diff options
Diffstat (limited to 'sysdeps/x86_64')
24 files changed, 2995 insertions, 17 deletions
diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile index 9477538af4..5ce858823a 100644 --- a/sysdeps/x86_64/multiarch/Makefile +++ b/sysdeps/x86_64/multiarch/Makefile @@ -39,7 +39,17 @@ sysdep_routines += strncat-c stpncpy-c strncpy-c \ memmove-avx512-unaligned-erms \ memset-sse2-unaligned-erms \ memset-avx2-unaligned-erms \ - memset-avx512-unaligned-erms + memset-avx512-unaligned-erms \ + memchr-evex \ + memrchr-evex \ + rawmemchr-evex \ + strchr-evex \ + strchrnul-evex \ + strcmp-evex \ + strlen-evex \ + strncmp-evex \ + strnlen-evex \ + strrchr-evex CFLAGS-varshift.c += -msse4 CFLAGS-strcspn-c.c += -msse4 CFLAGS-strpbrk-c.c += -msse4 @@ -56,7 +66,14 @@ sysdep_routines += wmemcmp-sse4 wmemcmp-ssse3 wmemcmp-c \ wcschr-sse2 wcschr-avx2 \ wcsrchr-sse2 wcsrchr-avx2 \ wcsnlen-sse4_1 wcsnlen-c \ - wcslen-sse2 wcslen-avx2 wcsnlen-avx2 + wcslen-sse2 wcslen-avx2 wcsnlen-avx2 \ + wcschr-evex \ + wcscmp-evex \ + wcslen-evex \ + wcsncmp-evex \ + wcsnlen-evex \ + wcsrchr-evex \ + wmemchr-evex endif ifeq ($(subdir),debug) diff --git a/sysdeps/x86_64/multiarch/ifunc-avx2.h b/sysdeps/x86_64/multiarch/ifunc-avx2.h index bbaf5dcf17..634c3c3c91 100644 --- a/sysdeps/x86_64/multiarch/ifunc-avx2.h +++ b/sysdeps/x86_64/multiarch/ifunc-avx2.h @@ -21,16 +21,24 @@ extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (avx2) attribute_hidden; +extern __typeof (REDIRECT_NAME) OPTIMIZE (evex) attribute_hidden; static inline void * IFUNC_SELECTOR (void) { const struct cpu_features* cpu_features = __get_cpu_features (); - if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER) - && CPU_FEATURE_USABLE_P (cpu_features, AVX2) + if (CPU_FEATURE_USABLE_P (cpu_features, AVX2) && CPU_FEATURES_ARCH_P (cpu_features, AVX_Fast_Unaligned_Load)) - return OPTIMIZE (avx2); + { + if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL) + && CPU_FEATURE_USABLE_P (cpu_features, AVX512BW) + && CPU_FEATURE_USABLE_P (cpu_features, BMI2)) + return OPTIMIZE (evex); + + if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER)) + return OPTIMIZE (avx2); + } return OPTIMIZE (sse2); } diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c index 1be5dd0323..3bf10d3714 100644 --- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c +++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c @@ -43,6 +43,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, memchr, CPU_FEATURE_USABLE (AVX2), __memchr_avx2) + IFUNC_IMPL_ADD (array, i, memchr, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW) + && CPU_FEATURE_USABLE (BMI2)), + __memchr_evex) IFUNC_IMPL_ADD (array, i, memchr, 1, __memchr_sse2)) /* Support sysdeps/x86_64/multiarch/memcmp.c. */ @@ -121,6 +126,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, memrchr, CPU_FEATURE_USABLE (AVX2), __memrchr_avx2) + IFUNC_IMPL_ADD (array, i, memrchr, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW)), + __memrchr_evex) + IFUNC_IMPL_ADD (array, i, memrchr, 1, __memrchr_sse2)) #ifdef SHARED @@ -179,6 +189,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, rawmemchr, CPU_FEATURE_USABLE (AVX2), __rawmemchr_avx2) + IFUNC_IMPL_ADD (array, i, rawmemchr, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW) + && CPU_FEATURE_USABLE (BMI2)), + __rawmemchr_evex) IFUNC_IMPL_ADD (array, i, rawmemchr, 1, __rawmemchr_sse2)) /* Support sysdeps/x86_64/multiarch/strlen.c. */ @@ -186,6 +201,10 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, strlen, CPU_FEATURE_USABLE (AVX2), __strlen_avx2) + IFUNC_IMPL_ADD (array, i, strlen, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW)), + __strlen_evex) IFUNC_IMPL_ADD (array, i, strlen, 1, __strlen_sse2)) /* Support sysdeps/x86_64/multiarch/strnlen.c. */ @@ -193,6 +212,10 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, strnlen, CPU_FEATURE_USABLE (AVX2), __strnlen_avx2) + IFUNC_IMPL_ADD (array, i, strnlen, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW)), + __strnlen_evex) IFUNC_IMPL_ADD (array, i, strnlen, 1, __strnlen_sse2)) /* Support sysdeps/x86_64/multiarch/stpncpy.c. */ @@ -255,6 +278,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, strchr, CPU_FEATURE_USABLE (AVX2), __strchr_avx2) + IFUNC_IMPL_ADD (array, i, strchr, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW) + && CPU_FEATURE_USABLE (BMI2)), + __strchr_evex) IFUNC_IMPL_ADD (array, i, strchr, 1, __strchr_sse2_no_bsf) IFUNC_IMPL_ADD (array, i, strchr, 1, __strchr_sse2)) @@ -263,6 +291,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, strchrnul, CPU_FEATURE_USABLE (AVX2), __strchrnul_avx2) + IFUNC_IMPL_ADD (array, i, strchrnul, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW) + && CPU_FEATURE_USABLE (BMI2)), + __strchrnul_evex) IFUNC_IMPL_ADD (array, i, strchrnul, 1, __strchrnul_sse2)) /* Support sysdeps/x86_64/multiarch/strrchr.c. */ @@ -270,6 +303,10 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, strrchr, CPU_FEATURE_USABLE (AVX2), __strrchr_avx2) + IFUNC_IMPL_ADD (array, i, strrchr, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW)), + __strrchr_evex) IFUNC_IMPL_ADD (array, i, strrchr, 1, __strrchr_sse2)) /* Support sysdeps/x86_64/multiarch/strcmp.c. */ @@ -277,6 +314,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, strcmp, CPU_FEATURE_USABLE (AVX2), __strcmp_avx2) + IFUNC_IMPL_ADD (array, i, strcmp, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW) + && CPU_FEATURE_USABLE (BMI2)), + __strcmp_evex) IFUNC_IMPL_ADD (array, i, strcmp, CPU_FEATURE_USABLE (SSE4_2), __strcmp_sse42) IFUNC_IMPL_ADD (array, i, strcmp, CPU_FEATURE_USABLE (SSSE3), @@ -370,6 +412,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, wcschr, CPU_FEATURE_USABLE (AVX2), __wcschr_avx2) + IFUNC_IMPL_ADD (array, i, wcschr, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW) + && CPU_FEATURE_USABLE (BMI2)), + __wcschr_evex) IFUNC_IMPL_ADD (array, i, wcschr, 1, __wcschr_sse2)) /* Support sysdeps/x86_64/multiarch/wcsrchr.c. */ @@ -377,6 +424,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, wcsrchr, CPU_FEATURE_USABLE (AVX2), __wcsrchr_avx2) + IFUNC_IMPL_ADD (array, i, wcsrchr, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW) + && CPU_FEATURE_USABLE (BMI2)), + __wcsrchr_evex) IFUNC_IMPL_ADD (array, i, wcsrchr, 1, __wcsrchr_sse2)) /* Support sysdeps/x86_64/multiarch/wcscmp.c. */ @@ -384,6 +436,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, wcscmp, CPU_FEATURE_USABLE (AVX2), __wcscmp_avx2) + IFUNC_IMPL_ADD (array, i, wcscmp, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW) + && CPU_FEATURE_USABLE (BMI2)), + __wcscmp_evex) IFUNC_IMPL_ADD (array, i, wcscmp, 1, __wcscmp_sse2)) /* Support sysdeps/x86_64/multiarch/wcsncmp.c. */ @@ -391,6 +448,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, wcsncmp, CPU_FEATURE_USABLE (AVX2), __wcsncmp_avx2) + IFUNC_IMPL_ADD (array, i, wcsncmp, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW) + && CPU_FEATURE_USABLE (BMI2)), + __wcsncmp_evex) IFUNC_IMPL_ADD (array, i, wcsncmp, 1, __wcsncmp_sse2)) /* Support sysdeps/x86_64/multiarch/wcscpy.c. */ @@ -404,6 +466,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, wcslen, CPU_FEATURE_USABLE (AVX2), __wcslen_avx2) + IFUNC_IMPL_ADD (array, i, wcslen, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW) + && CPU_FEATURE_USABLE (BMI2)), + __wcslen_evex) IFUNC_IMPL_ADD (array, i, wcslen, 1, __wcslen_sse2)) /* Support sysdeps/x86_64/multiarch/wcsnlen.c. */ @@ -412,6 +479,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, CPU_FEATURE_USABLE (AVX2), __wcsnlen_avx2) IFUNC_IMPL_ADD (array, i, wcsnlen, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW) + && CPU_FEATURE_USABLE (BMI2)), + __wcsnlen_evex) + IFUNC_IMPL_ADD (array, i, wcsnlen, CPU_FEATURE_USABLE (SSE4_1), __wcsnlen_sse4_1) IFUNC_IMPL_ADD (array, i, wcsnlen, 1, __wcsnlen_sse2)) @@ -421,6 +493,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, wmemchr, CPU_FEATURE_USABLE (AVX2), __wmemchr_avx2) + IFUNC_IMPL_ADD (array, i, wmemchr, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW) + && CPU_FEATURE_USABLE (BMI2)), + __wmemchr_evex) IFUNC_IMPL_ADD (array, i, wmemchr, 1, __wmemchr_sse2)) /* Support sysdeps/x86_64/multiarch/wmemcmp.c. */ @@ -568,6 +645,10 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, strncmp, CPU_FEATURE_USABLE (AVX2), __strncmp_avx2) + IFUNC_IMPL_ADD (array, i, strncmp, + (CPU_FEATURE_USABLE (AVX512VL) + && CPU_FEATURE_USABLE (AVX512BW)), + __strncmp_evex) IFUNC_IMPL_ADD (array, i, strncmp, CPU_FEATURE_USABLE (SSE4_2), __strncmp_sse42) IFUNC_IMPL_ADD (array, i, strncmp, CPU_FEATURE_USABLE (SSSE3), diff --git a/sysdeps/x86_64/multiarch/memchr-evex.S b/sysdeps/x86_64/multiarch/memchr-evex.S new file mode 100644 index 0000000000..6dd5d67b90 --- /dev/null +++ b/sysdeps/x86_64/multiarch/memchr-evex.S @@ -0,0 +1,381 @@ +/* memchr/wmemchr optimized with 256-bit EVEX instructions. + Copyright (C) 2021 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <https://www.gnu.org/licenses/>. */ + +#if IS_IN (libc) + +# include <sysdep.h> + +# ifndef MEMCHR +# define MEMCHR __memchr_evex +# endif + +# ifdef USE_AS_WMEMCHR +# define VPBROADCAST vpbroadcastd +# define VPCMP vpcmpd +# define SHIFT_REG r8d +# else +# define VPBROADCAST vpbroadcastb +# define VPCMP vpcmpb +# define SHIFT_REG ecx +# endif + +# define XMMMATCH xmm16 +# define YMMMATCH ymm16 +# define YMM1 ymm17 +# define YMM2 ymm18 +# define YMM3 ymm19 +# define YMM4 ymm20 +# define YMM5 ymm21 +# define YMM6 ymm22 + +# define VEC_SIZE 32 + + .section .text.evex,"ax",@progbits +ENTRY (MEMCHR) +# ifndef USE_AS_RAWMEMCHR + /* Check for zero length. */ + test %RDX_LP, %RDX_LP + jz L(zero) +# endif + movl %edi, %ecx +# ifdef USE_AS_WMEMCHR + shl $2, %RDX_LP +# else +# ifdef __ILP32__ + /* Clear the upper 32 bits. */ + movl %edx, %edx +# endif +# endif + /* Broadcast CHAR to YMMMATCH. */ + VPBROADCAST %esi, %YMMMATCH + /* Check if we may cross page boundary with one vector load. */ + andl $(2 * VEC_SIZE - 1), %ecx + cmpl $VEC_SIZE, %ecx + ja L(cros_page_boundary) + + /* Check the first VEC_SIZE bytes. */ + VPCMP $0, (%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + +# ifndef USE_AS_RAWMEMCHR + jnz L(first_vec_x0_check) + /* Adjust length and check the end of data. */ + subq $VEC_SIZE, %rdx + jbe L(zero) +# else + jnz L(first_vec_x0) +# endif + + /* Align data for aligned loads in the loop. */ + addq $VEC_SIZE, %rdi + andl $(VEC_SIZE - 1), %ecx + andq $-VEC_SIZE, %rdi + +# ifndef USE_AS_RAWMEMCHR + /* Adjust length. */ + addq %rcx, %rdx + + subq $(VEC_SIZE * 4), %rdx + jbe L(last_4x_vec_or_less) +# endif + jmp L(more_4x_vec) + + .p2align 4 +L(cros_page_boundary): + andl $(VEC_SIZE - 1), %ecx +# ifdef USE_AS_WMEMCHR + /* NB: Divide shift count by 4 since each bit in K1 represent 4 + bytes. */ + movl %ecx, %SHIFT_REG + sarl $2, %SHIFT_REG +# endif + andq $-VEC_SIZE, %rdi + VPCMP $0, (%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + /* Remove the leading bytes. */ + sarxl %SHIFT_REG, %eax, %eax + testl %eax, %eax + jz L(aligned_more) + tzcntl %eax, %eax +# ifdef USE_AS_WMEMCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %eax +# endif +# ifndef USE_AS_RAWMEMCHR + /* Check the end of data. */ + cmpq %rax, %rdx + jbe L(zero) +# endif + addq %rdi, %rax + addq %rcx, %rax + ret + + .p2align 4 +L(aligned_more): +# ifndef USE_AS_RAWMEMCHR + /* Calculate "rdx + rcx - VEC_SIZE" with "rdx - (VEC_SIZE - rcx)" + instead of "(rdx + rcx) - VEC_SIZE" to void possible addition + overflow. */ + negq %rcx + addq $VEC_SIZE, %rcx + + /* Check the end of data. */ + subq %rcx, %rdx + jbe L(zero) +# endif + + addq $VEC_SIZE, %rdi + +# ifndef USE_AS_RAWMEMCHR + subq $(VEC_SIZE * 4), %rdx + jbe L(last_4x_vec_or_less) +# endif + +L(more_4x_vec): + /* Check the first 4 * VEC_SIZE. Only one VEC_SIZE at a time + since data is only aligned to VEC_SIZE. */ + VPCMP $0, (%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + jnz L(first_vec_x0) + + VPCMP $0, VEC_SIZE(%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + jnz L(first_vec_x1) + + VPCMP $0, (VEC_SIZE * 2)(%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + jnz L(first_vec_x2) + + VPCMP $0, (VEC_SIZE * 3)(%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + jnz L(first_vec_x3) + + addq $(VEC_SIZE * 4), %rdi + +# ifndef USE_AS_RAWMEMCHR + subq $(VEC_SIZE * 4), %rdx + jbe L(last_4x_vec_or_less) +# endif + + /* Align data to 4 * VEC_SIZE. */ + movq %rdi, %rcx + andl $(4 * VEC_SIZE - 1), %ecx + andq $-(4 * VEC_SIZE), %rdi + +# ifndef USE_AS_RAWMEMCHR + /* Adjust length. */ + addq %rcx, %rdx +# endif + + .p2align 4 +L(loop_4x_vec): + /* Compare 4 * VEC at a time forward. */ + VPCMP $0, (%rdi), %YMMMATCH, %k1 + VPCMP $0, VEC_SIZE(%rdi), %YMMMATCH, %k2 + kord %k1, %k2, %k5 + VPCMP $0, (VEC_SIZE * 2)(%rdi), %YMMMATCH, %k3 + VPCMP $0, (VEC_SIZE * 3)(%rdi), %YMMMATCH, %k4 + + kord %k3, %k4, %k6 + kortestd %k5, %k6 + jnz L(4x_vec_end) + + addq $(VEC_SIZE * 4), %rdi + +# ifdef USE_AS_RAWMEMCHR + jmp L(loop_4x_vec) +# else + subq $(VEC_SIZE * 4), %rdx + ja L(loop_4x_vec) + +L(last_4x_vec_or_less): + /* Less than 4 * VEC and aligned to VEC_SIZE. */ + addl $(VEC_SIZE * 2), %edx + jle L(last_2x_vec) + + VPCMP $0, (%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + jnz L(first_vec_x0) + + VPCMP $0, VEC_SIZE(%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + jnz L(first_vec_x1) + + VPCMP $0, (VEC_SIZE * 2)(%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + + jnz L(first_vec_x2_check) + subl $VEC_SIZE, %edx + jle L(zero) + + VPCMP $0, (VEC_SIZE * 3)(%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + + jnz L(first_vec_x3_check) + xorl %eax, %eax + ret + + .p2align 4 +L(last_2x_vec): + addl $(VEC_SIZE * 2), %edx + VPCMP $0, (%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + + jnz L(first_vec_x0_check) + subl $VEC_SIZE, %edx + jle L(zero) + + VPCMP $0, VEC_SIZE(%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + jnz L(first_vec_x1_check) + xorl %eax, %eax + ret + + .p2align 4 +L(first_vec_x0_check): + tzcntl %eax, %eax +# ifdef USE_AS_WMEMCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %eax +# endif + /* Check the end of data. */ + cmpq %rax, %rdx + jbe L(zero) + addq %rdi, %rax + ret + + .p2align 4 +L(first_vec_x1_check): + tzcntl %eax, %eax +# ifdef USE_AS_WMEMCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %eax +# endif + /* Check the end of data. */ + cmpq %rax, %rdx + jbe L(zero) + addq $VEC_SIZE, %rax + addq %rdi, %rax + ret + + .p2align 4 +L(first_vec_x2_check): + tzcntl %eax, %eax +# ifdef USE_AS_WMEMCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %eax +# endif + /* Check the end of data. */ + cmpq %rax, %rdx + jbe L(zero) + addq $(VEC_SIZE * 2), %rax + addq %rdi, %rax + ret + + .p2align 4 +L(first_vec_x3_check): + tzcntl %eax, %eax +# ifdef USE_AS_WMEMCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %eax +# endif + /* Check the end of data. */ + cmpq %rax, %rdx + jbe L(zero) + addq $(VEC_SIZE * 3), %rax + addq %rdi, %rax + ret + + .p2align 4 +L(zero): + xorl %eax, %eax + ret +# endif + + .p2align 4 +L(first_vec_x0): + tzcntl %eax, %eax +# ifdef USE_AS_WMEMCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + leaq (%rdi, %rax, 4), %rax +# else + addq %rdi, %rax +# endif + ret + + .p2align 4 +L(first_vec_x1): + tzcntl %eax, %eax +# ifdef USE_AS_WMEMCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + leaq VEC_SIZE(%rdi, %rax, 4), %rax +# else + addq $VEC_SIZE, %rax + addq %rdi, %rax +# endif + ret + + .p2align 4 +L(first_vec_x2): + tzcntl %eax, %eax +# ifdef USE_AS_WMEMCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + leaq (VEC_SIZE * 2)(%rdi, %rax, 4), %rax +# else + addq $(VEC_SIZE * 2), %rax + addq %rdi, %rax +# endif + ret + + .p2align 4 +L(4x_vec_end): + kmovd %k1, %eax + testl %eax, %eax + jnz L(first_vec_x0) + kmovd %k2, %eax + testl %eax, %eax + jnz L(first_vec_x1) + kmovd %k3, %eax + testl %eax, %eax + jnz L(first_vec_x2) + kmovd %k4, %eax + testl %eax, %eax +L(first_vec_x3): + tzcntl %eax, %eax +# ifdef USE_AS_WMEMCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + leaq (VEC_SIZE * 3)(%rdi, %rax, 4), %rax +# else + addq $(VEC_SIZE * 3), %rax + addq %rdi, %rax +# endif + ret + +END (MEMCHR) +#endif diff --git a/sysdeps/x86_64/multiarch/memrchr-evex.S b/sysdeps/x86_64/multiarch/memrchr-evex.S new file mode 100644 index 0000000000..16bf8e02b1 --- /dev/null +++ b/sysdeps/x86_64/multiarch/memrchr-evex.S @@ -0,0 +1,337 @@ +/* memrchr optimized with 256-bit EVEX instructions. + Copyright (C) 2021 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <https://www.gnu.org/licenses/>. */ + +#if IS_IN (libc) + +# include <sysdep.h> + +# define VMOVA vmovdqa64 + +# define YMMMATCH ymm16 + +# define VEC_SIZE 32 + + .section .text.evex,"ax",@progbits +ENTRY (__memrchr_evex) + /* Broadcast CHAR to YMMMATCH. */ + vpbroadcastb %esi, %YMMMATCH + + sub $VEC_SIZE, %RDX_LP + jbe L(last_vec_or_less) + + add %RDX_LP, %RDI_LP + + /* Check the last VEC_SIZE bytes. */ + vpcmpb $0, (%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + jnz L(last_vec_x0) + + subq $(VEC_SIZE * 4), %rdi + movl %edi, %ecx + andl $(VEC_SIZE - 1), %ecx + jz L(aligned_more) + + /* Align data for aligned loads in the loop. */ + addq $VEC_SIZE, %rdi + addq $VEC_SIZE, %rdx + andq $-VEC_SIZE, %rdi + subq %rcx, %rdx + + .p2align 4 +L(aligned_more): + subq $(VEC_SIZE * 4), %rdx + jbe L(last_4x_vec_or_less) + + /* Check the last 4 * VEC_SIZE. Only one VEC_SIZE at a time + since data is only aligned to VEC_SIZE. */ + vpcmpb $0, (VEC_SIZE * 3)(%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + jnz L(last_vec_x3) + + vpcmpb $0, (VEC_SIZE * 2)(%rdi), %YMMMATCH, %k2 + kmovd %k2, %eax + testl %eax, %eax + jnz L(last_vec_x2) + + vpcmpb $0, VEC_SIZE(%rdi), %YMMMATCH, %k3 + kmovd %k3, %eax + testl %eax, %eax + jnz L(last_vec_x1) + + vpcmpb $0, (%rdi), %YMMMATCH, %k4 + kmovd %k4, %eax + testl %eax, %eax + jnz L(last_vec_x0) + + /* Align data to 4 * VEC_SIZE for loop with fewer branches. + There are some overlaps with above if data isn't aligned + to 4 * VEC_SIZE. */ + movl %edi, %ecx + andl $(VEC_SIZE * 4 - 1), %ecx + jz L(loop_4x_vec) + + addq $(VEC_SIZE * 4), %rdi + addq $(VEC_SIZE * 4), %rdx + andq $-(VEC_SIZE * 4), %rdi + subq %rcx, %rdx + + .p2align 4 +L(loop_4x_vec): + /* Compare 4 * VEC at a time forward. */ + subq $(VEC_SIZE * 4), %rdi + subq $(VEC_SIZE * 4), %rdx + jbe L(last_4x_vec_or_less) + + vpcmpb $0, (%rdi), %YMMMATCH, %k1 + vpcmpb $0, VEC_SIZE(%rdi), %YMMMATCH, %k2 + kord %k1, %k2, %k5 + vpcmpb $0, (VEC_SIZE * 2)(%rdi), %YMMMATCH, %k3 + vpcmpb $0, (VEC_SIZE * 3)(%rdi), %YMMMATCH, %k4 + + kord %k3, %k4, %k6 + kortestd %k5, %k6 + jz L(loop_4x_vec) + + /* There is a match. */ + kmovd %k4, %eax + testl %eax, %eax + jnz L(last_vec_x3) + + kmovd %k3, %eax + testl %eax, %eax + jnz L(last_vec_x2) + + kmovd %k2, %eax + testl %eax, %eax + jnz L(last_vec_x1) + + kmovd %k1, %eax + bsrl %eax, %eax + addq %rdi, %rax + ret + + .p2align 4 +L(last_4x_vec_or_less): + addl $(VEC_SIZE * 4), %edx + cmpl $(VEC_SIZE * 2), %edx + jbe L(last_2x_vec) + + vpcmpb $0, (VEC_SIZE * 3)(%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + jnz L(last_vec_x3) + + vpcmpb $0, (VEC_SIZE * 2)(%rdi), %YMMMATCH, %k2 + kmovd %k2, %eax + testl %eax, %eax + jnz L(last_vec_x2) + + vpcmpb $0, VEC_SIZE(%rdi), %YMMMATCH, %k3 + kmovd %k3, %eax + testl %eax, %eax + jnz L(last_vec_x1_check) + cmpl $(VEC_SIZE * 3), %edx + jbe L(zero) + + vpcmpb $0, (%rdi), %YMMMATCH, %k4 + kmovd %k4, %eax + testl %eax, %eax + jz L(zero) + bsrl %eax, %eax + subq $(VEC_SIZE * 4), %rdx + addq %rax, %rdx + jl L(zero) + addq %rdi, %rax + ret + + .p2align 4 +L(last_2x_vec): + vpcmpb $0, (VEC_SIZE * 3)(%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + jnz L(last_vec_x3_check) + cmpl $VEC_SIZE, %edx + jbe L(zero) + + vpcmpb $0, (VEC_SIZE * 2)(%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + testl %eax, %eax + jz L(zero) + bsrl %eax, %eax + subq $(VEC_SIZE * 2), %rdx + addq %rax, %rdx + jl L(zero) + addl $(VEC_SIZE * 2), %eax + addq %rdi, %rax + ret + + .p2align 4 +L(last_vec_x0): + bsrl %eax, %eax + addq %rdi, %rax + ret + + .p2align 4 +L(last_vec_x1): + bsrl %eax, %eax + addl $VEC_SIZE, %eax + addq %rdi, %rax + ret + + .p2align 4 +L(last_vec_x2): + bsrl %eax, %eax + addl $(VEC_SIZE * 2), %eax + addq %rdi, %rax + ret + + .p2align 4 +L(last_vec_x3): + bsrl %eax, %eax + addl $(VEC_SIZE * 3), %eax + addq %rdi, %rax + ret + + .p2align 4 +L(last_vec_x1_check): + bsrl %eax, %eax + subq $(VEC_SIZE * 3), %rdx + addq %rax, %rdx + jl L(zero) + addl $VEC_SIZE, %eax + addq %rdi, %rax + ret + + .p2align 4 +L(last_vec_x3_check): + bsrl %eax, %eax + subq $VEC_SIZE, %rdx + addq %rax, %rdx + jl L(zero) + addl $(VEC_SIZE * 3), %eax + addq %rdi, %rax + ret + + .p2align 4 +L(zero): + xorl %eax, %eax + ret + + .p2align 4 +L(last_vec_or_less_aligned): + movl %edx, %ecx + + vpcmpb $0, (%rdi), %YMMMATCH, %k1 + + movl $1, %edx + /* Support rdx << 32. */ + salq %cl, %rdx + subq $1, %rdx + + kmovd %k1, %eax + + /* Remove the trailing bytes. */ + andl %edx, %eax + testl %eax, %eax + jz L(zero) + + bsrl %eax, %eax + addq %rdi, %rax + ret + + .p2align 4 +L(last_vec_or_less): + addl $VEC_SIZE, %edx + + /* Check for zero length. */ + testl %edx, %edx + jz L(zero) + + movl %edi, %ecx + andl $(VEC_SIZE - 1), %ecx + jz L(last_vec_or_less_aligned) + + movl %ecx, %esi + movl %ecx, %r8d + addl %edx, %esi + andq $-VEC_SIZE, %rdi + + subl $VEC_SIZE, %esi + ja L(last_vec_2x_aligned) + + /* Check the last VEC. */ + vpcmpb $0, (%rdi), %YMMMATCH, %k1 + kmovd %k1, %eax + + /* Remove the leading and trailing bytes. */ + sarl %cl, %eax + movl %edx, %ecx + + movl $1, %edx + sall %cl, %edx + subl $1, %edx + + andl %edx, %eax + testl %eax, %eax + jz L(zero) + + bsrl %eax, %eax + addq %rdi, %rax + addq %r8, %rax + ret + + .p2align 4 +L(last_vec_2x_aligned): + movl %esi, %ecx + + /* Check the last VEC. */ + vpcmpb $0, VEC_SIZE(%rdi), %YMMMATCH, %k1 + + movl $1, %edx + sall %cl, %edx + subl $1, %edx + + kmovd %k1, %eax + + /* Remove the trailing bytes. */ + andl %edx, %eax + + testl %eax, %eax + jnz L(last_vec_x1) + + /* Check the second last VEC. */ + vpcmpb $0, (%rdi), %YMMMATCH, %k1 + + movl %r8d, %ecx + + kmovd %k1, %eax + + /* Remove the leading bytes. Must use unsigned right shift for + bsrl below. */ + shrl %cl, %eax + testl %eax, %eax + jz L(zero) + + bsrl %eax, %eax + addq %rdi, %rax + addq %r8, %rax + ret +END (__memrchr_evex) +#endif diff --git a/sysdeps/x86_64/multiarch/rawmemchr-evex.S b/sysdeps/x86_64/multiarch/rawmemchr-evex.S new file mode 100644 index 0000000000..ec942b77ba --- /dev/null +++ b/sysdeps/x86_64/multiarch/rawmemchr-evex.S @@ -0,0 +1,4 @@ +#define MEMCHR __rawmemchr_evex +#define USE_AS_RAWMEMCHR 1 + +#include "memchr-evex.S" diff --git a/sysdeps/x86_64/multiarch/strchr-evex.S b/sysdeps/x86_64/multiarch/strchr-evex.S new file mode 100644 index 0000000000..ddc86a7058 --- /dev/null +++ b/sysdeps/x86_64/multiarch/strchr-evex.S @@ -0,0 +1,335 @@ +/* strchr/strchrnul optimized with 256-bit EVEX instructions. + Copyright (C) 2021 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <https://www.gnu.org/licenses/>. */ + +#if IS_IN (libc) + +# include <sysdep.h> + +# ifndef STRCHR +# define STRCHR __strchr_evex +# endif + +# define VMOVU vmovdqu64 +# define VMOVA vmovdqa64 + +# ifdef USE_AS_WCSCHR +# define VPBROADCAST vpbroadcastd +# define VPCMP vpcmpd +# define VPMINU vpminud +# define CHAR_REG esi +# define SHIFT_REG r8d +# else +# define VPBROADCAST vpbroadcastb +# define VPCMP vpcmpb +# define VPMINU vpminub +# define CHAR_REG sil +# define SHIFT_REG ecx +# endif + +# define XMMZERO xmm16 + +# define YMMZERO ymm16 +# define YMM0 ymm17 +# define YMM1 ymm18 +# define YMM2 ymm19 +# define YMM3 ymm20 +# define YMM4 ymm21 +# define YMM5 ymm22 +# define YMM6 ymm23 +# define YMM7 ymm24 +# define YMM8 ymm25 + +# define VEC_SIZE 32 +# define PAGE_SIZE 4096 + + .section .text.evex,"ax",@progbits +ENTRY (STRCHR) + movl %edi, %ecx +# ifndef USE_AS_STRCHRNUL + xorl %edx, %edx +# endif + + /* Broadcast CHAR to YMM0. */ + VPBROADCAST %esi, %YMM0 + + vpxorq %XMMZERO, %XMMZERO, %XMMZERO + + /* Check if we cross page boundary with one vector load. */ + andl $(PAGE_SIZE - 1), %ecx + cmpl $(PAGE_SIZE - VEC_SIZE), %ecx + ja L(cross_page_boundary) + + /* Check the first VEC_SIZE bytes. Search for both CHAR and the + null bytes. */ + VMOVU (%rdi), %YMM1 + + /* Leaves only CHARS matching esi as 0. */ + vpxorq %YMM1, %YMM0, %YMM2 + VPMINU %YMM2, %YMM1, %YMM2 + /* Each bit in K0 represents a CHAR or a null byte in YMM1. */ + VPCMP $0, %YMMZERO, %YMM2, %k0 + ktestd %k0, %k0 + jz L(more_vecs) + kmovd %k0, %eax + tzcntl %eax, %eax + /* Found CHAR or the null byte. */ +# ifdef USE_AS_WCSCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + leaq (%rdi, %rax, 4), %rax +# else + addq %rdi, %rax +# endif +# ifndef USE_AS_STRCHRNUL + cmp (%rax), %CHAR_REG + cmovne %rdx, %rax +# endif + ret + + .p2align 4 +L(more_vecs): + /* Align data for aligned loads in the loop. */ + andq $-VEC_SIZE, %rdi +L(aligned_more): + + /* Check the next 4 * VEC_SIZE. Only one VEC_SIZE at a time + since data is only aligned to VEC_SIZE. */ + VMOVA VEC_SIZE(%rdi), %YMM1 + addq $VEC_SIZE, %rdi + + /* Leaves only CHARS matching esi as 0. */ + vpxorq %YMM1, %YMM0, %YMM2 + VPMINU %YMM2, %YMM1, %YMM2 + /* Each bit in K0 represents a CHAR or a null byte in YMM1. */ + VPCMP $0, %YMMZERO, %YMM2, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x0) + + VMOVA VEC_SIZE(%rdi), %YMM1 + /* Leaves only CHARS matching esi as 0. */ + vpxorq %YMM1, %YMM0, %YMM2 + VPMINU %YMM2, %YMM1, %YMM2 + /* Each bit in K0 represents a CHAR or a null byte in YMM1. */ + VPCMP $0, %YMMZERO, %YMM2, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x1) + + VMOVA (VEC_SIZE * 2)(%rdi), %YMM1 + /* Leaves only CHARS matching esi as 0. */ + vpxorq %YMM1, %YMM0, %YMM2 + VPMINU %YMM2, %YMM1, %YMM2 + /* Each bit in K0 represents a CHAR or a null byte in YMM1. */ + VPCMP $0, %YMMZERO, %YMM2, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x2) + + VMOVA (VEC_SIZE * 3)(%rdi), %YMM1 + /* Leaves only CHARS matching esi as 0. */ + vpxorq %YMM1, %YMM0, %YMM2 + VPMINU %YMM2, %YMM1, %YMM2 + /* Each bit in K0 represents a CHAR or a null byte in YMM1. */ + VPCMP $0, %YMMZERO, %YMM2, %k0 + ktestd %k0, %k0 + jz L(prep_loop_4x) + + kmovd %k0, %eax + tzcntl %eax, %eax + /* Found CHAR or the null byte. */ +# ifdef USE_AS_WCSCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + leaq (VEC_SIZE * 3)(%rdi, %rax, 4), %rax +# else + leaq (VEC_SIZE * 3)(%rdi, %rax), %rax +# endif +# ifndef USE_AS_STRCHRNUL + cmp (%rax), %CHAR_REG + cmovne %rdx, %rax +# endif + ret + + .p2align 4 +L(first_vec_x0): + tzcntl %eax, %eax + /* Found CHAR or the null byte. */ +# ifdef USE_AS_WCSCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + leaq (%rdi, %rax, 4), %rax +# else + addq %rdi, %rax +# endif +# ifndef USE_AS_STRCHRNUL + cmp (%rax), %CHAR_REG + cmovne %rdx, %rax +# endif + ret + + .p2align 4 +L(first_vec_x1): + tzcntl %eax, %eax + /* Found CHAR or the null byte. */ +# ifdef USE_AS_WCSCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + leaq VEC_SIZE(%rdi, %rax, 4), %rax +# else + leaq VEC_SIZE(%rdi, %rax), %rax +# endif +# ifndef USE_AS_STRCHRNUL + cmp (%rax), %CHAR_REG + cmovne %rdx, %rax +# endif + ret + + .p2align 4 +L(first_vec_x2): + tzcntl %eax, %eax + /* Found CHAR or the null byte. */ +# ifdef USE_AS_WCSCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + leaq (VEC_SIZE * 2)(%rdi, %rax, 4), %rax +# else + leaq (VEC_SIZE * 2)(%rdi, %rax), %rax +# endif +# ifndef USE_AS_STRCHRNUL + cmp (%rax), %CHAR_REG + cmovne %rdx, %rax +# endif + ret + +L(prep_loop_4x): + /* Align data to 4 * VEC_SIZE. */ + andq $-(VEC_SIZE * 4), %rdi + + .p2align 4 +L(loop_4x_vec): + /* Compare 4 * VEC at a time forward. */ + VMOVA (VEC_SIZE * 4)(%rdi), %YMM1 + VMOVA (VEC_SIZE * 5)(%rdi), %YMM2 + VMOVA (VEC_SIZE * 6)(%rdi), %YMM3 + VMOVA (VEC_SIZE * 7)(%rdi), %YMM4 + + /* Leaves only CHARS matching esi as 0. */ + vpxorq %YMM1, %YMM0, %YMM5 + vpxorq %YMM2, %YMM0, %YMM6 + vpxorq %YMM3, %YMM0, %YMM7 + vpxorq %YMM4, %YMM0, %YMM8 + + VPMINU %YMM5, %YMM1, %YMM5 + VPMINU %YMM6, %YMM2, %YMM6 + VPMINU %YMM7, %YMM3, %YMM7 + VPMINU %YMM8, %YMM4, %YMM8 + + VPMINU %YMM5, %YMM6, %YMM1 + VPMINU %YMM7, %YMM8, %YMM2 + + VPMINU %YMM1, %YMM2, %YMM1 + + /* Each bit in K0 represents a CHAR or a null byte. */ + VPCMP $0, %YMMZERO, %YMM1, %k0 + + addq $(VEC_SIZE * 4), %rdi + + ktestd %k0, %k0 + jz L(loop_4x_vec) + + /* Each bit in K0 represents a CHAR or a null byte in YMM1. */ + VPCMP $0, %YMMZERO, %YMM5, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x0) + + /* Each bit in K1 represents a CHAR or a null byte in YMM2. */ + VPCMP $0, %YMMZERO, %YMM6, %k1 + kmovd %k1, %eax + testl %eax, %eax + jnz L(first_vec_x1) + + /* Each bit in K2 represents a CHAR or a null byte in YMM3. */ + VPCMP $0, %YMMZERO, %YMM7, %k2 + /* Each bit in K3 represents a CHAR or a null byte in YMM4. */ + VPCMP $0, %YMMZERO, %YMM8, %k3 + +# ifdef USE_AS_WCSCHR + /* NB: Each bit in K2/K3 represents 4-byte element. */ + kshiftlw $8, %k3, %k1 +# else + kshiftlq $32, %k3, %k1 +# endif + + /* Each bit in K1 represents a NULL or a mismatch. */ + korq %k1, %k2, %k1 + kmovq %k1, %rax + + tzcntq %rax, %rax +# ifdef USE_AS_WCSCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + leaq (VEC_SIZE * 2)(%rdi, %rax, 4), %rax +# else + leaq (VEC_SIZE * 2)(%rdi, %rax), %rax +# endif +# ifndef USE_AS_STRCHRNUL + cmp (%rax), %CHAR_REG + cmovne %rdx, %rax +# endif + ret + + /* Cold case for crossing page with first load. */ + .p2align 4 +L(cross_page_boundary): + andq $-VEC_SIZE, %rdi + andl $(VEC_SIZE - 1), %ecx + + VMOVA (%rdi), %YMM1 + + /* Leaves only CHARS matching esi as 0. */ + vpxorq %YMM1, %YMM0, %YMM2 + VPMINU %YMM2, %YMM1, %YMM2 + /* Each bit in K0 represents a CHAR or a null byte in YMM1. */ + VPCMP $0, %YMMZERO, %YMM2, %k0 + kmovd %k0, %eax + testl %eax, %eax + +# ifdef USE_AS_WCSCHR + /* NB: Divide shift count by 4 since each bit in K1 represent 4 + bytes. */ + movl %ecx, %SHIFT_REG + sarl $2, %SHIFT_REG +# endif + + /* Remove the leading bits. */ + sarxl %SHIFT_REG, %eax, %eax + testl %eax, %eax + + jz L(aligned_more) + tzcntl %eax, %eax + addq %rcx, %rdi +# ifdef USE_AS_WCSCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + leaq (%rdi, %rax, 4), %rax +# else + addq %rdi, %rax +# endif +# ifndef USE_AS_STRCHRNUL + cmp (%rax), %CHAR_REG + cmovne %rdx, %rax +# endif + ret + +END (STRCHR) +# endif diff --git a/sysdeps/x86_64/multiarch/strchr.c b/sysdeps/x86_64/multiarch/strchr.c index 4dfbe3b58b..2c0a3e78fa 100644 --- a/sysdeps/x86_64/multiarch/strchr.c +++ b/sysdeps/x86_64/multiarch/strchr.c @@ -29,17 +29,24 @@ extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2_no_bsf) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (avx2) attribute_hidden; +extern __typeof (REDIRECT_NAME) OPTIMIZE (evex) attribute_hidden; static inline void * IFUNC_SELECTOR (void) { const struct cpu_features* cpu_features = __get_cpu_features (); - if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER) - && CPU_FEATURE_USABLE_P (cpu_features, AVX2) + if (CPU_FEATURE_USABLE_P (cpu_features, AVX2) && CPU_FEATURE_USABLE_P (cpu_features, BMI2) && CPU_FEATURES_ARCH_P (cpu_features, AVX_Fast_Unaligned_Load)) - return OPTIMIZE (avx2); + { + if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL) + && CPU_FEATURE_USABLE_P (cpu_features, AVX512BW)) + return OPTIMIZE (evex); + + if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER)) + return OPTIMIZE (avx2); + } if (CPU_FEATURES_ARCH_P (cpu_features, Slow_BSF)) return OPTIMIZE (sse2_no_bsf); diff --git a/sysdeps/x86_64/multiarch/strchrnul-evex.S b/sysdeps/x86_64/multiarch/strchrnul-evex.S new file mode 100644 index 0000000000..064fe7ca9e --- /dev/null +++ b/sysdeps/x86_64/multiarch/strchrnul-evex.S @@ -0,0 +1,3 @@ +#define STRCHR __strchrnul_evex +#define USE_AS_STRCHRNUL 1 +#include "strchr-evex.S" diff --git a/sysdeps/x86_64/multiarch/strcmp-evex.S b/sysdeps/x86_64/multiarch/strcmp-evex.S new file mode 100644 index 0000000000..459eeed09f --- /dev/null +++ b/sysdeps/x86_64/multiarch/strcmp-evex.S @@ -0,0 +1,1043 @@ +/* strcmp/wcscmp/strncmp/wcsncmp optimized with 256-bit EVEX instructions. + Copyright (C) 2021 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <https://www.gnu.org/licenses/>. */ + +#if IS_IN (libc) + +# include <sysdep.h> + +# ifndef STRCMP +# define STRCMP __strcmp_evex +# endif + +# define PAGE_SIZE 4096 + +/* VEC_SIZE = Number of bytes in a ymm register */ +# define VEC_SIZE 32 + +/* Shift for dividing by (VEC_SIZE * 4). */ +# define DIVIDE_BY_VEC_4_SHIFT 7 +# if (VEC_SIZE * 4) != (1 << DIVIDE_BY_VEC_4_SHIFT) +# error (VEC_SIZE * 4) != (1 << DIVIDE_BY_VEC_4_SHIFT) +# endif + +# define VMOVU vmovdqu64 +# define VMOVA vmovdqa64 + +# ifdef USE_AS_WCSCMP +/* Compare packed dwords. */ +# define VPCMP vpcmpd +# define SHIFT_REG32 r8d +# define SHIFT_REG64 r8 +/* 1 dword char == 4 bytes. */ +# define SIZE_OF_CHAR 4 +# else +/* Compare packed bytes. */ +# define VPCMP vpcmpb +# define SHIFT_REG32 ecx +# define SHIFT_REG64 rcx +/* 1 byte char == 1 byte. */ +# define SIZE_OF_CHAR 1 +# endif + +# define XMMZERO xmm16 +# define XMM0 xmm17 +# define XMM1 xmm18 + +# define YMMZERO ymm16 +# define YMM0 ymm17 +# define YMM1 ymm18 +# define YMM2 ymm19 +# define YMM3 ymm20 +# define YMM4 ymm21 +# define YMM5 ymm22 +# define YMM6 ymm23 +# define YMM7 ymm24 + +/* Warning! + wcscmp/wcsncmp have to use SIGNED comparison for elements. + strcmp/strncmp have to use UNSIGNED comparison for elements. +*/ + +/* The main idea of the string comparison (byte or dword) using 256-bit + EVEX instructions consists of comparing (VPCMP) two ymm vectors. The + latter can be on either packed bytes or dwords depending on + USE_AS_WCSCMP. In order to check the null char, algorithm keeps the + matched bytes/dwords, requiring 5 EVEX instructions (3 VPCMP and 2 + KORD). In general, the costs of comparing VEC_SIZE bytes (32-bytes) + are 3 VPCMP and 2 KORD instructions, together with VMOVU and ktestd + instructions. Main loop (away from from page boundary) compares 4 + vectors are a time, effectively comparing 4 x VEC_SIZE bytes (128 + bytes) on each loop. + + The routine strncmp/wcsncmp (enabled by defining USE_AS_STRNCMP) logic + is the same as strcmp, except that an a maximum offset is tracked. If + the maximum offset is reached before a difference is found, zero is + returned. */ + + .section .text.evex,"ax",@progbits +ENTRY (STRCMP) +# ifdef USE_AS_STRNCMP + /* Check for simple cases (0 or 1) in offset. */ + cmp $1, %RDX_LP + je L(char0) + jb L(zero) +# ifdef USE_AS_WCSCMP + /* Convert units: from wide to byte char. */ + shl $2, %RDX_LP +# endif + /* Register %r11 tracks the maximum offset. */ + mov %RDX_LP, %R11_LP +# endif + movl %edi, %eax + xorl %edx, %edx + /* Make %XMMZERO (%YMMZERO) all zeros in this function. */ + vpxorq %XMMZERO, %XMMZERO, %XMMZERO + orl %esi, %eax + andl $(PAGE_SIZE - 1), %eax + cmpl $(PAGE_SIZE - (VEC_SIZE * 4)), %eax + jg L(cross_page) + /* Start comparing 4 vectors. */ + VMOVU (%rdi), %YMM0 + VMOVU (%rsi), %YMM1 + + /* Each bit in K0 represents a mismatch in YMM0 and YMM1. */ + VPCMP $4, %YMM0, %YMM1, %k0 + + /* Check for NULL in YMM0. */ + VPCMP $0, %YMMZERO, %YMM0, %k1 + /* Check for NULL in YMM1. */ + VPCMP $0, %YMMZERO, %YMM1, %k2 + /* Each bit in K1 represents a NULL in YMM0 or YMM1. */ + kord %k1, %k2, %k1 + + /* Each bit in K1 represents: + 1. A mismatch in YMM0 and YMM1. Or + 2. A NULL in YMM0 or YMM1. + */ + kord %k0, %k1, %k1 + + ktestd %k1, %k1 + je L(next_3_vectors) + kmovd %k1, %ecx + tzcntl %ecx, %edx +# ifdef USE_AS_WCSCMP + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %edx +# endif +# ifdef USE_AS_STRNCMP + /* Return 0 if the mismatched index (%rdx) is after the maximum + offset (%r11). */ + cmpq %r11, %rdx + jae L(zero) +# endif +# ifdef USE_AS_WCSCMP + xorl %eax, %eax + movl (%rdi, %rdx), %ecx + cmpl (%rsi, %rdx), %ecx + je L(return) +L(wcscmp_return): + setl %al + negl %eax + orl $1, %eax +L(return): +# else + movzbl (%rdi, %rdx), %eax + movzbl (%rsi, %rdx), %edx + subl %edx, %eax +# endif + ret + + .p2align 4 +L(return_vec_size): + kmovd %k1, %ecx + tzcntl %ecx, %edx +# ifdef USE_AS_WCSCMP + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %edx +# endif +# ifdef USE_AS_STRNCMP + /* Return 0 if the mismatched index (%rdx + VEC_SIZE) is after + the maximum offset (%r11). */ + addq $VEC_SIZE, %rdx + cmpq %r11, %rdx + jae L(zero) +# ifdef USE_AS_WCSCMP + xorl %eax, %eax + movl (%rdi, %rdx), %ecx + cmpl (%rsi, %rdx), %ecx + jne L(wcscmp_return) +# else + movzbl (%rdi, %rdx), %eax + movzbl (%rsi, %rdx), %edx + subl %edx, %eax +# endif +# else +# ifdef USE_AS_WCSCMP + xorl %eax, %eax + movl VEC_SIZE(%rdi, %rdx), %ecx + cmpl VEC_SIZE(%rsi, %rdx), %ecx + jne L(wcscmp_return) +# else + movzbl VEC_SIZE(%rdi, %rdx), %eax + movzbl VEC_SIZE(%rsi, %rdx), %edx + subl %edx, %eax +# endif +# endif + ret + + .p2align 4 +L(return_2_vec_size): + kmovd %k1, %ecx + tzcntl %ecx, %edx +# ifdef USE_AS_WCSCMP + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %edx +# endif +# ifdef USE_AS_STRNCMP + /* Return 0 if the mismatched index (%rdx + 2 * VEC_SIZE) is + after the maximum offset (%r11). */ + addq $(VEC_SIZE * 2), %rdx + cmpq %r11, %rdx + jae L(zero) +# ifdef USE_AS_WCSCMP + xorl %eax, %eax + movl (%rdi, %rdx), %ecx + cmpl (%rsi, %rdx), %ecx + jne L(wcscmp_return) +# else + movzbl (%rdi, %rdx), %eax + movzbl (%rsi, %rdx), %edx + subl %edx, %eax +# endif +# else +# ifdef USE_AS_WCSCMP + xorl %eax, %eax + movl (VEC_SIZE * 2)(%rdi, %rdx), %ecx + cmpl (VEC_SIZE * 2)(%rsi, %rdx), %ecx + jne L(wcscmp_return) +# else + movzbl (VEC_SIZE * 2)(%rdi, %rdx), %eax + movzbl (VEC_SIZE * 2)(%rsi, %rdx), %edx + subl %edx, %eax +# endif +# endif + ret + + .p2align 4 +L(return_3_vec_size): + kmovd %k1, %ecx + tzcntl %ecx, %edx +# ifdef USE_AS_WCSCMP + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %edx +# endif +# ifdef USE_AS_STRNCMP + /* Return 0 if the mismatched index (%rdx + 3 * VEC_SIZE) is + after the maximum offset (%r11). */ + addq $(VEC_SIZE * 3), %rdx + cmpq %r11, %rdx + jae L(zero) +# ifdef USE_AS_WCSCMP + xorl %eax, %eax + movl (%rdi, %rdx), %ecx + cmpl (%rsi, %rdx), %ecx + jne L(wcscmp_return) +# else + movzbl (%rdi, %rdx), %eax + movzbl (%rsi, %rdx), %edx + subl %edx, %eax +# endif +# else +# ifdef USE_AS_WCSCMP + xorl %eax, %eax + movl (VEC_SIZE * 3)(%rdi, %rdx), %ecx + cmpl (VEC_SIZE * 3)(%rsi, %rdx), %ecx + jne L(wcscmp_return) +# else + movzbl (VEC_SIZE * 3)(%rdi, %rdx), %eax + movzbl (VEC_SIZE * 3)(%rsi, %rdx), %edx + subl %edx, %eax +# endif +# endif + ret + + .p2align 4 +L(next_3_vectors): + VMOVU VEC_SIZE(%rdi), %YMM0 + VMOVU VEC_SIZE(%rsi), %YMM1 + /* Each bit in K0 represents a mismatch in YMM0 and YMM1. */ + VPCMP $4, %YMM0, %YMM1, %k0 + VPCMP $0, %YMMZERO, %YMM0, %k1 + VPCMP $0, %YMMZERO, %YMM1, %k2 + /* Each bit in K1 represents a NULL in YMM0 or YMM1. */ + kord %k1, %k2, %k1 + /* Each bit in K1 represents a NULL or a mismatch. */ + kord %k0, %k1, %k1 + ktestd %k1, %k1 + jne L(return_vec_size) + + VMOVU (VEC_SIZE * 2)(%rdi), %YMM2 + VMOVU (VEC_SIZE * 3)(%rdi), %YMM3 + VMOVU (VEC_SIZE * 2)(%rsi), %YMM4 + VMOVU (VEC_SIZE * 3)(%rsi), %YMM5 + + /* Each bit in K0 represents a mismatch in YMM2 and YMM4. */ + VPCMP $4, %YMM2, %YMM4, %k0 + VPCMP $0, %YMMZERO, %YMM2, %k1 + VPCMP $0, %YMMZERO, %YMM4, %k2 + /* Each bit in K1 represents a NULL in YMM2 or YMM4. */ + kord %k1, %k2, %k1 + /* Each bit in K1 represents a NULL or a mismatch. */ + kord %k0, %k1, %k1 + ktestd %k1, %k1 + jne L(return_2_vec_size) + + /* Each bit in K0 represents a mismatch in YMM3 and YMM5. */ + VPCMP $4, %YMM3, %YMM5, %k0 + VPCMP $0, %YMMZERO, %YMM3, %k1 + VPCMP $0, %YMMZERO, %YMM5, %k2 + /* Each bit in K1 represents a NULL in YMM3 or YMM5. */ + kord %k1, %k2, %k1 + /* Each bit in K1 represents a NULL or a mismatch. */ + kord %k0, %k1, %k1 + ktestd %k1, %k1 + jne L(return_3_vec_size) +L(main_loop_header): + leaq (VEC_SIZE * 4)(%rdi), %rdx + movl $PAGE_SIZE, %ecx + /* Align load via RAX. */ + andq $-(VEC_SIZE * 4), %rdx + subq %rdi, %rdx + leaq (%rdi, %rdx), %rax +# ifdef USE_AS_STRNCMP + /* Starting from this point, the maximum offset, or simply the + 'offset', DECREASES by the same amount when base pointers are + moved forward. Return 0 when: + 1) On match: offset <= the matched vector index. + 2) On mistmach, offset is before the mistmatched index. + */ + subq %rdx, %r11 + jbe L(zero) +# endif + addq %rsi, %rdx + movq %rdx, %rsi + andl $(PAGE_SIZE - 1), %esi + /* Number of bytes before page crossing. */ + subq %rsi, %rcx + /* Number of VEC_SIZE * 4 blocks before page crossing. */ + shrq $DIVIDE_BY_VEC_4_SHIFT, %rcx + /* ESI: Number of VEC_SIZE * 4 blocks before page crossing. */ + movl %ecx, %esi + jmp L(loop_start) + + .p2align 4 +L(loop): +# ifdef USE_AS_STRNCMP + /* Base pointers are moved forward by 4 * VEC_SIZE. Decrease + the maximum offset (%r11) by the same amount. */ + subq $(VEC_SIZE * 4), %r11 + jbe L(zero) +# endif + addq $(VEC_SIZE * 4), %rax + addq $(VEC_SIZE * 4), %rdx +L(loop_start): + testl %esi, %esi + leal -1(%esi), %esi + je L(loop_cross_page) +L(back_to_loop): + /* Main loop, comparing 4 vectors are a time. */ + VMOVA (%rax), %YMM0 + VMOVA VEC_SIZE(%rax), %YMM2 + VMOVA (VEC_SIZE * 2)(%rax), %YMM4 + VMOVA (VEC_SIZE * 3)(%rax), %YMM6 + VMOVU (%rdx), %YMM1 + VMOVU VEC_SIZE(%rdx), %YMM3 + VMOVU (VEC_SIZE * 2)(%rdx), %YMM5 + VMOVU (VEC_SIZE * 3)(%rdx), %YMM7 + + VPCMP $4, %YMM0, %YMM1, %k0 + VPCMP $0, %YMMZERO, %YMM0, %k1 + VPCMP $0, %YMMZERO, %YMM1, %k2 + kord %k1, %k2, %k1 + /* Each bit in K4 represents a NULL or a mismatch in YMM0 and + YMM1. */ + kord %k0, %k1, %k4 + + VPCMP $4, %YMM2, %YMM3, %k0 + VPCMP $0, %YMMZERO, %YMM2, %k1 + VPCMP $0, %YMMZERO, %YMM3, %k2 + kord %k1, %k2, %k1 + /* Each bit in K5 represents a NULL or a mismatch in YMM2 and + YMM3. */ + kord %k0, %k1, %k5 + + VPCMP $4, %YMM4, %YMM5, %k0 + VPCMP $0, %YMMZERO, %YMM4, %k1 + VPCMP $0, %YMMZERO, %YMM5, %k2 + kord %k1, %k2, %k1 + /* Each bit in K6 represents a NULL or a mismatch in YMM4 and + YMM5. */ + kord %k0, %k1, %k6 + + VPCMP $4, %YMM6, %YMM7, %k0 + VPCMP $0, %YMMZERO, %YMM6, %k1 + VPCMP $0, %YMMZERO, %YMM7, %k2 + kord %k1, %k2, %k1 + /* Each bit in K7 represents a NULL or a mismatch in YMM6 and + YMM7. */ + kord %k0, %k1, %k7 + + kord %k4, %k5, %k0 + kord %k6, %k7, %k1 + + /* Test each mask (32 bits) individually because for VEC_SIZE + == 32 is not possible to OR the four masks and keep all bits + in a 64-bit integer register, differing from SSE2 strcmp + where ORing is possible. */ + kortestd %k0, %k1 + je L(loop) + ktestd %k4, %k4 + je L(test_vec) + kmovd %k4, %edi + tzcntl %edi, %ecx +# ifdef USE_AS_WCSCMP + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %ecx +# endif +# ifdef USE_AS_STRNCMP + cmpq %rcx, %r11 + jbe L(zero) +# ifdef USE_AS_WCSCMP + movq %rax, %rsi + xorl %eax, %eax + movl (%rsi, %rcx), %edi + cmpl (%rdx, %rcx), %edi + jne L(wcscmp_return) +# else + movzbl (%rax, %rcx), %eax + movzbl (%rdx, %rcx), %edx + subl %edx, %eax +# endif +# else +# ifdef USE_AS_WCSCMP + movq %rax, %rsi + xorl %eax, %eax + movl (%rsi, %rcx), %edi + cmpl (%rdx, %rcx), %edi + jne L(wcscmp_return) +# else + movzbl (%rax, %rcx), %eax + movzbl (%rdx, %rcx), %edx + subl %edx, %eax +# endif +# endif + ret + + .p2align 4 +L(test_vec): +# ifdef USE_AS_STRNCMP + /* The first vector matched. Return 0 if the maximum offset + (%r11) <= VEC_SIZE. */ + cmpq $VEC_SIZE, %r11 + jbe L(zero) +# endif + ktestd %k5, %k5 + je L(test_2_vec) + kmovd %k5, %ecx + tzcntl %ecx, %edi +# ifdef USE_AS_WCSCMP + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %edi +# endif +# ifdef USE_AS_STRNCMP + addq $VEC_SIZE, %rdi + cmpq %rdi, %r11 + jbe L(zero) +# ifdef USE_AS_WCSCMP + movq %rax, %rsi + xorl %eax, %eax + movl (%rsi, %rdi), %ecx + cmpl (%rdx, %rdi), %ecx + jne L(wcscmp_return) +# else + movzbl (%rax, %rdi), %eax + movzbl (%rdx, %rdi), %edx + subl %edx, %eax +# endif +# else +# ifdef USE_AS_WCSCMP + movq %rax, %rsi + xorl %eax, %eax + movl VEC_SIZE(%rsi, %rdi), %ecx + cmpl VEC_SIZE(%rdx, %rdi), %ecx + jne L(wcscmp_return) +# else + movzbl VEC_SIZE(%rax, %rdi), %eax + movzbl VEC_SIZE(%rdx, %rdi), %edx + subl %edx, %eax +# endif +# endif + ret + + .p2align 4 +L(test_2_vec): +# ifdef USE_AS_STRNCMP + /* The first 2 vectors matched. Return 0 if the maximum offset + (%r11) <= 2 * VEC_SIZE. */ + cmpq $(VEC_SIZE * 2), %r11 + jbe L(zero) +# endif + ktestd %k6, %k6 + je L(test_3_vec) + kmovd %k6, %ecx + tzcntl %ecx, %edi +# ifdef USE_AS_WCSCMP + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %edi +# endif +# ifdef USE_AS_STRNCMP + addq $(VEC_SIZE * 2), %rdi + cmpq %rdi, %r11 + jbe L(zero) +# ifdef USE_AS_WCSCMP + movq %rax, %rsi + xorl %eax, %eax + movl (%rsi, %rdi), %ecx + cmpl (%rdx, %rdi), %ecx + jne L(wcscmp_return) +# else + movzbl (%rax, %rdi), %eax + movzbl (%rdx, %rdi), %edx + subl %edx, %eax +# endif +# else +# ifdef USE_AS_WCSCMP + movq %rax, %rsi + xorl %eax, %eax + movl (VEC_SIZE * 2)(%rsi, %rdi), %ecx + cmpl (VEC_SIZE * 2)(%rdx, %rdi), %ecx + jne L(wcscmp_return) +# else + movzbl (VEC_SIZE * 2)(%rax, %rdi), %eax + movzbl (VEC_SIZE * 2)(%rdx, %rdi), %edx + subl %edx, %eax +# endif +# endif + ret + + .p2align 4 +L(test_3_vec): +# ifdef USE_AS_STRNCMP + /* The first 3 vectors matched. Return 0 if the maximum offset + (%r11) <= 3 * VEC_SIZE. */ + cmpq $(VEC_SIZE * 3), %r11 + jbe L(zero) +# endif + kmovd %k7, %esi + tzcntl %esi, %ecx +# ifdef USE_AS_WCSCMP + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %ecx +# endif +# ifdef USE_AS_STRNCMP + addq $(VEC_SIZE * 3), %rcx + cmpq %rcx, %r11 + jbe L(zero) +# ifdef USE_AS_WCSCMP + movq %rax, %rsi + xorl %eax, %eax + movl (%rsi, %rcx), %esi + cmpl (%rdx, %rcx), %esi + jne L(wcscmp_return) +# else + movzbl (%rax, %rcx), %eax + movzbl (%rdx, %rcx), %edx + subl %edx, %eax +# endif +# else +# ifdef USE_AS_WCSCMP + movq %rax, %rsi + xorl %eax, %eax + movl (VEC_SIZE * 3)(%rsi, %rcx), %esi + cmpl (VEC_SIZE * 3)(%rdx, %rcx), %esi + jne L(wcscmp_return) +# else + movzbl (VEC_SIZE * 3)(%rax, %rcx), %eax + movzbl (VEC_SIZE * 3)(%rdx, %rcx), %edx + subl %edx, %eax +# endif +# endif + ret + + .p2align 4 +L(loop_cross_page): + xorl %r10d, %r10d + movq %rdx, %rcx + /* Align load via RDX. We load the extra ECX bytes which should + be ignored. */ + andl $((VEC_SIZE * 4) - 1), %ecx + /* R10 is -RCX. */ + subq %rcx, %r10 + + /* This works only if VEC_SIZE * 2 == 64. */ +# if (VEC_SIZE * 2) != 64 +# error (VEC_SIZE * 2) != 64 +# endif + + /* Check if the first VEC_SIZE * 2 bytes should be ignored. */ + cmpl $(VEC_SIZE * 2), %ecx + jge L(loop_cross_page_2_vec) + + VMOVU (%rax, %r10), %YMM2 + VMOVU VEC_SIZE(%rax, %r10), %YMM3 + VMOVU (%rdx, %r10), %YMM4 + VMOVU VEC_SIZE(%rdx, %r10), %YMM5 + + VPCMP $4, %YMM4, %YMM2, %k0 + VPCMP $0, %YMMZERO, %YMM2, %k1 + VPCMP $0, %YMMZERO, %YMM4, %k2 + kord %k1, %k2, %k1 + /* Each bit in K1 represents a NULL or a mismatch in YMM2 and + YMM4. */ + kord %k0, %k1, %k1 + + VPCMP $4, %YMM5, %YMM3, %k3 + VPCMP $0, %YMMZERO, %YMM3, %k4 + VPCMP $0, %YMMZERO, %YMM5, %k5 + kord %k4, %k5, %k4 + /* Each bit in K3 represents a NULL or a mismatch in YMM3 and + YMM5. */ + kord %k3, %k4, %k3 + +# ifdef USE_AS_WCSCMP + /* NB: Each bit in K1/K3 represents 4-byte element. */ + kshiftlw $8, %k3, %k2 + /* NB: Divide shift count by 4 since each bit in K1 represent 4 + bytes. */ + movl %ecx, %SHIFT_REG32 + sarl $2, %SHIFT_REG32 +# else + kshiftlq $32, %k3, %k2 +# endif + + /* Each bit in K1 represents a NULL or a mismatch. */ + korq %k1, %k2, %k1 + kmovq %k1, %rdi + + /* Since ECX < VEC_SIZE * 2, simply skip the first ECX bytes. */ + shrxq %SHIFT_REG64, %rdi, %rdi + testq %rdi, %rdi + je L(loop_cross_page_2_vec) + tzcntq %rdi, %rcx +# ifdef USE_AS_WCSCMP + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %ecx +# endif +# ifdef USE_AS_STRNCMP + cmpq %rcx, %r11 + jbe L(zero) +# ifdef USE_AS_WCSCMP + movq %rax, %rsi + xorl %eax, %eax + movl (%rsi, %rcx), %edi + cmpl (%rdx, %rcx), %edi + jne L(wcscmp_return) +# else + movzbl (%rax, %rcx), %eax + movzbl (%rdx, %rcx), %edx + subl %edx, %eax +# endif +# else +# ifdef USE_AS_WCSCMP + movq %rax, %rsi + xorl %eax, %eax + movl (%rsi, %rcx), %edi + cmpl (%rdx, %rcx), %edi + jne L(wcscmp_return) +# else + movzbl (%rax, %rcx), %eax + movzbl (%rdx, %rcx), %edx + subl %edx, %eax +# endif +# endif + ret + + .p2align 4 +L(loop_cross_page_2_vec): + /* The first VEC_SIZE * 2 bytes match or are ignored. */ + VMOVU (VEC_SIZE * 2)(%rax, %r10), %YMM0 + VMOVU (VEC_SIZE * 3)(%rax, %r10), %YMM1 + VMOVU (VEC_SIZE * 2)(%rdx, %r10), %YMM2 + VMOVU (VEC_SIZE * 3)(%rdx, %r10), %YMM3 + + VPCMP $4, %YMM0, %YMM2, %k0 + VPCMP $0, %YMMZERO, %YMM0, %k1 + VPCMP $0, %YMMZERO, %YMM2, %k2 + kord %k1, %k2, %k1 + /* Each bit in K1 represents a NULL or a mismatch in YMM0 and + YMM2. */ + kord %k0, %k1, %k1 + + VPCMP $4, %YMM1, %YMM3, %k3 + VPCMP $0, %YMMZERO, %YMM1, %k4 + VPCMP $0, %YMMZERO, %YMM3, %k5 + kord %k4, %k5, %k4 + /* Each bit in K3 represents a NULL or a mismatch in YMM1 and + YMM3. */ + kord %k3, %k4, %k3 + +# ifdef USE_AS_WCSCMP + /* NB: Each bit in K1/K3 represents 4-byte element. */ + kshiftlw $8, %k3, %k2 +# else + kshiftlq $32, %k3, %k2 +# endif + + /* Each bit in K1 represents a NULL or a mismatch. */ + korq %k1, %k2, %k1 + kmovq %k1, %rdi + + xorl %r8d, %r8d + /* If ECX > VEC_SIZE * 2, skip ECX - (VEC_SIZE * 2) bytes. */ + subl $(VEC_SIZE * 2), %ecx + jle 1f + /* R8 has number of bytes skipped. */ + movl %ecx, %r8d +# ifdef USE_AS_WCSCMP + /* NB: Divide shift count by 4 since each bit in K1 represent 4 + bytes. */ + sarl $2, %ecx +# endif + /* Skip ECX bytes. */ + shrq %cl, %rdi +1: + /* Before jumping back to the loop, set ESI to the number of + VEC_SIZE * 4 blocks before page crossing. */ + movl $(PAGE_SIZE / (VEC_SIZE * 4) - 1), %esi + + testq %rdi, %rdi +# ifdef USE_AS_STRNCMP + /* At this point, if %rdi value is 0, it already tested + VEC_SIZE*4+%r10 byte starting from %rax. This label + checks whether strncmp maximum offset reached or not. */ + je L(string_nbyte_offset_check) +# else + je L(back_to_loop) +# endif + tzcntq %rdi, %rcx +# ifdef USE_AS_WCSCMP + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %ecx +# endif + addq %r10, %rcx + /* Adjust for number of bytes skipped. */ + addq %r8, %rcx +# ifdef USE_AS_STRNCMP + addq $(VEC_SIZE * 2), %rcx + subq %rcx, %r11 + jbe L(zero) +# ifdef USE_AS_WCSCMP + movq %rax, %rsi + xorl %eax, %eax + movl (%rsi, %rcx), %edi + cmpl (%rdx, %rcx), %edi + jne L(wcscmp_return) +# else + movzbl (%rax, %rcx), %eax + movzbl (%rdx, %rcx), %edx + subl %edx, %eax +# endif +# else +# ifdef USE_AS_WCSCMP + movq %rax, %rsi + xorl %eax, %eax + movl (VEC_SIZE * 2)(%rsi, %rcx), %edi + cmpl (VEC_SIZE * 2)(%rdx, %rcx), %edi + jne L(wcscmp_return) +# else + movzbl (VEC_SIZE * 2)(%rax, %rcx), %eax + movzbl (VEC_SIZE * 2)(%rdx, %rcx), %edx + subl %edx, %eax +# endif +# endif + ret + +# ifdef USE_AS_STRNCMP +L(string_nbyte_offset_check): + leaq (VEC_SIZE * 4)(%r10), %r10 + cmpq %r10, %r11 + jbe L(zero) + jmp L(back_to_loop) +# endif + + .p2align 4 +L(cross_page_loop): + /* Check one byte/dword at a time. */ +# ifdef USE_AS_WCSCMP + cmpl %ecx, %eax +# else + subl %ecx, %eax +# endif + jne L(different) + addl $SIZE_OF_CHAR, %edx + cmpl $(VEC_SIZE * 4), %edx + je L(main_loop_header) +# ifdef USE_AS_STRNCMP + cmpq %r11, %rdx + jae L(zero) +# endif +# ifdef USE_AS_WCSCMP + movl (%rdi, %rdx), %eax + movl (%rsi, %rdx), %ecx +# else + movzbl (%rdi, %rdx), %eax + movzbl (%rsi, %rdx), %ecx +# endif + /* Check null char. */ + testl %eax, %eax + jne L(cross_page_loop) + /* Since %eax == 0, subtract is OK for both SIGNED and UNSIGNED + comparisons. */ + subl %ecx, %eax +# ifndef USE_AS_WCSCMP +L(different): +# endif + ret + +# ifdef USE_AS_WCSCMP + .p2align 4 +L(different): + /* Use movl to avoid modifying EFLAGS. */ + movl $0, %eax + setl %al + negl %eax + orl $1, %eax + ret +# endif + +# ifdef USE_AS_STRNCMP + .p2align 4 +L(zero): + xorl %eax, %eax + ret + + .p2align 4 +L(char0): +# ifdef USE_AS_WCSCMP + xorl %eax, %eax + movl (%rdi), %ecx + cmpl (%rsi), %ecx + jne L(wcscmp_return) +# else + movzbl (%rsi), %ecx + movzbl (%rdi), %eax + subl %ecx, %eax +# endif + ret +# endif + + .p2align 4 +L(last_vector): + addq %rdx, %rdi + addq %rdx, %rsi +# ifdef USE_AS_STRNCMP + subq %rdx, %r11 +# endif + tzcntl %ecx, %edx +# ifdef USE_AS_WCSCMP + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %edx +# endif +# ifdef USE_AS_STRNCMP + cmpq %r11, %rdx + jae L(zero) +# endif +# ifdef USE_AS_WCSCMP + xorl %eax, %eax + movl (%rdi, %rdx), %ecx + cmpl (%rsi, %rdx), %ecx + jne L(wcscmp_return) +# else + movzbl (%rdi, %rdx), %eax + movzbl (%rsi, %rdx), %edx + subl %edx, %eax +# endif + ret + + /* Comparing on page boundary region requires special treatment: + It must done one vector at the time, starting with the wider + ymm vector if possible, if not, with xmm. If fetching 16 bytes + (xmm) still passes the boundary, byte comparison must be done. + */ + .p2align 4 +L(cross_page): + /* Try one ymm vector at a time. */ + cmpl $(PAGE_SIZE - VEC_SIZE), %eax + jg L(cross_page_1_vector) +L(loop_1_vector): + VMOVU (%rdi, %rdx), %YMM0 + VMOVU (%rsi, %rdx), %YMM1 + + /* Each bit in K0 represents a mismatch in YMM0 and YMM1. */ + VPCMP $4, %YMM0, %YMM1, %k0 + VPCMP $0, %YMMZERO, %YMM0, %k1 + VPCMP $0, %YMMZERO, %YMM1, %k2 + /* Each bit in K1 represents a NULL in YMM0 or YMM1. */ + kord %k1, %k2, %k1 + /* Each bit in K1 represents a NULL or a mismatch. */ + kord %k0, %k1, %k1 + kmovd %k1, %ecx + testl %ecx, %ecx + jne L(last_vector) + + addl $VEC_SIZE, %edx + + addl $VEC_SIZE, %eax +# ifdef USE_AS_STRNCMP + /* Return 0 if the current offset (%rdx) >= the maximum offset + (%r11). */ + cmpq %r11, %rdx + jae L(zero) +# endif + cmpl $(PAGE_SIZE - VEC_SIZE), %eax + jle L(loop_1_vector) +L(cross_page_1_vector): + /* Less than 32 bytes to check, try one xmm vector. */ + cmpl $(PAGE_SIZE - 16), %eax + jg L(cross_page_1_xmm) + VMOVU (%rdi, %rdx), %XMM0 + VMOVU (%rsi, %rdx), %XMM1 + + /* Each bit in K0 represents a mismatch in XMM0 and XMM1. */ + VPCMP $4, %XMM0, %XMM1, %k0 + VPCMP $0, %XMMZERO, %XMM0, %k1 + VPCMP $0, %XMMZERO, %XMM1, %k2 + /* Each bit in K1 represents a NULL in XMM0 or XMM1. */ + korw %k1, %k2, %k1 + /* Each bit in K1 represents a NULL or a mismatch. */ + korw %k0, %k1, %k1 + kmovw %k1, %ecx + testl %ecx, %ecx + jne L(last_vector) + + addl $16, %edx +# ifndef USE_AS_WCSCMP + addl $16, %eax +# endif +# ifdef USE_AS_STRNCMP + /* Return 0 if the current offset (%rdx) >= the maximum offset + (%r11). */ + cmpq %r11, %rdx + jae L(zero) +# endif + +L(cross_page_1_xmm): +# ifndef USE_AS_WCSCMP + /* Less than 16 bytes to check, try 8 byte vector. NB: No need + for wcscmp nor wcsncmp since wide char is 4 bytes. */ + cmpl $(PAGE_SIZE - 8), %eax + jg L(cross_page_8bytes) + vmovq (%rdi, %rdx), %XMM0 + vmovq (%rsi, %rdx), %XMM1 + + /* Each bit in K0 represents a mismatch in XMM0 and XMM1. */ + VPCMP $4, %XMM0, %XMM1, %k0 + VPCMP $0, %XMMZERO, %XMM0, %k1 + VPCMP $0, %XMMZERO, %XMM1, %k2 + /* Each bit in K1 represents a NULL in XMM0 or XMM1. */ + kord %k1, %k2, %k1 + /* Each bit in K1 represents a NULL or a mismatch. */ + kord %k0, %k1, %k1 + kmovd %k1, %ecx + +# ifdef USE_AS_WCSCMP + /* Only last 2 bits are valid. */ + andl $0x3, %ecx +# else + /* Only last 8 bits are valid. */ + andl $0xff, %ecx +# endif + + testl %ecx, %ecx + jne L(last_vector) + + addl $8, %edx + addl $8, %eax +# ifdef USE_AS_STRNCMP + /* Return 0 if the current offset (%rdx) >= the maximum offset + (%r11). */ + cmpq %r11, %rdx + jae L(zero) +# endif + +L(cross_page_8bytes): + /* Less than 8 bytes to check, try 4 byte vector. */ + cmpl $(PAGE_SIZE - 4), %eax + jg L(cross_page_4bytes) + vmovd (%rdi, %rdx), %XMM0 + vmovd (%rsi, %rdx), %XMM1 + + /* Each bit in K0 represents a mismatch in XMM0 and XMM1. */ + VPCMP $4, %XMM0, %XMM1, %k0 + VPCMP $0, %XMMZERO, %XMM0, %k1 + VPCMP $0, %XMMZERO, %XMM1, %k2 + /* Each bit in K1 represents a NULL in XMM0 or XMM1. */ + kord %k1, %k2, %k1 + /* Each bit in K1 represents a NULL or a mismatch. */ + kord %k0, %k1, %k1 + kmovd %k1, %ecx + +# ifdef USE_AS_WCSCMP + /* Only the last bit is valid. */ + andl $0x1, %ecx +# else + /* Only last 4 bits are valid. */ + andl $0xf, %ecx +# endif + + testl %ecx, %ecx + jne L(last_vector) + + addl $4, %edx +# ifdef USE_AS_STRNCMP + /* Return 0 if the current offset (%rdx) >= the maximum offset + (%r11). */ + cmpq %r11, %rdx + jae L(zero) +# endif + +L(cross_page_4bytes): +# endif + /* Less than 4 bytes to check, try one byte/dword at a time. */ +# ifdef USE_AS_STRNCMP + cmpq %r11, %rdx + jae L(zero) +# endif +# ifdef USE_AS_WCSCMP + movl (%rdi, %rdx), %eax + movl (%rsi, %rdx), %ecx +# else + movzbl (%rdi, %rdx), %eax + movzbl (%rsi, %rdx), %ecx +# endif + testl %eax, %eax + jne L(cross_page_loop) + subl %ecx, %eax + ret +END (STRCMP) +#endif diff --git a/sysdeps/x86_64/multiarch/strcmp.c b/sysdeps/x86_64/multiarch/strcmp.c index 6a9dca4389..1df75690d0 100644 --- a/sysdeps/x86_64/multiarch/strcmp.c +++ b/sysdeps/x86_64/multiarch/strcmp.c @@ -30,16 +30,25 @@ extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2_unaligned) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (ssse3) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (avx2) attribute_hidden; +extern __typeof (REDIRECT_NAME) OPTIMIZE (evex) attribute_hidden; static inline void * IFUNC_SELECTOR (void) { const struct cpu_features* cpu_features = __get_cpu_features (); - if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER) - && CPU_FEATURE_USABLE_P (cpu_features, AVX2) + if (CPU_FEATURE_USABLE_P (cpu_features, AVX2) && CPU_FEATURES_ARCH_P (cpu_features, AVX_Fast_Unaligned_Load)) - return OPTIMIZE (avx2); + { + if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL) + && CPU_FEATURE_USABLE_P (cpu_features, AVX512BW) + && CPU_FEATURE_USABLE_P (cpu_features, BMI2) + && !CPU_FEATURES_ARCH_P (cpu_features, Prefer_AVX2_STRCMP)) + return OPTIMIZE (evex); + + if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER)) + return OPTIMIZE (avx2); + } if (CPU_FEATURES_ARCH_P (cpu_features, Fast_Unaligned_Load)) return OPTIMIZE (sse2_unaligned); diff --git a/sysdeps/x86_64/multiarch/strlen-evex.S b/sysdeps/x86_64/multiarch/strlen-evex.S new file mode 100644 index 0000000000..cd022509cb --- /dev/null +++ b/sysdeps/x86_64/multiarch/strlen-evex.S @@ -0,0 +1,436 @@ +/* strlen/strnlen/wcslen/wcsnlen optimized with 256-bit EVEX instructions. + Copyright (C) 2021 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <https://www.gnu.org/licenses/>. */ + +#if IS_IN (libc) + +# include <sysdep.h> + +# ifndef STRLEN +# define STRLEN __strlen_evex +# endif + +# define VMOVA vmovdqa64 + +# ifdef USE_AS_WCSLEN +# define VPCMP vpcmpd +# define VPMINU vpminud +# define SHIFT_REG r9d +# else +# define VPCMP vpcmpb +# define VPMINU vpminub +# define SHIFT_REG ecx +# endif + +# define XMMZERO xmm16 +# define YMMZERO ymm16 +# define YMM1 ymm17 +# define YMM2 ymm18 +# define YMM3 ymm19 +# define YMM4 ymm20 +# define YMM5 ymm21 +# define YMM6 ymm22 + +# define VEC_SIZE 32 + + .section .text.evex,"ax",@progbits +ENTRY (STRLEN) +# ifdef USE_AS_STRNLEN + /* Check for zero length. */ + test %RSI_LP, %RSI_LP + jz L(zero) +# ifdef USE_AS_WCSLEN + shl $2, %RSI_LP +# elif defined __ILP32__ + /* Clear the upper 32 bits. */ + movl %esi, %esi +# endif + mov %RSI_LP, %R8_LP +# endif + movl %edi, %ecx + movq %rdi, %rdx + vpxorq %XMMZERO, %XMMZERO, %XMMZERO + + /* Check if we may cross page boundary with one vector load. */ + andl $(2 * VEC_SIZE - 1), %ecx + cmpl $VEC_SIZE, %ecx + ja L(cros_page_boundary) + + /* Check the first VEC_SIZE bytes. Each bit in K0 represents a + null byte. */ + VPCMP $0, (%rdi), %YMMZERO, %k0 + kmovd %k0, %eax + testl %eax, %eax + +# ifdef USE_AS_STRNLEN + jnz L(first_vec_x0_check) + /* Adjust length and check the end of data. */ + subq $VEC_SIZE, %rsi + jbe L(max) +# else + jnz L(first_vec_x0) +# endif + + /* Align data for aligned loads in the loop. */ + addq $VEC_SIZE, %rdi + andl $(VEC_SIZE - 1), %ecx + andq $-VEC_SIZE, %rdi + +# ifdef USE_AS_STRNLEN + /* Adjust length. */ + addq %rcx, %rsi + + subq $(VEC_SIZE * 4), %rsi + jbe L(last_4x_vec_or_less) +# endif + jmp L(more_4x_vec) + + .p2align 4 +L(cros_page_boundary): + andl $(VEC_SIZE - 1), %ecx + andq $-VEC_SIZE, %rdi + +# ifdef USE_AS_WCSLEN + /* NB: Divide shift count by 4 since each bit in K0 represent 4 + bytes. */ + movl %ecx, %SHIFT_REG + sarl $2, %SHIFT_REG +# endif + VPCMP $0, (%rdi), %YMMZERO, %k0 + kmovd %k0, %eax + + /* Remove the leading bytes. */ + sarxl %SHIFT_REG, %eax, %eax + testl %eax, %eax + jz L(aligned_more) + tzcntl %eax, %eax +# ifdef USE_AS_WCSLEN + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %eax +# endif +# ifdef USE_AS_STRNLEN + /* Check the end of data. */ + cmpq %rax, %rsi + jbe L(max) +# endif + addq %rdi, %rax + addq %rcx, %rax + subq %rdx, %rax +# ifdef USE_AS_WCSLEN + shrq $2, %rax +# endif + ret + + .p2align 4 +L(aligned_more): +# ifdef USE_AS_STRNLEN + /* "rcx" is less than VEC_SIZE. Calculate "rdx + rcx - VEC_SIZE" + with "rdx - (VEC_SIZE - rcx)" instead of "(rdx + rcx) - VEC_SIZE" + to void possible addition overflow. */ + negq %rcx + addq $VEC_SIZE, %rcx + + /* Check the end of data. */ + subq %rcx, %rsi + jbe L(max) +# endif + + addq $VEC_SIZE, %rdi + +# ifdef USE_AS_STRNLEN + subq $(VEC_SIZE * 4), %rsi + jbe L(last_4x_vec_or_less) +# endif + +L(more_4x_vec): + /* Check the first 4 * VEC_SIZE. Only one VEC_SIZE at a time + since data is only aligned to VEC_SIZE. */ + VPCMP $0, (%rdi), %YMMZERO, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x0) + + VPCMP $0, VEC_SIZE(%rdi), %YMMZERO, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x1) + + VPCMP $0, (VEC_SIZE * 2)(%rdi), %YMMZERO, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x2) + + VPCMP $0, (VEC_SIZE * 3)(%rdi), %YMMZERO, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x3) + + addq $(VEC_SIZE * 4), %rdi + +# ifdef USE_AS_STRNLEN + subq $(VEC_SIZE * 4), %rsi + jbe L(last_4x_vec_or_less) +# endif + + /* Align data to 4 * VEC_SIZE. */ + movq %rdi, %rcx + andl $(4 * VEC_SIZE - 1), %ecx + andq $-(4 * VEC_SIZE), %rdi + +# ifdef USE_AS_STRNLEN + /* Adjust length. */ + addq %rcx, %rsi +# endif + + .p2align 4 +L(loop_4x_vec): + /* Compare 4 * VEC at a time forward. */ + VMOVA (%rdi), %YMM1 + VMOVA VEC_SIZE(%rdi), %YMM2 + VMOVA (VEC_SIZE * 2)(%rdi), %YMM3 + VMOVA (VEC_SIZE * 3)(%rdi), %YMM4 + + VPMINU %YMM1, %YMM2, %YMM5 + VPMINU %YMM3, %YMM4, %YMM6 + + VPMINU %YMM5, %YMM6, %YMM5 + VPCMP $0, %YMM5, %YMMZERO, %k0 + ktestd %k0, %k0 + jnz L(4x_vec_end) + + addq $(VEC_SIZE * 4), %rdi + +# ifndef USE_AS_STRNLEN + jmp L(loop_4x_vec) +# else + subq $(VEC_SIZE * 4), %rsi + ja L(loop_4x_vec) + +L(last_4x_vec_or_less): + /* Less than 4 * VEC and aligned to VEC_SIZE. */ + addl $(VEC_SIZE * 2), %esi + jle L(last_2x_vec) + + VPCMP $0, (%rdi), %YMMZERO, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x0) + + VPCMP $0, VEC_SIZE(%rdi), %YMMZERO, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x1) + + VPCMP $0, (VEC_SIZE * 2)(%rdi), %YMMZERO, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x2_check) + subl $VEC_SIZE, %esi + jle L(max) + + VPCMP $0, (VEC_SIZE * 3)(%rdi), %YMMZERO, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x3_check) + movq %r8, %rax +# ifdef USE_AS_WCSLEN + shrq $2, %rax +# endif + ret + + .p2align 4 +L(last_2x_vec): + addl $(VEC_SIZE * 2), %esi + + VPCMP $0, (%rdi), %YMMZERO, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x0_check) + subl $VEC_SIZE, %esi + jle L(max) + + VPCMP $0, VEC_SIZE(%rdi), %YMMZERO, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x1_check) + movq %r8, %rax +# ifdef USE_AS_WCSLEN + shrq $2, %rax +# endif + ret + + .p2align 4 +L(first_vec_x0_check): + tzcntl %eax, %eax +# ifdef USE_AS_WCSLEN + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %eax +# endif + /* Check the end of data. */ + cmpq %rax, %rsi + jbe L(max) + addq %rdi, %rax + subq %rdx, %rax +# ifdef USE_AS_WCSLEN + shrq $2, %rax +# endif + ret + + .p2align 4 +L(first_vec_x1_check): + tzcntl %eax, %eax +# ifdef USE_AS_WCSLEN + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %eax +# endif + /* Check the end of data. */ + cmpq %rax, %rsi + jbe L(max) + addq $VEC_SIZE, %rax + addq %rdi, %rax + subq %rdx, %rax +# ifdef USE_AS_WCSLEN + shrq $2, %rax +# endif + ret + + .p2align 4 +L(first_vec_x2_check): + tzcntl %eax, %eax +# ifdef USE_AS_WCSLEN + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %eax +# endif + /* Check the end of data. */ + cmpq %rax, %rsi + jbe L(max) + addq $(VEC_SIZE * 2), %rax + addq %rdi, %rax + subq %rdx, %rax +# ifdef USE_AS_WCSLEN + shrq $2, %rax +# endif + ret + + .p2align 4 +L(first_vec_x3_check): + tzcntl %eax, %eax +# ifdef USE_AS_WCSLEN + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %eax +# endif + /* Check the end of data. */ + cmpq %rax, %rsi + jbe L(max) + addq $(VEC_SIZE * 3), %rax + addq %rdi, %rax + subq %rdx, %rax +# ifdef USE_AS_WCSLEN + shrq $2, %rax +# endif + ret + + .p2align 4 +L(max): + movq %r8, %rax +# ifdef USE_AS_WCSLEN + shrq $2, %rax +# endif + ret + + .p2align 4 +L(zero): + xorl %eax, %eax + ret +# endif + + .p2align 4 +L(first_vec_x0): + tzcntl %eax, %eax +# ifdef USE_AS_WCSLEN + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %eax +# endif + addq %rdi, %rax + subq %rdx, %rax +# ifdef USE_AS_WCSLEN + shrq $2, %rax +# endif + ret + + .p2align 4 +L(first_vec_x1): + tzcntl %eax, %eax +# ifdef USE_AS_WCSLEN + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %eax +# endif + addq $VEC_SIZE, %rax + addq %rdi, %rax + subq %rdx, %rax +# ifdef USE_AS_WCSLEN + shrq $2, %rax +# endif + ret + + .p2align 4 +L(first_vec_x2): + tzcntl %eax, %eax +# ifdef USE_AS_WCSLEN + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %eax +# endif + addq $(VEC_SIZE * 2), %rax + addq %rdi, %rax + subq %rdx, %rax +# ifdef USE_AS_WCSLEN + shrq $2, %rax +# endif + ret + + .p2align 4 +L(4x_vec_end): + VPCMP $0, %YMM1, %YMMZERO, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x0) + VPCMP $0, %YMM2, %YMMZERO, %k1 + kmovd %k1, %eax + testl %eax, %eax + jnz L(first_vec_x1) + VPCMP $0, %YMM3, %YMMZERO, %k2 + kmovd %k2, %eax + testl %eax, %eax + jnz L(first_vec_x2) + VPCMP $0, %YMM4, %YMMZERO, %k3 + kmovd %k3, %eax +L(first_vec_x3): + tzcntl %eax, %eax +# ifdef USE_AS_WCSLEN + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + sall $2, %eax +# endif + addq $(VEC_SIZE * 3), %rax + addq %rdi, %rax + subq %rdx, %rax +# ifdef USE_AS_WCSLEN + shrq $2, %rax +# endif + ret + +END (STRLEN) +#endif diff --git a/sysdeps/x86_64/multiarch/strncmp-evex.S b/sysdeps/x86_64/multiarch/strncmp-evex.S new file mode 100644 index 0000000000..a1d53e8c9f --- /dev/null +++ b/sysdeps/x86_64/multiarch/strncmp-evex.S @@ -0,0 +1,3 @@ +#define STRCMP __strncmp_evex +#define USE_AS_STRNCMP 1 +#include "strcmp-evex.S" diff --git a/sysdeps/x86_64/multiarch/strncmp.c b/sysdeps/x86_64/multiarch/strncmp.c index a565626aed..617c4e1065 100644 --- a/sysdeps/x86_64/multiarch/strncmp.c +++ b/sysdeps/x86_64/multiarch/strncmp.c @@ -30,16 +30,25 @@ extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (ssse3) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (sse42) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (avx2) attribute_hidden; +extern __typeof (REDIRECT_NAME) OPTIMIZE (evex) attribute_hidden; static inline void * IFUNC_SELECTOR (void) { const struct cpu_features* cpu_features = __get_cpu_features (); - if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER) - && CPU_FEATURE_USABLE_P (cpu_features, AVX2) + if (CPU_FEATURE_USABLE_P (cpu_features, AVX2) && CPU_FEATURES_ARCH_P (cpu_features, AVX_Fast_Unaligned_Load)) - return OPTIMIZE (avx2); + { + if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL) + && CPU_FEATURE_USABLE_P (cpu_features, AVX512BW) + && CPU_FEATURE_USABLE_P (cpu_features, BMI2) + && !CPU_FEATURES_ARCH_P (cpu_features, Prefer_AVX2_STRCMP)) + return OPTIMIZE (evex); + + if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER)) + return OPTIMIZE (avx2); + } if (CPU_FEATURE_USABLE_P (cpu_features, SSE4_2) && !CPU_FEATURES_ARCH_P (cpu_features, Slow_SSE4_2)) diff --git a/sysdeps/x86_64/multiarch/strnlen-evex.S b/sysdeps/x86_64/multiarch/strnlen-evex.S new file mode 100644 index 0000000000..722022f303 --- /dev/null +++ b/sysdeps/x86_64/multiarch/strnlen-evex.S @@ -0,0 +1,4 @@ +#define STRLEN __strnlen_evex +#define USE_AS_STRNLEN 1 + +#include "strlen-evex.S" diff --git a/sysdeps/x86_64/multiarch/strrchr-evex.S b/sysdeps/x86_64/multiarch/strrchr-evex.S new file mode 100644 index 0000000000..f920b5a584 --- /dev/null +++ b/sysdeps/x86_64/multiarch/strrchr-evex.S @@ -0,0 +1,265 @@ +/* strrchr/wcsrchr optimized with 256-bit EVEX instructions. + Copyright (C) 2021 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <https://www.gnu.org/licenses/>. */ + +#if IS_IN (libc) + +# include <sysdep.h> + +# ifndef STRRCHR +# define STRRCHR __strrchr_evex +# endif + +# define VMOVU vmovdqu64 +# define VMOVA vmovdqa64 + +# ifdef USE_AS_WCSRCHR +# define VPBROADCAST vpbroadcastd +# define VPCMP vpcmpd +# define SHIFT_REG r8d +# else +# define VPBROADCAST vpbroadcastb +# define VPCMP vpcmpb +# define SHIFT_REG ecx +# endif + +# define XMMZERO xmm16 +# define YMMZERO ymm16 +# define YMMMATCH ymm17 +# define YMM1 ymm18 + +# define VEC_SIZE 32 + + .section .text.evex,"ax",@progbits +ENTRY (STRRCHR) + movl %edi, %ecx + /* Broadcast CHAR to YMMMATCH. */ + VPBROADCAST %esi, %YMMMATCH + + vpxorq %XMMZERO, %XMMZERO, %XMMZERO + + /* Check if we may cross page boundary with one vector load. */ + andl $(2 * VEC_SIZE - 1), %ecx + cmpl $VEC_SIZE, %ecx + ja L(cros_page_boundary) + + VMOVU (%rdi), %YMM1 + + /* Each bit in K0 represents a null byte in YMM1. */ + VPCMP $0, %YMMZERO, %YMM1, %k0 + /* Each bit in K1 represents a CHAR in YMM1. */ + VPCMP $0, %YMMMATCH, %YMM1, %k1 + kmovd %k0, %ecx + kmovd %k1, %eax + + addq $VEC_SIZE, %rdi + + testl %eax, %eax + jnz L(first_vec) + + testl %ecx, %ecx + jnz L(return_null) + + andq $-VEC_SIZE, %rdi + xorl %edx, %edx + jmp L(aligned_loop) + + .p2align 4 +L(first_vec): + /* Check if there is a null byte. */ + testl %ecx, %ecx + jnz L(char_and_nul_in_first_vec) + + /* Remember the match and keep searching. */ + movl %eax, %edx + movq %rdi, %rsi + andq $-VEC_SIZE, %rdi + jmp L(aligned_loop) + + .p2align 4 +L(cros_page_boundary): + andl $(VEC_SIZE - 1), %ecx + andq $-VEC_SIZE, %rdi + +# ifdef USE_AS_WCSRCHR + /* NB: Divide shift count by 4 since each bit in K1 represent 4 + bytes. */ + movl %ecx, %SHIFT_REG + sarl $2, %SHIFT_REG +# endif + + VMOVA (%rdi), %YMM1 + + /* Each bit in K0 represents a null byte in YMM1. */ + VPCMP $0, %YMMZERO, %YMM1, %k0 + /* Each bit in K1 represents a CHAR in YMM1. */ + VPCMP $0, %YMMMATCH, %YMM1, %k1 + kmovd %k0, %edx + kmovd %k1, %eax + + shrxl %SHIFT_REG, %edx, %edx + shrxl %SHIFT_REG, %eax, %eax + addq $VEC_SIZE, %rdi + + /* Check if there is a CHAR. */ + testl %eax, %eax + jnz L(found_char) + + testl %edx, %edx + jnz L(return_null) + + jmp L(aligned_loop) + + .p2align 4 +L(found_char): + testl %edx, %edx + jnz L(char_and_nul) + + /* Remember the match and keep searching. */ + movl %eax, %edx + leaq (%rdi, %rcx), %rsi + + .p2align 4 +L(aligned_loop): + VMOVA (%rdi), %YMM1 + addq $VEC_SIZE, %rdi + + /* Each bit in K0 represents a null byte in YMM1. */ + VPCMP $0, %YMMZERO, %YMM1, %k0 + /* Each bit in K1 represents a CHAR in YMM1. */ + VPCMP $0, %YMMMATCH, %YMM1, %k1 + kmovd %k0, %ecx + kmovd %k1, %eax + orl %eax, %ecx + jnz L(char_nor_null) + + VMOVA (%rdi), %YMM1 + add $VEC_SIZE, %rdi + + /* Each bit in K0 represents a null byte in YMM1. */ + VPCMP $0, %YMMZERO, %YMM1, %k0 + /* Each bit in K1 represents a CHAR in YMM1. */ + VPCMP $0, %YMMMATCH, %YMM1, %k1 + kmovd %k0, %ecx + kmovd %k1, %eax + orl %eax, %ecx + jnz L(char_nor_null) + + VMOVA (%rdi), %YMM1 + addq $VEC_SIZE, %rdi + + /* Each bit in K0 represents a null byte in YMM1. */ + VPCMP $0, %YMMZERO, %YMM1, %k0 + /* Each bit in K1 represents a CHAR in YMM1. */ + VPCMP $0, %YMMMATCH, %YMM1, %k1 + kmovd %k0, %ecx + kmovd %k1, %eax + orl %eax, %ecx + jnz L(char_nor_null) + + VMOVA (%rdi), %YMM1 + addq $VEC_SIZE, %rdi + + /* Each bit in K0 represents a null byte in YMM1. */ + VPCMP $0, %YMMZERO, %YMM1, %k0 + /* Each bit in K1 represents a CHAR in YMM1. */ + VPCMP $0, %YMMMATCH, %YMM1, %k1 + kmovd %k0, %ecx + kmovd %k1, %eax + orl %eax, %ecx + jz L(aligned_loop) + + .p2align 4 +L(char_nor_null): + /* Find a CHAR or a null byte in a loop. */ + testl %eax, %eax + jnz L(match) +L(return_value): + testl %edx, %edx + jz L(return_null) + movl %edx, %eax + movq %rsi, %rdi + bsrl %eax, %eax +# ifdef USE_AS_WCSRCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + leaq -VEC_SIZE(%rdi, %rax, 4), %rax +# else + leaq -VEC_SIZE(%rdi, %rax), %rax +# endif + ret + + .p2align 4 +L(match): + /* Find a CHAR. Check if there is a null byte. */ + kmovd %k0, %ecx + testl %ecx, %ecx + jnz L(find_nul) + + /* Remember the match and keep searching. */ + movl %eax, %edx + movq %rdi, %rsi + jmp L(aligned_loop) + + .p2align 4 +L(find_nul): + /* Mask out any matching bits after the null byte. */ + movl %ecx, %r8d + subl $1, %r8d + xorl %ecx, %r8d + andl %r8d, %eax + testl %eax, %eax + /* If there is no CHAR here, return the remembered one. */ + jz L(return_value) + bsrl %eax, %eax +# ifdef USE_AS_WCSRCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + leaq -VEC_SIZE(%rdi, %rax, 4), %rax +# else + leaq -VEC_SIZE(%rdi, %rax), %rax +# endif + ret + + .p2align 4 +L(char_and_nul): + /* Find both a CHAR and a null byte. */ + addq %rcx, %rdi + movl %edx, %ecx +L(char_and_nul_in_first_vec): + /* Mask out any matching bits after the null byte. */ + movl %ecx, %r8d + subl $1, %r8d + xorl %ecx, %r8d + andl %r8d, %eax + testl %eax, %eax + /* Return null pointer if the null byte comes first. */ + jz L(return_null) + bsrl %eax, %eax +# ifdef USE_AS_WCSRCHR + /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ + leaq -VEC_SIZE(%rdi, %rax, 4), %rax +# else + leaq -VEC_SIZE(%rdi, %rax), %rax +# endif + ret + + .p2align 4 +L(return_null): + xorl %eax, %eax + ret + +END (STRRCHR) +#endif diff --git a/sysdeps/x86_64/multiarch/wcschr-evex.S b/sysdeps/x86_64/multiarch/wcschr-evex.S new file mode 100644 index 0000000000..7cb8f1e41a --- /dev/null +++ b/sysdeps/x86_64/multiarch/wcschr-evex.S @@ -0,0 +1,3 @@ +#define STRCHR __wcschr_evex +#define USE_AS_WCSCHR 1 +#include "strchr-evex.S" diff --git a/sysdeps/x86_64/multiarch/wcscmp-evex.S b/sysdeps/x86_64/multiarch/wcscmp-evex.S new file mode 100644 index 0000000000..42e73e51eb --- /dev/null +++ b/sysdeps/x86_64/multiarch/wcscmp-evex.S @@ -0,0 +1,4 @@ +#define STRCMP __wcscmp_evex +#define USE_AS_WCSCMP 1 + +#include "strcmp-evex.S" diff --git a/sysdeps/x86_64/multiarch/wcslen-evex.S b/sysdeps/x86_64/multiarch/wcslen-evex.S new file mode 100644 index 0000000000..bdafa83bd5 --- /dev/null +++ b/sysdeps/x86_64/multiarch/wcslen-evex.S @@ -0,0 +1,4 @@ +#define STRLEN __wcslen_evex +#define USE_AS_WCSLEN 1 + +#include "strlen-evex.S" diff --git a/sysdeps/x86_64/multiarch/wcsncmp-evex.S b/sysdeps/x86_64/multiarch/wcsncmp-evex.S new file mode 100644 index 0000000000..8a8e310713 --- /dev/null +++ b/sysdeps/x86_64/multiarch/wcsncmp-evex.S @@ -0,0 +1,5 @@ +#define STRCMP __wcsncmp_evex +#define USE_AS_STRNCMP 1 +#define USE_AS_WCSCMP 1 + +#include "strcmp-evex.S" diff --git a/sysdeps/x86_64/multiarch/wcsnlen-evex.S b/sysdeps/x86_64/multiarch/wcsnlen-evex.S new file mode 100644 index 0000000000..24773bb4e2 --- /dev/null +++ b/sysdeps/x86_64/multiarch/wcsnlen-evex.S @@ -0,0 +1,5 @@ +#define STRLEN __wcsnlen_evex +#define USE_AS_WCSLEN 1 +#define USE_AS_STRNLEN 1 + +#include "strlen-evex.S" diff --git a/sysdeps/x86_64/multiarch/wcsnlen.c b/sysdeps/x86_64/multiarch/wcsnlen.c index 81b1a221ff..19bc6fd938 100644 --- a/sysdeps/x86_64/multiarch/wcsnlen.c +++ b/sysdeps/x86_64/multiarch/wcsnlen.c @@ -29,16 +29,24 @@ extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (sse4_1) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (avx2) attribute_hidden; +extern __typeof (REDIRECT_NAME) OPTIMIZE (evex) attribute_hidden; static inline void * IFUNC_SELECTOR (void) { const struct cpu_features* cpu_features = __get_cpu_features (); - if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER) - && CPU_FEATURE_USABLE_P (cpu_features, AVX2) + if (CPU_FEATURE_USABLE_P (cpu_features, AVX2) && CPU_FEATURES_ARCH_P (cpu_features, AVX_Fast_Unaligned_Load)) - return OPTIMIZE (avx2); + { + if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL) + && CPU_FEATURE_USABLE_P (cpu_features, AVX512BW) + && CPU_FEATURE_USABLE_P (cpu_features, BMI2)) + return OPTIMIZE (evex); + + if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER)) + return OPTIMIZE (avx2); + } if (CPU_FEATURE_USABLE_P (cpu_features, SSE4_1)) return OPTIMIZE (sse4_1); diff --git a/sysdeps/x86_64/multiarch/wcsrchr-evex.S b/sysdeps/x86_64/multiarch/wcsrchr-evex.S new file mode 100644 index 0000000000..c64602f7dc --- /dev/null +++ b/sysdeps/x86_64/multiarch/wcsrchr-evex.S @@ -0,0 +1,3 @@ +#define STRRCHR __wcsrchr_evex +#define USE_AS_WCSRCHR 1 +#include "strrchr-evex.S" diff --git a/sysdeps/x86_64/multiarch/wmemchr-evex.S b/sysdeps/x86_64/multiarch/wmemchr-evex.S new file mode 100644 index 0000000000..06cd0f9f5a --- /dev/null +++ b/sysdeps/x86_64/multiarch/wmemchr-evex.S @@ -0,0 +1,4 @@ +#define MEMCHR __wmemchr_evex +#define USE_AS_WMEMCHR 1 + +#include "memchr-evex.S" |