aaru-dps/Aaru.Checksums.Native
1
0
Fork 0
mirror of https://github.com/aaru-dps/Aaru.Checksums.Native.git synced 2025-12-16 19:24:29 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix compilation on ARM and ARM64 using MSVC.

Browse Source
This commit is contained in:
Natalia Portillo
2021-10-13 00:41:58 +01:00
parent 6c10f3e58d
commit 8d77838be2
9 changed files with 84 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
adler32_neon.c
View File

@@ -2,7 +2,7 @@
// Created by claunia on 28/9/21.
//
#if defined(__aarch64__) || defined(_M_ARM64) || defined(__arm__) || defined(_M_ARM)
#if defined(__aarch64__) || defined(_M_ARM64) || ((defined(__arm__) || defined(_M_ARM)) && !defined(_WIN32))
#include <arm_neon.h>
@@ -45,8 +45,13 @@ TARGET_WITH_SIMD void adler32_neon(uint16_t* sum1, uint16_t* sum2, const unsigne
* Process n blocks of data. At most NMAX data bytes can be
* processed before s2 must be reduced modulo ADLER_MODULE.
*/
#ifdef _WIN32
uint32x4_t v_s2 = {.n128_u32 = {0, 0, 0, s1 * n}};
uint32x4_t v_s1 = {.n128_u32 = {0, 0, 0, 0}};
#else
uint32x4_t v_s2 = (uint32x4_t){0, 0, 0, s1 * n};
uint32x4_t v_s1 = (uint32x4_t){0, 0, 0, 0};
#endif
uint16x8_t v_column_sum_1 = vdupq_n_u16(0);
uint16x8_t v_column_sum_2 = vdupq_n_u16(0);
uint16x8_t v_column_sum_3 = vdupq_n_u16(0);
@@ -78,6 +83,16 @@ TARGET_WITH_SIMD void adler32_neon(uint16_t* sum1, uint16_t* sum2, const unsigne
/*
* Multiply-add bytes by [ 32, 31, 30, ... ] for s2.
*/
#ifdef _WIN32
v_s2 = vmlal_u16(v_s2, vget_low_u16(v_column_sum_1), neon_ld1m_16((uint16_t[]) {32, 31, 30, 29}));
v_s2 = vmlal_u16(v_s2, vget_high_u16(v_column_sum_1), neon_ld1m_16((uint16_t[]){28, 27, 26, 25}));
v_s2 = vmlal_u16(v_s2, vget_low_u16(v_column_sum_2), neon_ld1m_16((uint16_t[]) {24, 23, 22, 21}));
v_s2 = vmlal_u16(v_s2, vget_high_u16(v_column_sum_2), neon_ld1m_16((uint16_t[]) {20, 19, 18, 17}));
v_s2 = vmlal_u16(v_s2, vget_low_u16(v_column_sum_3), neon_ld1m_16((uint16_t[]) {16, 15, 14, 13}));
v_s2 = vmlal_u16(v_s2, vget_high_u16(v_column_sum_3), neon_ld1m_16((uint16_t[]) {12, 11, 10, 9}));
v_s2 = vmlal_u16(v_s2, vget_low_u16(v_column_sum_4), neon_ld1m_16((uint16_t[]) {8, 7, 6, 5}));
v_s2 = vmlal_u16(v_s2, vget_high_u16(v_column_sum_4), neon_ld1m_16((uint16_t[]) {4, 3, 2, 1}));
#else
v_s2 = vmlal_u16(v_s2, vget_low_u16(v_column_sum_1), (uint16x4_t){32, 31, 30, 29});
v_s2 = vmlal_u16(v_s2, vget_high_u16(v_column_sum_1), (uint16x4_t){28, 27, 26, 25});
v_s2 = vmlal_u16(v_s2, vget_low_u16(v_column_sum_2), (uint16x4_t){24, 23, 22, 21});
@@ -86,6 +101,7 @@ TARGET_WITH_SIMD void adler32_neon(uint16_t* sum1, uint16_t* sum2, const unsigne
v_s2 = vmlal_u16(v_s2, vget_high_u16(v_column_sum_3), (uint16x4_t){12, 11, 10, 9});
v_s2 = vmlal_u16(v_s2, vget_low_u16(v_column_sum_4), (uint16x4_t){8, 7, 6, 5});
v_s2 = vmlal_u16(v_s2, vget_high_u16(v_column_sum_4), (uint16x4_t){4, 3, 2, 1});
#endif
/*
* Sum epi32 ints v_s1(s2) and accumulate in s1(s2).
*/