE, AO subq $ -4 * SIZE, BO subq $1, %rax BRANCH jg .L92 ALIGN_3 .L95: #ifndef TRMMKERNEL movq K, %rax #else movq KKK, %rax #endif andq $3, %rax # if (k & 1) BRANCH je .L98 ALIGN_3 .L96: pshufd $0x00, %xmm2, %xmm3 movss -31 * SIZE(BO), %xmm2 mulps %xmm3, %xmm0 addps %xmm0, %xmm8 movaps -24 * SIZE(AO), %xmm0 mulps %xmm3, %xmm1 addps %xmm1, %xmm9 movaps -20 * SIZE(AO), %xmm1 addq $8 * SIZE, AO addq $1 * SIZE, BO subq $1, %rax BRANCH jg .L96 ALIGN_3 .L98: movups ALPHA_R, %xmm7 addps %xmm10, %xmm8 addps %xmm11, %xmm9 movsd 0 * SIZE(CO1), %xmm0 movhps 2 * SIZE(CO1), %xmm0 movsd 4 * SIZE(CO1), %xmm1 movhps 6 * SIZE(CO1), %xmm1 movsd 8 * SIZE(CO1), %xmm2 movhps 10 * SIZE(CO1), %xmm2 movsd 12 * SIZE(CO1), %xmm3 movhps 14 * SIZE(CO1), %xmm3 pshufd $0x50, %xmm8, %xmm4 pshufd $0xfa, %xmm8, %xmm8 pshufd $0x50, %xmm9, %xmm5 pshufd $0xfa, %xmm9, %xmm9 mulps %xmm7, %xmm4 mulps %xmm7, %xmm8 mulps %xmm7, %xmm5 mulps %xmm7, %xmm9 addps %xmm0, %xmm4 addps %xmm1, %xmm8 addps %xmm2, %xmm5 addps %xmm3, %xmm9 movlps %xmm4, 0 * SIZE(CO1) movhps %xmm4, 2 * SIZE(CO1) movlps %xmm8, 4 * SIZE(CO1) movhps %xmm8, 6 * SIZE(CO1) movlps %xmm5, 8 * SIZE(CO1) movhps %xmm5, 10 * SIZE(CO1) movlps %xmm9, 12 * SIZE(CO1) movhps %xmm9, 14 * SIZE(CO1) addq $16 * SIZE, CO1 decq I BRANCH jg .L91 ALIGN_4 .L100: testq $4, M BRANCH jle .L110 #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movq B, BO #else movq B, BO movq KK, %rax leaq (, %rax, SIZE), %rax leaq (AO, %rax, 4), AO leaq (BO, %rax, 1), BO #endif movaps -32 * SIZE(AO), %xmm0 xorps %xmm8, %xmm8 movsd -32 * SIZE(BO), %xmm2 xorps %xmm9, %xmm9 #ifndef TRMMKERNEL movq K, %rax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movq K, %rax subq KK, %rax movq %rax, KKK #else movq KK, %rax #ifdef LEFT addq $4, %rax #else addq $1, %rax #endif movq %rax, KKK #endif sarq $2, %rax NOBRANCH jle .L105 ALIGN_3 .L102: PREFETCH (PREFETCHSIZE + 0) * SIZE(AO) pshufd $0x00, %xmm2, %xmm3 mulps %xmm0, %xmm3 movaps -28 * SIZE(AO), %xmm0 addps %xmm3, %xmm8 pshufd $0x55, %xmm2, %xmm3 movsd -30 * SIZE(BO), %xmm2 mulps %xmm0, %xmm3 movaps -24 * SIZE(AO), %xmm0 addps %xmm3, %xmm9 pshufd $0x00, %xmm2, %xmm3 mulps %xmm0, %xmm3 movaps -20 * SIZE(AO), %xmm0 addps %xmm3, %xmm8 pshufd $0x55, %xmm2, %xmm3 movsd -28 * SIZE(BO), %xmm2 mulps %xmm0, %xmm3 movaps -16 * SIZE(AO), %xmm0 addps %xmm3, %xmm9 subq $-16 * SIZE, AO subq $ -4 * SIZE, BO subq $1, %rax BRANCH jg .L102 ALIGN_3 .L105: #ifndef TRMMKERNEL movq K, %rax #else movq KKK, %rax #endif andq $3, %rax BRANCH je .L108 ALIGN_3 .L106: pshufd $0x00, %xmm2, %xmm3 movss -31 * SIZE(BO), %xmm2 mulps %xmm0, %xmm3 movaps -28 * SIZE(AO), %xmm0 addps %xmm3, %xmm8 addq $4 * SIZE, AO addq $1 * SIZE, BO subq $1, %rax BRANCH jg .L106 ALIGN_3 .L108: movups ALPHA_R, %xmm7 addps %xmm9, %xmm8 movsd 0 * SIZE(CO1), %xmm0 movhps 2 * SIZE(CO1), %xmm0 movsd 4 * SIZE(CO1), %xmm1 movhps 6 * SIZE(CO1), %xmm1 pshufd $0x50, %xmm8, %xmm4 pshufd $0xfa, %xmm8, %xmm8 mulps %xmm7, %xmm4 mulps %xmm7, %xmm8 addps %xmm0, %xmm4 addps %xmm1, %xmm8 movlps %xmm4, 0 * SIZE(CO1) movhps %xmm4, 2 * SIZE(CO1) movlps %xmm8, 4 * SIZE(CO1) movhps %xmm8, 6 * SIZE(CO1) addq $8 * SIZE, CO1 ALIGN_4 .L110: testq $2, M BRANCH jle .L120 #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movq B, BO #else movq B, BO movq KK, %rax leaq (, %rax, SIZE), %rax leaq (AO, %rax, 2), AO leaq (BO, %rax, 1), BO #endif movaps -32 * SIZE(AO), %xmm0 xorps %xmm3, %xmm3 movsd -32 * SIZE(BO), %xmm2 xorps %xmm8, %xmm8 xorps %xmm9, %xmm9 #ifndef TRMMKERNEL movq K, %rax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movq K, %rax subq KK, %rax movq %rax, KKK #else movq KK, %rax #ifdef LEFT addq $2, %rax #else addq $1, %rax #endif movq %rax, KKK #endif sarq $2, %rax NOBRANCH jle .L115 ALIGN_3 .L112: PREFETCH (PREFETCHSIZE + 0) * SIZE(AO) pshufd $0x00, %xmm2, %xmm3 mulps %xmm0, %xmm3 movsd -30 * SIZE(AO), %xmm0 addps %xmm3, %xmm8 pshufd $0x55, %xmm2, %xmm3 movsd -30 * SIZE(BO), %xmm2 mulps %xmm0, %xmm3 movsd -28 * SIZE(AO), %xmm0 addps %xmm3, %xmm8 pshufd $0x00, %xmm2, %xmm3 mulps %xmm0, %xmm3 movsd -26 * SIZE(AO), %xmm0 addps %xmm3, %xmm8 pshufd $0x55, %xmm2, %xmm3 movsd -28 * SIZE(BO), %xmm2 mulps %xmm0, %xmm3 movsd -24 * SIZE(AO), %xmm0 addps %xmm3, %xmm8 subq $-8 * SIZE, AO subq $-4 * SIZE, BO subq $1, %rax BRANCH jg .L112 ALIGN_3 .L115: #ifndef TRMMKERNEL movq K, %rax #else movq KKK, %rax #endif andq $3, %rax BRANCH je .L118 ALIGN_3 .L116: pshufd $0x00, %xmm2, %xmm3 movss -31 * SIZE(BO), %xmm2 mulps %xmm0, %xmm3 movsd -30 * SIZE(AO), %xmm0 addps %xmm3, %xmm8 addq $2 * SIZE, AO addq $1 * SIZE, BO subq $1, %rax BRANCH jg .L116 ALIGN_3 .L118: movups ALPHA_R, %xmm7 movsd 0 * SIZE(CO1), %xmm0 movhps 2 * SIZE(CO1), %xmm0 pshufd $0x50, %xmm8, %xmm4 mulps %xmm7, %xmm4 addps %xmm0, %xmm4 movlps %xmm4, 0 * SIZE(CO1) movhps %xmm4, 2 * SIZE(CO1) addq $4 * SIZE, CO1 ALIGN_4 .L120: testq $1, M BRANCH jle .L999 #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movq B, BO #else movq B, BO movq KK, %rax leaq (, %rax, SIZE), %rax leaq (AO, %rax, 1), AO leaq (BO, %rax, 1), BO #endif movss -32 * SIZE(AO), %xmm0 xorps %xmm8, %xmm8 movss -32 * SIZE(BO), %xmm2 xorps %xmm9, %xmm9 #ifndef TRMMKERNEL movq K, %rax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movq K, %rax subq KK, %rax movq %rax, KKK #else movq KK, %rax #ifdef LEFT addq $1, %rax #else addq $1, %rax #endif movq %rax, KKK #endif sarq $2, %rax NOBRANCH jle .L125 ALIGN_3 .L122: mulss %xmm0, %xmm2 movss -31 * SIZE(AO), %xmm0 addss %xmm2, %xmm8 movss -31 * SIZE(BO), %xmm2 mulss %xmm0, %xmm2 movss -30 * SIZE(AO), %xmm0 addss %xmm2, %xmm9 movss -30 * SIZE(BO), %xmm2 mulss %xmm0, %xmm2 movss -29 * SIZE(AO), %xmm0 addss %xmm2, %xmm8 movss -29 * SIZE(BO), %xmm2 mulss %xmm0, %xmm2 movss -28 * SIZE(AO), %xmm0 addss %xmm2, %xmm9 movss -28 * SIZE(BO), %xmm2 subq $-4 * SIZE, AO subq $-4 * SIZE, BO subq $1, %rax BRANCH jg .L122 ALIGN_3 .L125: #ifndef TRMMKERNEL movq K, %rax #else movq KKK, %rax #endif andq $3, %rax BRANCH je .L128 ALIGN_3 .L126: mulss %xmm0, %xmm2 movss -31 * SIZE(AO), %xmm0 addss %xmm2, %xmm8 movss -31 * SIZE(BO), %xmm2 addq $1 * SIZE, AO addq $1 * SIZE, BO subq $1, %rax BRANCH jg .L126 ALIGN_3 .L128: movups ALPHA_R, %xmm7 addss %xmm9, %xmm8 movsd 0 * SIZE(CO1), %xmm0 pshufd $0x50, %xmm8, %xmm4 mulps %xmm7, %xmm4 addps %xmm0, %xmm4 movlps %xmm4, 0 * SIZE(CO1) ALIGN_4 .L999: movq 0(%rsp), %rbx movq 8(%rsp), %rbp movq 16(%rsp), %r12 movq 24(%rsp), %r13 movq 32(%rsp), %r14 movq 40(%rsp), %r15 #ifdef WINDOWS_ABI movq 48(%rsp), %rdi movq 56(%rsp), %rsi movups 64(%rsp), %xmm6 movups 80(%rsp), %xmm7 movups 96(%rsp), %xmm8 movups 112(%rsp), %xmm9 movups 128(%rsp), %xmm10 movups 144(%rsp), %xmm11 movups 160(%rsp), %xmm12 movups 176(%rsp), %xmm13 movups 192(%rsp), %xmm14 movups 208(%rsp), %xmm15 #endif addq $STACKSIZE, %rsp ret EPILOGUE