/*********************************************************************/ /* 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 STACK 16 #define ARGS 16 #define M 4 + STACK + ARGS(%esp) #define N 8 + STACK + ARGS(%esp) #define K 12 + STACK + ARGS(%esp) #define ALPHA 16 + STACK + ARGS(%esp) #define A 32 + STACK + ARGS(%esp) #define ARG_B 36 + STACK + ARGS(%esp) #define C 40 + STACK + ARGS(%esp) #define ARG_LDC 44 + STACK + ARGS(%esp) #define J 0 + STACK(%esp) #define BX 4 + STACK(%esp) #define KK 8 + STACK(%esp) #define KKK 12 + STACK(%esp) #define AA %edx #define BB %ecx #define LDC %ebp #define B %edi #define C1 %esi #define I %ebx #ifdef NANO #define PREFETCHSIZE (8 * 3 + 4) #define PREFETCHW prefetcht0 #define PREFETCHB prefetcht0 #endif #ifndef PREFETCH #define PREFETCH prefetcht0 #endif #ifndef PREFETCHW #define PREFETCHW prefetcht2 #endif #ifndef PREFETCHB #define PREFETCHB prefetcht2 #endif #ifndef PREFETCHSIZE #define PREFETCHSIZE (8 * 21 + 4) #endif PROLOGUE subl $ARGS, %esp # Generate Stack Frame pushl %ebp pushl %edi pushl %esi pushl %ebx PROFCODE movl ARG_B, B movl ARG_LDC, LDC #ifdef TRMMKERNEL movl OFFSET, %eax #ifndef LEFT negl %eax #endif movl %eax, KK #endif subl $-16 * SIZE, A subl $-16 * SIZE, B sall $ZBASE_SHIFT, LDC movl N, %eax sarl $2, %eax movl %eax, J jle .L30 ALIGN_4 .L01: #if defined(TRMMKERNEL) && defined(LEFT) movl OFFSET, %eax movl %eax, KK #endif movl K, %eax sall $BASE_SHIFT + 2, %eax leal (B, %eax), %eax movl %eax, BX movl C, C1 movl A, AA movl M, I sarl $1, I jle .L20 ALIGN_4 .L11: #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movl B, BB #else movl B, BB movl KK, %eax leal (, %eax, SIZE), %eax leal (AA, %eax, 2), AA leal (BB, %eax, 4), BB #endif movl BX, %eax PREFETCHB -16 * SIZE(%eax) subl $-8 * SIZE, BX leal (C1, LDC, 2), %eax movaps -16 * SIZE(AA), %xmm0 pxor %xmm2, %xmm2 movaps -16 * SIZE(BB), %xmm1 pxor %xmm3, %xmm3 pxor %xmm4, %xmm4 PREFETCHW 1 * SIZE(C1) pxor %xmm5, %xmm5 PREFETCHW 1 * SIZE(C1, LDC) pxor %xmm6, %xmm6 PREFETCHW 1 * SIZE(%eax) pxor %xmm7, %xmm7 PREFETCHW 1 * SIZE(%eax, LDC) #ifndef TRMMKERNEL movl K, %eax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movl K, %eax subl KK, %eax movl %eax, KKK #else movl KK, %eax #ifdef LEFT addl $2, %eax #else addl $4, %eax #endif movl %eax, KKK #endif sarl $3, %eax je .L15 ALIGN_4 .L12: PREFETCH (PREFETCHSIZE + 0) * SIZE(AA) addpd %xmm3, %xmm7 movaps -14 * SIZE(BB), %xmm3 addpd %xmm2, %xmm6 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 addpd %xmm1, %xmm5 movaps -12 * SIZE(BB), %xmm1 addpd %xmm2, %xmm4 pshufd $0x4e, %xmm3, %xmm2 mulpd %xmm0, %xmm3 mulpd %xmm0, %xmm2 movaps -14 * SIZE(AA), %xmm0 addpd %xmm3, %xmm7 movaps -10 * SIZE(BB), %xmm3 addpd %xmm2, %xmm6 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 addpd %xmm1, %xmm5 movaps -8 * SIZE(BB), %xmm1 addpd %xmm2, %xmm4 pshufd $0x4e, %xmm3, %xmm2 mulpd %xmm0, %xmm3 mulpd %xmm0, %xmm2 movaps -12 * SIZE(AA), %xmm0 addpd %xmm3, %xmm7 movaps -6 * SIZE(BB), %xmm3 addpd %xmm2, %xmm6 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 addpd %xmm1, %xmm5 movaps -4 * SIZE(BB), %xmm1 addpd %xmm2, %xmm4 pshufd $0x4e, %xmm3, %xmm2 mulpd %xmm0, %xmm3 mulpd %xmm0, %xmm2 movaps -10 * SIZE(AA), %xmm0 addpd %xmm3, %xmm7 movaps -2 * SIZE(BB), %xmm3 addpd %xmm2, %xmm6 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 addpd %xmm1, %xmm5 movaps 0 * SIZE(BB), %xmm1 addpd %xmm2, %xmm4 pshufd $0x4e, %xmm3, %xmm2 mulpd %xmm0, %xmm3 mulpd %xmm0, %xmm2 movaps -8 * SIZE(AA), %xmm0 PREFETCH (PREFETCHSIZE + 8) * SIZE(AA) addpd %xmm3, %xmm7 movaps 2 * SIZE(BB), %xmm3 addpd %xmm2, %xmm6 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 addpd %xmm1, %xmm5 movaps 4 * SIZE(BB), %xmm1 addpd %xmm2, %xmm4 pshufd $0x4e, %xmm3, %xmm2 mulpd %xmm0, %xmm3 mulpd %xmm0, %xmm2 movaps -6 * SIZE(AA), %xmm0 addpd %xmm3, %xmm7 movaps 6 * SIZE(BB), %xmm3 addpd %xmm2, %xmm6 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 addpd %xmm1, %xmm5 movaps 8 * SIZE(BB), %xmm1 addpd %xmm2, %xmm4 pshufd $0x4e, %xmm3, %xmm2 mulpd %xmm0, %xmm3 mulpd %xmm0, %xmm2 movaps -4 * SIZE(AA), %xmm0 addpd %xmm3, %xmm7 movaps 10 * SIZE(BB), %xmm3 addpd %xmm2, %xmm6 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 addpd %xmm1, %xmm5 movaps 12 * SIZE(BB), %xmm1 addpd %xmm2, %xmm4 pshufd $0x4e, %xmm3, %xmm2 mulpd %xmm0, %xmm3 mulpd %xmm0, %xmm2 movaps -2 * SIZE(AA), %xmm0 addpd %xmm3, %xmm7 movaps 14 * SIZE(BB), %xmm3 addpd %xmm2, %xmm6 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 addpd %xmm1, %xmm5 movaps 16 * SIZE(BB), %xmm1 addpd %xmm2, %xmm4 pshufd $0x4e, %xmm3, %xmm2 mulpd %xmm0, %xmm3 subl $-32 * SIZE, BB mulpd %xmm0, %xmm2 movaps 0 * SIZE(AA), %xmm0 subl $-16 * SIZE, AA subl $1, %eax BRANCH jne .L12 ALIGN_4 .L15: #ifndef TRMMKERNEL movl K, %eax #else movl KKK, %eax #endif andl $7, %eax BRANCH je .L18 ALIGN_4 .L16: addpd %xmm3, %xmm7 movaps -14 * SIZE(BB), %xmm3 addpd %xmm2, %xmm6 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 addpd %xmm1, %xmm5 movaps -12 * SIZE(BB), %xmm1 addpd %xmm2, %xmm4 pshufd $0x4e, %xmm3, %xmm2 mulpd %xmm0, %xmm3 mulpd %xmm0, %xmm2 movaps -14 * SIZE(AA), %xmm0 addl $2 * SIZE, AA addl $4 * SIZE, BB decl %eax jg .L16 ALIGN_4 .L18: addpd %xmm2, %xmm6 addpd %xmm3, %xmm7 movups ALPHA, %xmm3 movaps %xmm4, %xmm0 movsd %xmm5, %xmm4 movsd %xmm0, %xmm5 movaps %xmm6, %xmm0 movsd %xmm7, %xmm6 movsd %xmm0, %xmm7 leal (C1, LDC, 2), %eax movsd 0 * SIZE(C1), %xmm0 movhps 1 * SIZE(C1), %xmm0 movsd 2 * SIZE(C1), %xmm1 movhps 3 * SIZE(C1), %xmm1 pshufd $0x44, %xmm4, %xmm2 unpckhpd %xmm4, %xmm4 mulpd %xmm3, %xmm2 addpd %xmm2, %xmm0 mulpd %xmm3, %xmm4 addpd %xmm4, %xmm1 movlps %xmm0, 0 * SIZE(C1) movhps %xmm0, 1 * SIZE(C1) movlps %xmm1, 2 * SIZE(C1) movhps %xmm1, 3 * SIZE(C1) movsd 0 * SIZE(C1, LDC), %xmm0 movhps 1 * SIZE(C1, LDC), %xmm0 movsd 2 * SIZE(C1, LDC), %xmm1 movhps 3 * SIZE(C1, LDC), %xmm1 pshufd $0x44, %xmm5, %xmm2 unpckhpd %xmm5, %xmm5 mulpd %xmm3, %xmm2 addpd %xmm2, %xmm0 mulpd %xmm3, %xmm5 addpd %xmm5, %xmm1 movlps %xmm0, 0 * SIZE(C1, LDC) movhps %xmm0, 1 * SIZE(C1, LDC) movlps %xmm1, 2 * SIZE(C1, LDC) movhps %xmm1, 3 * SIZE(C1, LDC) movsd 0 * SIZE(%eax), %xmm0 movhps 1 * SIZE(%eax), %xmm0 movsd 2 * SIZE(%eax), %xmm1 movhps 3 * SIZE(%eax), %xmm1 pshufd $0x44, %xmm6, %xmm2 unpckhpd %xmm6, %xmm6 mulpd %xmm3, %xmm2 addpd %xmm2, %xmm0 mulpd %xmm3, %xmm6 addpd %xmm6, %xmm1 movlps %xmm0, 0 * SIZE(%eax) movhps %xmm0, 1 * SIZE(%eax) movlps %xmm1, 2 * SIZE(%eax) movhps %xmm1, 3 * SIZE(%eax) movsd 0 * SIZE(%eax, LDC), %xmm0 movhps 1 * SIZE(%eax, LDC), %xmm0 movsd 2 * SIZE(%eax, LDC), %xmm1 movhps 3 * SIZE(%eax, LDC), %xmm1 pshufd $0x44, %xmm7, %xmm2 unpckhpd %xmm7, %xmm7 mulpd %xmm3, %xmm2 addpd %xmm2, %xmm0 mulpd %xmm3, %xmm7 addpd %xmm7, %xmm1 movlps %xmm0, 0 * SIZE(%eax, LDC) movhps %xmm0, 1 * SIZE(%eax, LDC) movlps %xmm1, 2 * SIZE(%eax, LDC) movhps %xmm1, 3 * SIZE(%eax, LDC) addl $4 * SIZE, C1 decl I jg .L11 ALIGN_4 .L20: movl M, I testl $1, I jle .L29 #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movl B, BB #else movl B, BB movl KK, %eax leal (, %eax, SIZE), %eax addl %eax, AA leal (BB, %eax, 4), BB #endif movaps -16 * SIZE(AA), %xmm0 pxor %xmm4, %xmm4 movaps -16 * SIZE(BB), %xmm2 pxor %xmm5, %xmm5 movaps -14 * SIZE(BB), %xmm3 pxor %xmm6, %xmm6 pxor %xmm7, %xmm7 #ifndef TRMMKERNEL movl K, %eax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movl K, %eax subl KK, %eax movl %eax, KKK #else movl KK, %eax #ifdef LEFT addl $1, %eax #else addl $4, %eax #endif movl %eax, KKK #endif sarl $3, %eax je .L25 ALIGN_4 .L22: PREFETCH (PREFETCHSIZE + 0) * SIZE(AA) pshufd $0x44, %xmm0, %xmm1 mulpd %xmm1, %xmm2 mulpd %xmm1, %xmm3 addpd %xmm2, %xmm4 movaps -12 * SIZE(BB), %xmm2 addpd %xmm3, %xmm5 movaps -10 * SIZE(BB), %xmm3 pshufd $0xee, %xmm0, %xmm1 movaps -14 * SIZE(AA), %xmm0 mulpd %xmm1, %xmm2 mulpd %xmm1, %xmm3 addpd %xmm2, %xmm6 movaps -8 * SIZE(BB), %xmm2 addpd %xmm3, %xmm7 movaps -6 * SIZE(BB), %xmm3 pshufd $0x44, %xmm0, %xmm1 mulpd %xmm1, %xmm2 mulpd %xmm1, %xmm3 addpd %xmm2, %xmm4 movaps -4 * SIZE(BB), %xmm2 addpd %xmm3, %xmm5 movaps -2 * SIZE(BB), %xmm3 pshufd $0xee, %xmm0, %xmm1 movaps -12 * SIZE(AA), %xmm0 mulpd %xmm1, %xmm2 mulpd %xmm1, %xmm3 addpd %xmm2, %xmm6 movaps 0 * SIZE(BB), %xmm2 addpd %xmm3, %xmm7 movaps 2 * SIZE(BB), %xmm3 pshufd $0x44, %xmm0, %xmm1 mulpd %xmm1, %xmm2 mulpd %xmm1, %xmm3 addpd %xmm2, %xmm4 movaps 4 * SIZE(BB), %xmm2 addpd %xmm3, %xmm5 movaps 6 * SIZE(BB), %xmm3 pshufd $0xee, %xmm0, %xmm1 movaps -10 * SIZE(AA), %xmm0 mulpd %xmm1, %xmm2 mulpd %xmm1, %xmm3 addpd %xmm2, %xmm6 movaps 8 * SIZE(BB), %xmm2 addpd %xmm3, %xmm7 movaps 10 * SIZE(BB), %xmm3 pshufd $0x44, %xmm0, %xmm1 mulpd %xmm1, %xmm2 mulpd %xmm1, %xmm3 addpd %xmm2, %xmm4 movaps 12 * SIZE(BB), %xmm2 addpd %xmm3, %xmm5 movaps 14 * SIZE(BB), %xmm3 pshufd $0xee, %xmm0, %xmm1 movaps -8 * SIZE(AA), %xmm0 mulpd %xmm1, %xmm2 mulpd %xmm1, %xmm3 addpd %xmm2, %xmm6 movaps 16 * SIZE(BB), %xmm2 addpd %xmm3, %xmm7 movaps 18 * SIZE(BB), %xmm3 subl $ -8 * SIZE, AA subl $-32 * SIZE, BB subl $1, %eax jne .L22 ALIGN_4 .L25: #ifndef TRMMKERNEL movl K, %eax #else movl KKK, %eax #endif andl $7, %eax BRANCH je .L28 ALIGN_4 .L26: pshufd $0x44, %xmm0, %xmm1 movsd -15 * SIZE(AA), %xmm0 mulpd %xmm1, %xmm2 mulpd %xmm1, %xmm3 addpd %xmm2, %xmm4 movaps -12 * SIZE(BB), %xmm2 addpd %xmm3, %xmm5 movaps -10 * SIZE(BB), %xmm3 addl $1 * SIZE, AA addl $4 * SIZE, BB decl %eax jg .L26 ALIGN_4 .L28: movups ALPHA, %xmm3 addpd %xmm6, %xmm4 addpd %xmm7, %xmm5 leal (C1, LDC, 2), %eax movsd 0 * SIZE(C1), %xmm0 movhps 1 * SIZE(C1), %xmm0 movsd 0 * SIZE(C1, LDC), %xmm1 movhps 1 * SIZE(C1, LDC), %xmm1 pshufd $0x44, %xmm4, %xmm2 unpckhpd %xmm4, %xmm4 mulpd %xmm3, %xmm2 addpd %xmm2, %xmm0 mulpd %xmm3, %xmm4 addpd %xmm4, %xmm1 movlps %xmm0, 0 * SIZE(C1) movhps %xmm0, 1 * SIZE(C1) movlps %xmm1, 0 * SIZE(C1, LDC) movhps %xmm1, 1 * SIZE(C1, LDC) movsd 0 * SIZE(%eax), %xmm0 movhps 1 * SIZE(%eax), %xmm0 movsd 0 * SIZE(%eax, LDC), %xmm1 movhps 1 * SIZE(%eax, LDC), %xmm1 pshufd $0x44, %xmm5, %xmm2 unpckhpd %xmm5, %xmm5 mulpd %xmm3, %xmm2 addpd %xmm2, %xmm0 mulpd %xmm3, %xmm5 addpd %xmm5, %xmm1 movlps %xmm0, 0 * SIZE(%eax) movhps %xmm0, 1 * SIZE(%eax) movlps %xmm1, 0 * SIZE(%eax, LDC) movhps %xmm1, 1 * SIZE(%eax, LDC) ALIGN_4 .L29: #if defined(TRMMKERNEL) && !defined(LEFT) addl $4, KK #endif movl BB, B leal (, LDC, 4), %eax addl %eax, C decl J jg .L01 ALIGN_4 .L30: movl N, %eax testl $2, %eax jle .L50 #if defined(TRMMKERNEL) && defined(LEFT) movl OFFSET, %eax movl %eax, KK #endif movl C, C1 movl A, AA movl M, I sarl $1, I jle .L40 ALIGN_4 .L31: #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movl B, BB #else movl B, BB movl KK, %eax leal (, %eax, SIZE), %eax leal (AA, %eax, 2), AA leal (BB, %eax, 2), BB #endif movaps -16 * SIZE(AA), %xmm0 pxor %xmm4, %xmm4 movaps -16 * SIZE(BB), %xmm1 pxor %xmm5, %xmm5 PREFETCHW 1 * SIZE(C1) pxor %xmm6, %xmm6 PREFETCHW 1 * SIZE(C1, LDC) pxor %xmm7, %xmm7 #ifndef TRMMKERNEL movl K, %eax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movl K, %eax subl KK, %eax movl %eax, KKK #else movl KK, %eax #ifdef LEFT addl $2, %eax #else addl $2, %eax #endif movl %eax, KKK #endif sarl $3, %eax je .L35 ALIGN_4 .L32: PREFETCH (PREFETCHSIZE + 0) * SIZE(AA) pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 movaps -14 * SIZE(AA), %xmm0 addpd %xmm1, %xmm5 movaps -14 * SIZE(BB), %xmm1 addpd %xmm2, %xmm4 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 movaps -12 * SIZE(AA), %xmm0 addpd %xmm1, %xmm7 movaps -12 * SIZE(BB), %xmm1 addpd %xmm2, %xmm6 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 movaps -10 * SIZE(AA), %xmm0 addpd %xmm1, %xmm5 movaps -10 * SIZE(BB), %xmm1 addpd %xmm2, %xmm4 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 movaps -8 * SIZE(AA), %xmm0 addpd %xmm1, %xmm7 movaps -8 * SIZE(BB), %xmm1 addpd %xmm2, %xmm6 PREFETCH (PREFETCHSIZE + 8) * SIZE(AA) pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 movaps -6 * SIZE(AA), %xmm0 addpd %xmm1, %xmm5 movaps -6 * SIZE(BB), %xmm1 addpd %xmm2, %xmm4 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 movaps -4 * SIZE(AA), %xmm0 addpd %xmm1, %xmm7 movaps -4 * SIZE(BB), %xmm1 addpd %xmm2, %xmm6 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 movaps -2 * SIZE(AA), %xmm0 addpd %xmm1, %xmm5 movaps -2 * SIZE(BB), %xmm1 addpd %xmm2, %xmm4 pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 movaps 0 * SIZE(AA), %xmm0 addpd %xmm1, %xmm7 movaps 0 * SIZE(BB), %xmm1 addpd %xmm2, %xmm6 subl $-16 * SIZE, AA subl $-16 * SIZE, BB subl $1, %eax jne .L32 ALIGN_4 .L35: #ifndef TRMMKERNEL movl K, %eax #else movl KKK, %eax #endif andl $7, %eax BRANCH je .L38 ALIGN_4 .L36: pshufd $0x4e, %xmm1, %xmm2 mulpd %xmm0, %xmm1 mulpd %xmm0, %xmm2 movaps -14 * SIZE(AA), %xmm0 addpd %xmm1, %xmm5 movaps -14 * SIZE(BB), %xmm1 addpd %xmm2, %xmm4 addl $2 * SIZE, AA addl $2 * SIZE, BB decl %eax jg .L36 ALIGN_4 .L38: movups ALPHA, %xmm3 addpd %xmm6, %xmm4 addpd %xmm7, %xmm5 movaps %xmm4, %xmm0 movsd %xmm5, %xmm4 movsd %xmm0, %xmm5 movsd 0 * SIZE(C1), %xmm0 movhps 1 * SIZE(C1), %xmm0 movsd 2 * SIZE(C1), %xmm1 movhps 3 * SIZE(C1), %xmm1 pshufd $0x44, %xmm4, %xmm2 unpckhpd %xmm4, %xmm4 mulpd %xmm3, %xmm2 addpd %xmm2, %xmm0 mulpd %xmm3, %xmm4 addpd %xmm4, %xmm1 movlps %xmm0, 0 * SIZE(C1) movhps %xmm0, 1 * SIZE(C1) movlps %xmm1, 2 * SIZE(C1) movhps %xmm1, 3 * SIZE(C1) movsd 0 * SIZE(C1, LDC), %xmm0 movhps 1 * SIZE(C1, LDC), %xmm0 movsd 2 * SIZE(C1, LDC), %xmm1 movhps 3 * SIZE(C1, LDC), %xmm1 pshufd $0x44, %xmm5, %xmm2 unpckhpd %xmm5, %xmm5 mulpd %xmm3, %xmm2 addpd %xmm2, %xmm0 mulpd %xmm3, %xmm5 addpd %xmm5, %xmm1 movlps %xmm0, 0 * SIZE(C1, LDC) movhps %xmm0, 1 * SIZE(C1, LDC) movlps %xmm1, 2 * SIZE(C1, LDC) movhps %xmm1, 3 * SIZE(C1, LDC) addl $4 * SIZE, C1 decl I jg .L31 ALIGN_4 .L40: movl M, I testl $1, I jle .L49 #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movl B, BB #else movl B, BB movl KK, %eax leal (, %eax, SIZE), %eax addl %eax, AA leal (BB, %eax, 2), BB #endif movaps -16 * SIZE(AA), %xmm0 pxor %xmm4, %xmm4 movaps -16 * SIZE(BB), %xmm2 pxor %xmm5, %xmm5 #ifndef TRMMKERNEL movl K, %eax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movl K, %eax subl KK, %eax movl %eax, KKK #else movl KK, %eax #ifdef LEFT addl $1, %eax #else addl $2, %eax #endif movl %eax, KKK #endif sarl $3, %eax je .L45 ALIGN_4 .L42: PREFETCH (PREFETCHSIZE + 0) * SIZE(AA) pshufd $0x44, %xmm0, %xmm1 mulpd %xmm1, %xmm2 addpd %xmm2, %xmm4 movaps -14 * SIZE(BB), %xmm2 pshufd $0xee, %xmm0, %xmm1 movaps -14 * SIZE(AA), %xmm0 mulpd %xmm1, %xmm2 addpd %xmm2, %xmm5 movaps -12 * SIZE(BB), %xmm2 pshufd $0x44, %xmm0, %xmm1 mulpd %xmm1, %xmm2 addpd %xmm2, %xmm4 movaps -10 * SIZE(BB), %xmm2 pshufd $0xee, %xmm0, %xmm1 movaps -12 * SIZE(AA), %xmm0 mulpd %xmm1, %xmm2 addpd %xmm2, %xmm5 movaps -8 * SIZE(BB), %xmm2 pshufd $0x44, %xmm0, %xmm1 mulpd %xmm1, %xmm2 addpd %xmm2, %xmm4 movaps -6 * SIZE(BB), %xmm2 pshufd $0xee, %xmm0, %xmm1 movaps -10 * SIZE(AA), %xmm0 mulpd %xmm1, %xmm2 addpd %xmm2, %xmm5 movaps -4 * SIZE(BB), %xmm2 pshufd $0x44, %xmm0, %xmm1 mulpd %xmm1, %xmm2 addpd %xmm2, %xmm4 movaps -2 * SIZE(BB), %xmm2 pshufd $0xee, %xmm0, %xmm1 movaps -8 * SIZE(AA), %xmm0 mulpd %xmm1, %xmm2 addpd %xmm2, %xmm5 movaps 0 * SIZE(BB), %xmm2 subl $ -8 * SIZE, AA subl $-16 * SIZE, BB subl $1, %eax jne .L42 ALIGN_4 .L45: #ifndef TRMMKERNEL movl K, %eax #else movl KKK, %eax #endif andl $7, %eax BRANCH je .L48 ALIGN_4 .L46: pshufd $0x44, %xmm0, %xmm1 movsd -15 * SIZE(AA), %xmm0 mulpd %xmm1, %xmm2 addpd %xmm2, %xmm4 movaps -14 * SIZE(BB), %xmm2 addl $1 * SIZE, AA addl $2 * SIZE, BB decl %eax jg .L46 ALIGN_4 .L48: movups ALPHA, %xmm3 addpd %xmm5, %xmm4 movsd 0 * SIZE(C1), %xmm0 movhps 1 * SIZE(C1), %xmm0 movsd 0 * SIZE(C1, LDC), %xmm1 movhps 1 * SIZE(C1, LDC), %xmm1 pshufd $0x44, %xmm4, %xmm2 unpckhpd %xmm4, %xmm4 mulpd %xmm3, %xmm2 addpd %xmm2, %xmm0 mulpd %xmm3, %xmm4 addpd %xmm4, %xmm1 movlps %xmm0, 0 * SIZE(C1) movhps %xmm0, 1 * SIZE(C1) movlps %xmm1, 0 * SIZE(C1, LDC) movhps %xmm1, 1 * SIZE(C1, LDC) ALIGN_4 .L49: #if defined(TRMMKERNEL) && !defined(LEFT) addl $2, KK #endif movl BB, B leal (, LDC, 2), %eax addl %eax, C ALIGN_4 .L50: movl N, %eax testl $1, %eax jle .L999 #if defined(TRMMKERNEL) && defined(LEFT) movl OFFSET, %eax movl %eax, KK #endif movl C, C1 movl A, AA movl M, I sarl $1, I jle .L60 ALIGN_4 .L51: #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movl B, BB #else movl B, BB movl KK, %eax leal (, %eax, SIZE), %eax leal (AA, %eax, 2), AA addl %eax, BB #endif movaps -16 * SIZE(AA), %xmm0 pxor %xmm4, %xmm4 movaps -16 * SIZE(BB), %xmm1 pxor %xmm5, %xmm5 PREFETCHW 1 * SIZE(C1) #ifndef TRMMKERNEL movl K, %eax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movl K, %eax subl KK, %eax movl %eax, KKK #else movl KK, %eax #ifdef LEFT addl $2, %eax #else addl $1, %eax #endif movl %eax, KKK #endif sarl $3, %eax je .L55 ALIGN_4 .L52: PREFETCH (PREFETCHSIZE + 0) * SIZE(AA) pshufd $0x44, %xmm1, %xmm2 mulpd %xmm0, %xmm2 movaps -14 * SIZE(AA), %xmm0 addpd %xmm2, %xmm4 pshufd $0xee, %xmm1, %xmm2 movaps -14 * SIZE(BB), %xmm1 mulpd %xmm0, %xmm2 movaps -12 * SIZE(AA), %xmm0 addpd %xmm2, %xmm5 pshufd $0x44, %xmm1, %xmm2 mulpd %xmm0, %xmm2 movaps -10 * SIZE(AA), %xmm0 addpd %xmm2, %xmm4 pshufd $0xee, %xmm1, %xmm2 movaps -12 * SIZE(BB), %xmm1 mulpd %xmm0, %xmm2 movaps -8 * SIZE(AA), %xmm0 addpd %xmm2, %xmm5 PREFETCH (PREFETCHSIZE + 8) * SIZE(AA) pshufd $0x44, %xmm1, %xmm2 mulpd %xmm0, %xmm2 movaps -6 * SIZE(AA), %xmm0 addpd %xmm2, %xmm4 pshufd $0xee, %xmm1, %xmm2 movaps -10 * SIZE(BB), %xmm1 mulpd %xmm0, %xmm2 movaps -4 * SIZE(AA), %xmm0 addpd %xmm2, %xmm5 pshufd $0x44, %xmm1, %xmm2 mulpd %xmm0, %xmm2 movaps -2 * SIZE(AA), %xmm0 addpd %xmm2, %xmm4 pshufd $0xee, %xmm1, %xmm2 movaps -8 * SIZE(BB), %xmm1 mulpd %xmm0, %xmm2 movaps 0 * SIZE(AA), %xmm0 addpd %xmm2, %xmm5 subl $-16 * SIZE, AA subl $ -8 * SIZE, BB subl $1, %eax jne .L52 ALIGN_4 .L55: #ifndef TRMMKERNEL movl K, %eax #else movl KKK, %eax #endif andl $7, %eax BRANCH je .L58 ALIGN_4 .L56: pshufd $0x44, %xmm1, %xmm2 movsd -15 * SIZE(BB), %xmm1 mulpd %xmm0, %xmm2 movaps -14 * SIZE(AA), %xmm0 addpd %xmm2, %xmm4 addl $2 * SIZE, AA addl $1 * SIZE, BB decl %eax jg .L56 ALIGN_4 .L58: movups ALPHA, %xmm3 addpd %xmm5, %xmm4 movsd 0 * SIZE(C1), %xmm0 movhps 1 * SIZE(C1), %xmm0 movsd 2 * SIZE(C1), %xmm1 movhps 3 * SIZE(C1), %xmm1 pshufd $0x44, %xmm4, %xmm2 unpckhpd %xmm4, %xmm4 mulpd %xmm3, %xmm2 addpd %xmm2, %xmm0 mulpd %xmm3, %xmm4 addpd %xmm4, %xmm1 movlps %xmm0, 0 * SIZE(C1) movhps %xmm0, 1 * SIZE(C1) movlps %xmm1, 2 * SIZE(C1) movhps %xmm1, 3 * SIZE(C1) addl $4 * SIZE, C1 decl I jg .L51 ALIGN_4 .L60: movl M, I testl $1, I jle .L999 #if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movl B, BB #else movl B, BB movl KK, %eax leal (, %eax, SIZE), %eax addl %eax, AA addl %eax, BB #endif movaps -16 * SIZE(AA), %xmm0 pxor %xmm4, %xmm4 movaps -16 * SIZE(BB), %xmm2 pxor %xmm5, %xmm5 #ifndef TRMMKERNEL movl K, %eax #elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movl K, %eax subl KK, %eax movl %eax, KKK #else movl KK, %eax #ifdef LEFT addl $1, %eax #else addl $1, %eax #endif movl %eax, KKK #endif sarl $3, %eax je .L65 ALIGN_4 .L62: PREFETCH (PREFETCHSIZE + 0) * SIZE(AA) mulpd %xmm0, %xmm2 movaps -14 * SIZE(AA), %xmm0 addpd %xmm2, %xmm4 movaps -14 * SIZE(BB), %xmm2 mulpd %xmm0, %xmm2 movaps -12 * SIZE(AA), %xmm0 addpd %xmm2, %xmm5 movaps -12 * SIZE(BB), %xmm2 mulpd %xmm0, %xmm2 movaps -10 * SIZE(AA), %xmm0 addpd %xmm2, %xmm4 movaps -10 * SIZE(BB), %xmm2 mulpd %xmm0, %xmm2 movaps -8 * SIZE(AA), %xmm0 addpd %xmm2, %xmm5 movaps -8 * SIZE(BB), %xmm2 subl $-8 * SIZE, AA subl $-8 * SIZE, BB subl $1, %eax jne .L62 ALIGN_4 .L65: #ifndef TRMMKERNEL movl K, %eax #else movl KKK, %eax #endif andl $7, %eax BRANCH je .L68 ALIGN_4 .L66: mulsd %xmm0, %xmm2 movsd -15 * SIZE(AA), %xmm0 addsd %xmm2, %xmm4 movsd -15 * SIZE(BB), %xmm2 addl $1 * SIZE, AA addl $1 * SIZE, BB decl %eax jg .L66 ALIGN_4 .L68: movups ALPHA, %xmm3 addpd %xmm5, %xmm4 haddpd %xmm4, %xmm4 movsd 0 * SIZE(C1), %xmm0 movhps 1 * SIZE(C1), %xmm0 pshufd $0x44, %xmm4, %xmm2 mulpd %xmm3, %xmm2 addpd %xmm2, %xmm0 movlps %xmm0, 0 * SIZE(C1) movhps %xmm0, 1 * SIZE(C1) ALIGN_4 .L999: popl %ebx popl %esi popl %edi popl %ebp addl $ARGS, %esp ret EPILOGUE