From c024a0b9cf26a4c1df4da7ee5fe4098df3f59d93 Mon Sep 17 00:00:00 2001 From: Noah Goldstein Date: Thu, 14 Apr 2022 11:47:35 -0500 Subject: x86: Remove str{n}{case}cmp-ssse3 With SSE2, SSE4.1, AVX2, and EVEX versions very few targets prefer SSSE3. As a result it is no longer worth it to keep the SSSE3 versions given the code size cost. Reviewed-by: H.J. Lu --- sysdeps/x86_64/strcmp.S | 155 ++++++++++-------------------------------------- 1 file changed, 30 insertions(+), 125 deletions(-) (limited to 'sysdeps/x86_64/strcmp.S') diff --git a/sysdeps/x86_64/strcmp.S b/sysdeps/x86_64/strcmp.S index 99d8b36f1d..c38dc627f9 100644 --- a/sysdeps/x86_64/strcmp.S +++ b/sysdeps/x86_64/strcmp.S @@ -59,12 +59,7 @@ # endif #endif -#ifndef USE_SSSE3 .text -#else - .section .text.ssse3,"ax",@progbits -#endif - #ifdef USE_AS_STRCASECMP_L # ifndef ENTRY2 # define ENTRY2(name) ENTRY (name) @@ -343,13 +338,10 @@ LABEL(gobble_ashr_1): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 /* store for next cycle */ -#ifndef USE_SSSE3 psrldq $1, %xmm3 pslldq $15, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $1, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -373,13 +365,10 @@ LABEL(gobble_ashr_1): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 /* store for next cycle */ -#ifndef USE_SSSE3 psrldq $1, %xmm3 pslldq $15, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $1, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -473,13 +462,10 @@ LABEL(gobble_ashr_2): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $2, %xmm3 pslldq $14, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $2, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -504,13 +490,10 @@ LABEL(gobble_ashr_2): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $2, %xmm3 pslldq $14, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $2, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -598,13 +581,10 @@ LABEL(gobble_ashr_3): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $3, %xmm3 pslldq $13, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $3, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -629,13 +609,10 @@ LABEL(gobble_ashr_3): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $3, %xmm3 pslldq $13, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $3, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -723,13 +700,10 @@ LABEL(gobble_ashr_4): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $4, %xmm3 pslldq $12, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $4, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -754,13 +728,10 @@ LABEL(gobble_ashr_4): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $4, %xmm3 pslldq $12, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $4, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -848,13 +819,10 @@ LABEL(gobble_ashr_5): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $5, %xmm3 pslldq $11, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $5, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -879,13 +847,10 @@ LABEL(gobble_ashr_5): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $5, %xmm3 pslldq $11, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $5, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -973,13 +938,10 @@ LABEL(gobble_ashr_6): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $6, %xmm3 pslldq $10, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $6, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1004,13 +966,10 @@ LABEL(gobble_ashr_6): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $6, %xmm3 pslldq $10, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $6, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1098,13 +1057,10 @@ LABEL(gobble_ashr_7): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $7, %xmm3 pslldq $9, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $7, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1129,13 +1085,10 @@ LABEL(gobble_ashr_7): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $7, %xmm3 pslldq $9, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $7, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1223,13 +1176,10 @@ LABEL(gobble_ashr_8): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $8, %xmm3 pslldq $8, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $8, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1254,13 +1204,10 @@ LABEL(gobble_ashr_8): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $8, %xmm3 pslldq $8, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $8, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1348,13 +1295,10 @@ LABEL(gobble_ashr_9): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $9, %xmm3 pslldq $7, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $9, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1379,13 +1323,10 @@ LABEL(gobble_ashr_9): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $9, %xmm3 pslldq $7, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $9, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1473,13 +1414,10 @@ LABEL(gobble_ashr_10): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $10, %xmm3 pslldq $6, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $10, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1504,13 +1442,10 @@ LABEL(gobble_ashr_10): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $10, %xmm3 pslldq $6, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $10, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1598,13 +1533,10 @@ LABEL(gobble_ashr_11): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $11, %xmm3 pslldq $5, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $11, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1629,13 +1561,10 @@ LABEL(gobble_ashr_11): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $11, %xmm3 pslldq $5, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $11, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1723,13 +1652,10 @@ LABEL(gobble_ashr_12): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $12, %xmm3 pslldq $4, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $12, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1754,13 +1680,10 @@ LABEL(gobble_ashr_12): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $12, %xmm3 pslldq $4, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $12, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1848,13 +1771,10 @@ LABEL(gobble_ashr_13): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $13, %xmm3 pslldq $3, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $13, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1879,13 +1799,10 @@ LABEL(gobble_ashr_13): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $13, %xmm3 pslldq $3, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $13, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -1973,13 +1890,10 @@ LABEL(gobble_ashr_14): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $14, %xmm3 pslldq $2, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $14, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -2004,13 +1918,10 @@ LABEL(gobble_ashr_14): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $14, %xmm3 pslldq $2, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $14, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -2100,13 +2011,10 @@ LABEL(gobble_ashr_15): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $15, %xmm3 pslldq $1, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $15, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 @@ -2131,13 +2039,10 @@ LABEL(gobble_ashr_15): movdqa (%rdi, %rcx), %xmm2 movdqa %xmm2, %xmm4 -#ifndef USE_SSSE3 psrldq $15, %xmm3 pslldq $1, %xmm2 por %xmm3, %xmm2 /* merge into one 16byte value */ -#else - palignr $15, %xmm3, %xmm2 /* merge into one 16byte value */ -#endif + TOLOWER (%xmm1, %xmm2) pcmpeqb %xmm1, %xmm0 -- cgit 1.4.1