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

Enable ADLER32 using NEON in 32-bit Windows on ARM.

Browse Source
This commit is contained in:
Natalia Portillo
2023-09-23 01:34:29 +01:00
parent e18fe4d973
commit 1bd06bb761
1 changed files with 46 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
adler32_neon.c
View File

@@ -84,7 +84,8 @@ TARGET_WITH_SIMD void adler32_neon(uint16_t* sum1, uint16_t* sum2, const uint8_t
uint16x8_t v_column_sum_2 = vdupq_n_u16(0); uint16x8_t v_column_sum_2 = vdupq_n_u16(0);
uint16x8_t v_column_sum_3 = vdupq_n_u16(0); uint16x8_t v_column_sum_3 = vdupq_n_u16(0);
uint16x8_t v_column_sum_4 = vdupq_n_u16(0); uint16x8_t v_column_sum_4 = vdupq_n_u16(0);
do { do
{
/* /*
* Load 32 input bytes. * Load 32 input bytes.
*/ */
@@ -112,6 +113,7 @@ TARGET_WITH_SIMD void adler32_neon(uint16_t* sum1, uint16_t* sum2, const uint8_t
* Multiply-add bytes by [ 32, 31, 30, ... ] for s2. * Multiply-add bytes by [ 32, 31, 30, ... ] for s2.
*/ */
#ifdef _MSC_VER #ifdef _MSC_VER
#ifdef _M_ARM64
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_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_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_low_u16(v_column_sum_2), neon_ld1m_16((uint16_t[]) {24, 23, 22, 21}));
@@ -120,6 +122,16 @@ TARGET_WITH_SIMD void adler32_neon(uint16_t* sum1, uint16_t* sum2, const uint8_t
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_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_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})); 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), vld1_u16(((uint16_t[]) {32, 31, 30, 29})));
v_s2 = vmlal_u16(v_s2, vget_high_u16(v_column_sum_1), vld1_u16(((uint16_t[]) {28, 27, 26, 25})));
v_s2 = vmlal_u16(v_s2, vget_low_u16(v_column_sum_2), vld1_u16(((uint16_t[]) {24, 23, 22, 21})));
v_s2 = vmlal_u16(v_s2, vget_high_u16(v_column_sum_2), vld1_u16(((uint16_t[]) {20, 19, 18, 17})));
v_s2 = vmlal_u16(v_s2, vget_low_u16(v_column_sum_3), vld1_u16(((uint16_t[]) {16, 15, 14, 13})));
v_s2 = vmlal_u16(v_s2, vget_high_u16(v_column_sum_3), vld1_u16(((uint16_t[]) {12, 11, 10, 9})));
v_s2 = vmlal_u16(v_s2, vget_low_u16(v_column_sum_4), vld1_u16(((uint16_t[]) {8, 7, 6, 5})));
v_s2 = vmlal_u16(v_s2, vget_high_u16(v_column_sum_4), vld1_u16(((uint16_t[]) {4, 3, 2, 1})));
#endif
#else #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_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_high_u16(v_column_sum_1), (uint16x4_t) {28, 27, 26, 25});
@@ -169,7 +181,8 @@ TARGET_WITH_SIMD void adler32_neon(uint16_t* sum1, uint16_t* sum2, const uint8_t
s2 += (s1 += *data++); s2 += (s1 += *data++);
len -= 16; len -= 16;
} }
while(len--) { s2 += (s1 += *data++); } while (len--)
{ s2 += (s1 += *data++); }
if (s1 >= ADLER_MODULE) s1 -= ADLER_MODULE; if (s1 >= ADLER_MODULE) s1 -= ADLER_MODULE;
s2 %= ADLER_MODULE; s2 %= ADLER_MODULE;
} }