/*********************************************************************/ /* 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" #ifdef PENTIUM4 #define PREFETCH prefetcht0 #define PREFETCHW prefetcht0 #define PREFETCHSIZE (8 * 2) #endif #if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM) || defined(SANDYBRIDGE) #define PREFETCH prefetcht0 #define PREFETCHW prefetcht0 #define PREFETCHSIZE (8 * 7) #endif #ifdef OPTERON #define PREFETCH prefetchnta #define PREFETCHW prefetchw #define PREFETCHSIZE (8 * 3) #define movsd movlps #endif #ifdef BARCELONA #define PREFETCH prefetchnta #define PREFETCHW prefetchw #define PREFETCHSIZE (8 * 5) #endif #ifdef ATOM #define PREFETCH prefetch #define PREFETCHW prefetcht0 #define PREFETCHSIZE (8 * 6) #endif #ifdef NANO #define PREFETCH prefetcht0 #define PREFETCHSIZE (8 * 4) #endif #define STACKSIZE 16 #define ARGS 16 #define M 4 + STACKSIZE+ARGS(%esp) #define N 8 + STACKSIZE+ARGS(%esp) #define ALPHA 16 + STACKSIZE+ARGS(%esp) #define A 24 + STACKSIZE+ARGS(%esp) #define STACK_LDA 28 + STACKSIZE+ARGS(%esp) #define STACK_X 32 + STACKSIZE+ARGS(%esp) #define STACK_INCX 36 + STACKSIZE+ARGS(%esp) #define Y 40 + STACKSIZE+ARGS(%esp) #define STACK_INCY 44 + STACKSIZE+ARGS(%esp) #define BUFFER 48 + STACKSIZE+ARGS(%esp) #define MMM 0+ARGS(%esp) #define YY 4+ARGS(%esp) #define AA 8+ARGS(%esp) #define I %eax #define J %ebx #define INCX %ecx #define INCY J #define A1 %esi #define X %edx #define Y1 %edi #define LDA %ebp PROLOGUE subl $ARGS,%esp pushl %ebp pushl %edi pushl %esi pushl %ebx PROFCODE movl Y,J movl J,YY # backup Y movl A,J movl J,AA # backup A movl M,J movl J,MMM # backup MM .L0t: xorl J,J addl $1,J sall $20,J subl J,MMM movl J,M jge .L00t ALIGN_4 movl MMM,%eax addl J,%eax jle .L999x movl %eax,M .L00t: movl AA,%eax movl %eax,A movl YY,J movl J,Y movl STACK_LDA, LDA movl STACK_X, X movl STACK_INCX, INCX leal (,INCX, SIZE), INCX leal (,LDA, SIZE), LDA subl $-16 * SIZE, A cmpl $0, N jle .L999 cmpl $0, M jle .L999 movl BUFFER, Y1 pxor %xmm7, %xmm7 movl M, %eax addl $16, %eax sarl $4, %eax ALIGN_3 .L01: movapd %xmm7, 0 * SIZE(Y1) movapd %xmm7, 2 * SIZE(Y1) movapd %xmm7, 4 * SIZE(Y1) movapd %xmm7, 6 * SIZE(Y1) movapd %xmm7, 8 * SIZE(Y1) movapd %xmm7, 10 * SIZE(Y1) movapd %xmm7, 12 * SIZE(Y1) movapd %xmm7, 14 * SIZE(Y1) subl $-16 * SIZE, Y1 decl %eax jg .L01 ALIGN_3 .L10: movl N, J sarl $1, J jle .L20 ALIGN_3 .L11: movl BUFFER, Y1 addl $16 * SIZE, Y1 movl A, A1 leal (A1, LDA, 2), %eax movl %eax, A #ifdef HAVE_SSE3 movddup (X), %xmm6 addl INCX, X movddup (X), %xmm7 addl INCX, X movddup ALPHA, %xmm0 mulpd %xmm0, %xmm6 mulpd %xmm0, %xmm7 #else movsd (X), %xmm6 addl INCX, X movsd (X), %xmm7 addl INCX, X movsd ALPHA, %xmm0 mulsd %xmm0, %xmm6 mulsd %xmm0, %xmm7 unpcklpd %xmm6, %xmm6 unpcklpd %xmm7, %xmm7 #endif ALIGN_3 movl M, I sarl $3, I jle .L15 movsd -16 * SIZE(A1), %xmm2 movhpd -15 * SIZE(A1), %xmm2 movsd -14 * SIZE(A1), %xmm3 movhpd -13 * SIZE(A1), %xmm3 movapd -16 * SIZE(Y1), %xmm0 movapd -14 * SIZE(Y1), %xmm1 movsd -16 * SIZE(A1, LDA), %xmm4 movhpd -15 * SIZE(A1, LDA), %xmm4 movsd -14 * SIZE(A1, LDA), %xmm5 movhpd -13 * SIZE(A1, LDA), %xmm5 decl I jle .L14 ALIGN_3 .L13: #ifdef PREFETCH PREFETCH (PREFETCHSIZE + 0) * SIZE(A1) #endif mulpd %xmm6, %xmm2 addpd %xmm2, %xmm0 movsd -12 * SIZE(A1), %xmm2 movhpd -11 * SIZE(A1), %xmm2 mulpd %xmm6, %xmm3 addpd %xmm3, %xmm1 movsd -10 * SIZE(A1), %xmm3 movhpd -9 * SIZE(A1), %xmm3 mulpd %xmm7, %xmm4 addpd %xmm4, %xmm0 movsd -12 * SIZE(A1, LDA), %xmm4 movhpd -11 * SIZE(A1, LDA), %xmm4 movapd %xmm0, -16 * SIZE(Y1) movapd -12 * SIZE(Y1), %xmm0 mulpd %xmm7, %xmm5 addpd %xmm5, %xmm1 movsd -10 * SIZE(A1, LDA), %xmm5 movhpd -9 * SIZE(A1, LDA), %xmm5 movapd %xmm1, -14 * SIZE(Y1) movapd -10 * SIZE(Y1), %xmm1 #ifdef PREFETCH PREFETCH (PREFETCHSIZE + 0) * SIZE(A1, LDA) #endif mulpd %xmm6, %xmm2 addpd %xmm2, %xmm0 movsd -8 * SIZE(A1), %xmm2 movhpd -7 * SIZE(A1), %xmm2 mulpd %xmm6, %xmm3 addpd %xmm3, %xmm1 movsd -6 * SIZE(A1), %xmm3 movhpd -5 * SIZE(A1), %xmm3 mulpd %xmm7, %xmm4 addpd %xmm4, %xmm0 movsd -8 * SIZE(A1, LDA), %xmm4 movhpd -7 * SIZE(A1, LDA), %xmm4 movapd %xmm0, -12 * SIZE(Y1) movapd -8 * SIZE(Y1), %xmm0 mulpd %xmm7, %xmm5 addpd %xmm5, %xmm1 movsd -6 * SIZE(A1, LDA), %xmm5 movhpd -5 * SIZE(A1, LDA), %xmm5 movapd %xmm1, -10 * SIZE(Y1) movapd -6 * SIZE(Y1), %xmm1 subl $-8 * SIZE, A1 subl $-8 * SIZE, Y1 subl $1, I BRANCH jg .L13 ALIGN_3 .L14: mulpd %xmm6, %xmm2 addpd %xmm2, %xmm0 movsd -12 * SIZE(A1), %xmm2 movhpd -11 * SIZE(A1), %xmm2 mulpd %xmm6, %xmm3 addpd %xmm3, %xmm1 movsd -10 * SIZE(A1), %xmm3 movhpd -9 * SIZE(A1), %xmm3 mulpd %xmm7, %xmm4 addpd %xmm4, %xmm0 movsd -12 * SIZE(A1, LDA), %xmm4 movhpd -11 * SIZE(A1, LDA), %xmm4 movapd %xmm0, -16 * SIZE(Y1) movapd -12 * SIZE(Y1), %xmm0 mulpd %xmm7, %xmm5 addpd %xmm5, %xmm1 movsd -10 * SIZE(A1, LDA), %xmm5 movhpd -9 * SIZE(A1, LDA), %xmm5 movapd %xmm1, -14 * SIZE(Y1) movapd -10 * SIZE(Y1), %xmm1 mulpd %xmm6, %xmm2 addpd %xmm2, %xmm0 mulpd %xmm6, %xmm3 addpd %xmm3, %xmm1 mulpd %xmm7, %xmm4 addpd %xmm4, %xmm0 movapd %xmm0, -12 * SIZE(Y1) mulpd %xmm7, %xmm5 addpd %xmm5, %xmm1 movapd %xmm1, -10 * SIZE(Y1) subl $-8 * SIZE, A1 subl $-8 * SIZE, Y1 ALIGN_3 .L15: testl $4, M je .L16 movsd -16 * SIZE(A1), %xmm2 movhpd -15 * SIZE(A1), %xmm2 movsd -14 * SIZE(A1), %xmm3 movhpd -13 * SIZE(A1), %xmm3 movapd -16 * SIZE(Y1), %xmm0 movapd -14 * SIZE(Y1), %xmm1 mulpd %xmm6, %xmm2 addpd %xmm2, %xmm0 mulpd %xmm6, %xmm3 addpd %xmm3, %xmm1 movsd -16 * SIZE(A1, LDA), %xmm4 movhpd -15 * SIZE(A1, LDA), %xmm4 movsd -14 * SIZE(A1, LDA), %xmm5 movhpd -13 * SIZE(A1, LDA), %xmm5 mulpd %xmm7, %xmm4 addpd %xmm4, %xmm0 mulpd %xmm7, %xmm5 addpd %xmm5, %xmm1 movapd %xmm0, -16 * SIZE(Y1) movapd %xmm1, -14 * SIZE(Y1) addl $4 * SIZE, A1 addl $4 * SIZE, Y1 ALIGN_3 .L16: testl $2, M je .L17 movsd -16 * SIZE(A1), %xmm2 movhpd -15 * SIZE(A1), %xmm2 movsd -16 * SIZE(A1, LDA), %xmm3 movhpd -15 * SIZE(A1, LDA), %xmm3 movapd -16 * SIZE(Y1), %xmm0 mulpd %xmm6, %xmm2 addpd %xmm2, %xmm0 mulpd %xmm7, %xmm3 addpd %xmm3, %xmm0 movapd %xmm0, -16 * SIZE(Y1) addl $2 * SIZE, A1 addl $2 * SIZE, Y1 ALIGN_3 .L17: testl $1, M je .L19 movsd -16 * SIZE(A1), %xmm2 movsd -16 * SIZE(A1, LDA), %xmm3 movsd -16 * SIZE(Y1), %xmm0 mulsd %xmm6, %xmm2 addsd %xmm2, %xmm0 mulsd %xmm7, %xmm3 addsd %xmm3, %xmm0 movsd %xmm0, -16 * SIZE(Y1) ALIGN_3 .L19: decl J jg .L11 ALIGN_4 .L20: testl $1, N jle .L990 movl BUFFER, Y1 addl $16 * SIZE, Y1 movl A, A1 #ifdef HAVE_SSE3 movddup (X), %xmm6 addl INCX, X movddup ALPHA, %xmm0 mulpd %xmm0, %xmm6 #else movsd (X), %xmm6 addl INCX, X movsd ALPHA, %xmm0 mulsd %xmm0, %xmm6 unpcklpd %xmm6, %xmm6 #endif ALIGN_3 movl M, I sarl $3, I jle .L25 movsd -16 * SIZE(A1), %xmm2 movhpd -15 * SIZE(A1), %xmm2 movsd -14 * SIZE(A1), %xmm3 movhpd -13 * SIZE(A1), %xmm3 movapd -16 * SIZE(Y1), %xmm0 movapd -14 * SIZE(Y1), %xmm1 decl I jle .L24 ALIGN_3 .L23: #ifdef PREFETCH PREFETCH (PREFETCHSIZE + 0) * SIZE(A1) #endif mulpd %xmm6, %xmm2 addpd %xmm2, %xmm0 movsd -12 * SIZE(A1), %xmm2 movhpd -11 * SIZE(A1), %xmm2 movapd %xmm0, -16 * SIZE(Y1) movapd -12 * SIZE(Y1), %xmm0 mulpd %xmm6, %xmm3 addpd %xmm3, %xmm1 movsd -10 * SIZE(A1), %xmm3 movhpd -9 * SIZE(A1), %xmm3 movapd %xmm1, -14 * SIZE(Y1) movapd -10 * SIZE(Y1), %xmm1 mulpd %xmm6, %xmm2 addpd %xmm2, %xmm0 movsd -8 * SIZE(A1), %xmm2 movhpd -7 * SIZE(A1), %xmm2 movapd %xmm0, -12 * SIZE(Y1) movapd -8 * SIZE(Y1), %xmm0 mulpd %xmm6, %xmm3 addpd %xmm3, %xmm1 movsd -6 * SIZE(A1), %xmm3 movhpd -5 * SIZE(A1), %xmm3 movapd %xmm1, -10 * SIZE(Y1) movapd -6 * SIZE(Y1), %xmm1 subl $-8 * SIZE, A1 subl $-8 * SIZE, Y1 subl $1, I BRANCH jg .L23 ALIGN_3 .L24: mulpd %xmm6, %xmm2 addpd %xmm2, %xmm0 movsd -12 * SIZE(A1), %xmm2 movhpd -11 * SIZE(A1), %xmm2 mulpd %xmm6, %xmm3 addpd %xmm3, %xmm1 movsd -10 * SIZE(A1), %xmm3 movhpd -9 * SIZE(A1), %xmm3 movapd %xmm0, -16 * SIZE(Y1) movapd -12 * SIZE(Y1), %xmm0 movapd %xmm1, -14 * SIZE(Y1) movapd -10 * SIZE(Y1), %xmm1 mulpd %xmm6, %xmm2 addpd %xmm2, %xmm0 movapd %xmm0, -12 * SIZE(Y1) mulpd %xmm6, %xmm3 addpd %xmm3, %xmm1 movapd %xmm1, -10 * SIZE(Y1) subl $-8 * SIZE, A1 subl $-8 * SIZE, Y1 ALIGN_3 .L25: testl $4, M je .L26 movsd -16 * SIZE(A1), %xmm2 movhpd -15 * SIZE(A1), %xmm2 movsd -14 * SIZE(A1), %xmm3 movhpd -13 * SIZE(A1), %xmm3 movapd -16 * SIZE(Y1), %xmm0 movapd -14 * SIZE(Y1), %xmm1 mulpd %xmm6, %xmm2 addpd %xmm2, %xmm0 mulpd %xmm6, %xmm3 addpd %xmm3, %xmm1 movapd %xmm0, -16 * SIZE(Y1) movapd %xmm1, -14 * SIZE(Y1) addl $4 * SIZE, A1 addl $4 * SIZE, Y1 ALIGN_3 .L26: testl $2, M je .L27 movsd -16 * SIZE(A1), %xmm2 movhpd -15 * SIZE(A1), %xmm2 movapd -16 * SIZE(Y1), %xmm0 mulpd %xmm6, %xmm2 addpd %xmm2, %xmm0 movapd %xmm0, -16 * SIZE(Y1) addl $2 * SIZE, A1 addl $2 * SIZE, Y1 ALIGN_3 .L27: testl $1, M je .L990 movsd -16 * SIZE(A1), %xmm2 movsd -16 * SIZE(Y1), %xmm0 mulsd %xmm6, %xmm2 addsd %xmm2, %xmm0 movsd %xmm0, -16 * SIZE(Y1) ALIGN_3 .L990: movl Y, Y1 movl BUFFER, X movl STACK_INCY, INCY sall $BASE_SHIFT, INCY movl M, %eax sarl $3, %eax jle .L994 ALIGN_3 .L992: movsd (Y1), %xmm0 movhpd (Y1, INCY), %xmm0 addpd 0 * SIZE(X), %xmm0 movlpd %xmm0, (Y1) movhpd %xmm0, (Y1, INCY) leal (Y1, INCY, 2), Y1 movsd (Y1), %xmm0 movhpd (Y1, INCY), %xmm0 addpd 2 * SIZE(X), %xmm0 movlpd %xmm0, (Y1) movhpd %xmm0, (Y1, INCY) leal (Y1, INCY, 2), Y1 movsd (Y1), %xmm0 movhpd (Y1, INCY), %xmm0 addpd 4 * SIZE(X), %xmm0 movlpd %xmm0, (Y1) movhpd %xmm0, (Y1, INCY) leal (Y1, INCY, 2), Y1 movsd (Y1), %xmm0 movhpd (Y1, INCY), %xmm0 addpd 6 * SIZE(X), %xmm0 movlpd %xmm0, (Y1) movhpd %xmm0, (Y1, INCY) leal (Y1, INCY, 2), Y1 addl $8 * SIZE, X decl %eax jg .L992 ALIGN_3 .L994: testl $7, M jle .L999 testl $4, M jle .L995 movsd (Y1), %xmm0 movhpd (Y1, INCY), %xmm0 addpd 0 * SIZE(X), %xmm0 movlpd %xmm0, (Y1) movhpd %xmm0, (Y1, INCY) leal (Y1, INCY, 2), Y1 movsd (Y1), %xmm0 movhpd (Y1, INCY), %xmm0 addpd 2 * SIZE(X), %xmm0 movlpd %xmm0, (Y1) movhpd %xmm0, (Y1, INCY) leal (Y1, INCY, 2), Y1 addl $4 * SIZE, X ALIGN_3 .L995: testl $2, M jle .L996 movsd (Y1), %xmm0 movhpd (Y1, INCY), %xmm0 addpd 0 * SIZE(X), %xmm0 movlpd %xmm0, (Y1) movhpd %xmm0, (Y1, INCY) leal (Y1, INCY, 2), Y1 addl $2 * SIZE, X ALIGN_3 .L996: testl $1, M jle .L999 movsd (Y1), %xmm0 movsd 0 * SIZE(X), %xmm4 addsd %xmm4, %xmm0 movlpd %xmm0, (Y1) ALIGN_3 .L999: movl M,J leal (,J,SIZE),%eax addl %eax,AA movl STACK_INCY,INCY imull INCY,%eax addl %eax,YY jmp .L0t ALIGN_4 .L999x: popl %ebx popl %esi popl %edi popl %ebp addl $ARGS,%esp ret EPILOGUE