|
- /*********************************************************************/
- /* 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"
-
- #if defined(BARCELONA) || defined(SHANGHAI) || defined(BULLDOZER)
- #define RPREFETCHSIZE (12 + 4)
- #define WPREFETCHSIZE (12 + 4)
- #define MOVNTQ MOVQ
- #else
- #define RPREFETCHSIZE (12 + 4)
- #define WPREFETCHSIZE (12 + 4)
- #define MOVNTQ MOVQ
- #endif
-
- #ifndef WINDOWS_ABI
-
- #define M ARG1 /* rdi */
- #define N ARG2 /* rsi */
- #define A ARG3 /* rdx */
- #define LDA ARG4 /* rcx */
- #define B ARG5 /* r8 */
-
- #define I %r10
- #define J %rbp
-
- #define AO1 %r9
- #define AO2 %r15
- #define AO3 %r11
- #define AO4 %r14
- #define BO1 %r13
- #define BO2 %r12
- #define M8 %rbx
- #define BO %rax
-
- #else
-
- #define STACKSIZE 256
-
- #define M ARG1 /* rcx */
- #define N ARG2 /* rdx */
- #define A ARG3 /* r8 */
- #define LDA ARG4 /* r9 */
- #define OLD_B 40 + 64 + STACKSIZE(%rsp)
-
- #define B %rdi
-
- #define I %r10
- #define J %r11
-
- #define AO1 %r12
- #define AO2 %r13
- #define AO3 %r14
- #define AO4 %r15
-
- #define BO1 %rsi
- #define BO2 %rbx
- #define M8 %rbp
- #define BO %rax
-
- #endif
-
- #if defined(BARCELONA) || defined(SHANGHAI) || defined(BULLDOZER)
- #define RPREFETCH prefetch
- #else
- #define RPREFETCH prefetch
- #endif
-
- PROLOGUE
- PROFCODE
-
- #ifdef WINDOWS_ABI
- pushq %rdi
- pushq %rsi
- #endif
- pushq %r15
- pushq %r14
- pushq %r13
- pushq %r12
- pushq %rbp
- pushq %rbx
-
- #ifdef WINDOWS_ABI
- subq $STACKSIZE, %rsp
-
- movups %xmm6, 0(%rsp)
- movups %xmm7, 16(%rsp)
- movups %xmm8, 32(%rsp)
- movups %xmm9, 48(%rsp)
- movups %xmm10, 64(%rsp)
- movups %xmm11, 80(%rsp)
- movups %xmm12, 96(%rsp)
- movups %xmm13, 112(%rsp)
- movups %xmm14, 128(%rsp)
- movups %xmm15, 144(%rsp)
-
- movq OLD_B, B
- #endif
-
- movq N, %rax
- movq N, %rbx
- andq $-4, %rax
- andq $-2, %rbx
- imulq M, %rax
- imulq M, %rbx
-
- EMMS
-
- leaq (B, %rax, SIZE), BO1
- leaq (B, %rbx, SIZE), BO2
-
- leaq (, LDA, SIZE), LDA
- leaq (, M, SIZE), M8
- movq M, J
- sarq $2, J
- jle .L20
- ALIGN_4
-
- .L11:
- #if 0
- movq A, AO1
- leaq (A, LDA, 1), AO2
- leaq (A, LDA, 2), AO3
- leaq (AO2, LDA, 2), AO4
-
- movq N, I
- sarq $3, I
- jle .L13
- ALIGN_4
-
- .L12:
- MOVQ 0 * SIZE(AO1), %mm0
- addq $8 * SIZE, AO1
- MOVQ 0 * SIZE(AO2), %mm1
- addq $8 * SIZE, AO2
- MOVQ 0 * SIZE(AO3), %mm2
- addq $8 * SIZE, AO3
- MOVQ 0 * SIZE(AO4), %mm3
- addq $8 * SIZE, AO4
-
- decq I
- jg .L12
- ALIGN_4
-
- .L13:
- #endif
-
- movq A, AO1
- leaq (A, LDA ), AO2
- leaq (A, LDA, 2), AO3
- leaq (AO2, LDA, 2), AO4
- leaq (A, LDA, 4), A
-
- movq B, BO
- addq $16 * SIZE, B
-
- movq N, I
- sarq $2, I
- jle .L15
- ALIGN_4
-
- .L14:
-
- RPREFETCH (RPREFETCHSIZE) * SIZE(AO1)
-
- MOVQ 0 * SIZE(AO1), %mm0
- MOVNTQ %mm0, 0 * SIZE(BO)
- MOVQ 1 * SIZE(AO1), %mm1
- MOVNTQ %mm1, 1 * SIZE(BO)
-
- RPREFETCH (RPREFETCHSIZE) * SIZE(AO2)
-
- MOVQ 2 * SIZE(AO1), %mm2
- MOVNTQ %mm2, 2 * SIZE(BO)
- MOVQ 3 * SIZE(AO1), %mm3
- MOVNTQ %mm3, 3 * SIZE(BO)
-
- prefetchw (WPREFETCHSIZE + 0) * SIZE(B)
- MOVQ 0 * SIZE(AO2), %mm4
- MOVNTQ %mm4, 4 * SIZE(BO)
- MOVQ 1 * SIZE(AO2), %mm5
- MOVNTQ %mm5, 5 * SIZE(BO)
- MOVQ 2 * SIZE(AO2), %mm6
- MOVNTQ %mm6, 6 * SIZE(BO)
- MOVQ 3 * SIZE(AO2), %mm7
- MOVNTQ %mm7, 7 * SIZE(BO)
-
-
- RPREFETCH (RPREFETCHSIZE) * SIZE(AO3)
-
- MOVQ 0 * SIZE(AO3), %mm0
- MOVNTQ %mm0, 8 * SIZE(BO)
- MOVQ 1 * SIZE(AO3), %mm1
- MOVNTQ %mm1, 9 * SIZE(BO)
-
- RPREFETCH (RPREFETCHSIZE) * SIZE(AO4)
-
- MOVQ 2 * SIZE(AO3), %mm2
- MOVNTQ %mm2, 10 * SIZE(BO)
- MOVQ 3 * SIZE(AO3), %mm3
- MOVNTQ %mm3, 11 * SIZE(BO)
-
- prefetchw (WPREFETCHSIZE + 8) * SIZE(B)
- MOVQ 0 * SIZE(AO4), %mm4
- MOVNTQ %mm4, 12 * SIZE(BO)
- MOVQ 1 * SIZE(AO4), %mm5
- MOVNTQ %mm5, 13 * SIZE(BO)
- MOVQ 2 * SIZE(AO4), %mm6
- MOVNTQ %mm6, 14 * SIZE(BO)
- MOVQ 3 * SIZE(AO4), %mm7
- MOVNTQ %mm7, 15 * SIZE(BO)
-
- addq $4 * SIZE, AO1
- addq $4 * SIZE, AO2
- addq $4 * SIZE, AO3
- addq $4 * SIZE, AO4
-
- leaq (BO, M8, 4), BO
- decq I
- jg .L14
- ALIGN_4
-
- .L15:
- testq $2, N
- jle .L16
-
- MOVQ 0 * SIZE(AO1), %mm0
- MOVQ 1 * SIZE(AO1), %mm1
- MOVQ 0 * SIZE(AO2), %mm2
- MOVQ 1 * SIZE(AO2), %mm3
-
- MOVQ 0 * SIZE(AO3), %mm4
- MOVQ 1 * SIZE(AO3), %mm5
- MOVQ 0 * SIZE(AO4), %mm6
- MOVQ 1 * SIZE(AO4), %mm7
-
- MOVNTQ %mm0, 0 * SIZE(BO1)
- MOVNTQ %mm1, 1 * SIZE(BO1)
- MOVNTQ %mm2, 2 * SIZE(BO1)
- MOVNTQ %mm3, 3 * SIZE(BO1)
- MOVNTQ %mm4, 4 * SIZE(BO1)
- MOVNTQ %mm5, 5 * SIZE(BO1)
- MOVNTQ %mm6, 6 * SIZE(BO1)
- MOVNTQ %mm7, 7 * SIZE(BO1)
-
- addq $2 * SIZE, AO1
- addq $2 * SIZE, AO2
- addq $2 * SIZE, AO3
- addq $2 * SIZE, AO4
- addq $8 * SIZE, BO1
- ALIGN_4
-
- .L16:
- testq $1, N
- jle .L19
-
- MOVQ 0 * SIZE(AO1), %mm0
- MOVQ 0 * SIZE(AO2), %mm1
- MOVQ 0 * SIZE(AO3), %mm2
- MOVQ 0 * SIZE(AO4), %mm3
-
- MOVNTQ %mm0, 0 * SIZE(BO2)
- MOVNTQ %mm1, 1 * SIZE(BO2)
- MOVNTQ %mm2, 2 * SIZE(BO2)
- MOVNTQ %mm3, 3 * SIZE(BO2)
-
- addq $4 * SIZE, BO2
- ALIGN_4
-
- .L19:
- decq J
- jg .L11
- ALIGN_4
-
- .L20:
- testq $2, M
- jle .L30
- ALIGN_4
-
- .L21:
- movq A, AO1
- leaq (A, LDA ), AO2
- leaq (A, LDA, 2), A
-
- movq B, BO
- addq $8 * SIZE, B
-
- movq N, I
- sarq $2, I
- jle .L23
- ALIGN_4
-
- .L22:
- RPREFETCH (RPREFETCHSIZE) * SIZE(AO1)
- MOVQ 0 * SIZE(AO1), %mm0
- MOVQ 1 * SIZE(AO1), %mm1
- MOVQ 2 * SIZE(AO1), %mm2
- MOVQ 3 * SIZE(AO1), %mm3
-
- RPREFETCH (RPREFETCHSIZE) * SIZE(AO2)
- MOVQ 0 * SIZE(AO2), %mm4
- MOVQ 1 * SIZE(AO2), %mm5
- MOVQ 2 * SIZE(AO2), %mm6
- MOVQ 3 * SIZE(AO2), %mm7
-
- prefetchw (WPREFETCHSIZE + 0) * SIZE(B)
- MOVNTQ %mm0, 0 * SIZE(BO)
- MOVNTQ %mm1, 1 * SIZE(BO)
- MOVNTQ %mm2, 2 * SIZE(BO)
- MOVNTQ %mm3, 3 * SIZE(BO)
- MOVNTQ %mm4, 4 * SIZE(BO)
- MOVNTQ %mm5, 5 * SIZE(BO)
- MOVNTQ %mm6, 6 * SIZE(BO)
- MOVNTQ %mm7, 7 * SIZE(BO)
-
- addq $4 * SIZE, AO1
- addq $4 * SIZE, AO2
- leaq (BO, M8, 4), BO
- decq I
- jg .L22
- ALIGN_4
-
- .L23:
- testq $2, N
- jle .L24
-
- MOVQ 0 * SIZE(AO1), %mm0
- MOVQ 1 * SIZE(AO1), %mm1
- MOVQ 0 * SIZE(AO2), %mm2
- MOVQ 1 * SIZE(AO2), %mm3
-
- MOVNTQ %mm0, 0 * SIZE(BO1)
- MOVNTQ %mm1, 1 * SIZE(BO1)
- MOVNTQ %mm2, 2 * SIZE(BO1)
- MOVNTQ %mm3, 3 * SIZE(BO1)
-
- addq $2 * SIZE, AO1
- addq $2 * SIZE, AO2
- addq $4 * SIZE, BO1
- ALIGN_4
-
- .L24:
- testq $1, N
- jle .L30
-
- MOVQ 0 * SIZE(AO1), %mm0
- MOVQ 0 * SIZE(AO2), %mm1
-
- MOVNTQ %mm0, 0 * SIZE(BO2)
- MOVNTQ %mm1, 1 * SIZE(BO2)
-
- addq $2 * SIZE, BO2
- ALIGN_4
-
- .L30:
- testq $1, M
- jle .L999
- ALIGN_4
-
- .L31:
- movq A, AO1
- movq B, BO
-
- movq N, I
- sarq $2, I
- jle .L33
- ALIGN_4
-
- .L32:
- MOVQ 0 * SIZE(AO1), %mm0
- MOVQ 1 * SIZE(AO1), %mm1
- MOVQ 2 * SIZE(AO1), %mm2
- MOVQ 3 * SIZE(AO1), %mm3
-
- MOVNTQ %mm0, 0 * SIZE(BO)
- MOVNTQ %mm1, 1 * SIZE(BO)
- MOVNTQ %mm2, 2 * SIZE(BO)
- MOVNTQ %mm3, 3 * SIZE(BO)
-
- addq $4 * SIZE, AO1
- leaq (BO, M8, 4), BO
- decq I
- jg .L32
- ALIGN_4
-
- .L33:
- testq $2, N
- jle .L34
-
- MOVQ 0 * SIZE(AO1), %mm0
- MOVQ 1 * SIZE(AO1), %mm1
-
- MOVNTQ %mm0, 0 * SIZE(BO1)
- MOVNTQ %mm1, 1 * SIZE(BO1)
-
- addq $2 * SIZE, AO1
- addq $2 * SIZE, BO1
- ALIGN_4
-
- .L34:
- testq $1, N
- jle .L999
-
- MOVQ 0 * SIZE(AO1), %mm0
- MOVNTQ %mm0, 0 * SIZE(BO2)
-
- addq $1 * SIZE, BO2
- ALIGN_4
-
- .L999:
- EMMS
-
- #ifdef WINDOWS_ABI
- movups 0(%rsp), %xmm6
- movups 16(%rsp), %xmm7
- movups 32(%rsp), %xmm8
- movups 48(%rsp), %xmm9
- movups 64(%rsp), %xmm10
- movups 80(%rsp), %xmm11
- movups 96(%rsp), %xmm12
- movups 112(%rsp), %xmm13
- movups 128(%rsp), %xmm14
- movups 144(%rsp), %xmm15
-
- addq $STACKSIZE, %rsp
- #endif
-
- popq %rbx
- popq %rbp
- popq %r12
- popq %r13
- popq %r14
- popq %r15
- #ifdef WINDOWS_ABI
- popq %rsi
- popq %rdi
- #endif
- ret
-
- EPILOGUE
|