popov/cpuminer-opt-gpu
1
0
Fork 0
mirror of https://github.com/JayDDee/cpuminer-opt.git synced 2025-09-17 23:44:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

v24.4

Browse Source
This commit is contained in:
Jay D Dee
2024-07-01 00:33:19 -04:00
parent c47c4a8885
commit 47e24b50e8
23 changed files with 2097 additions and 2855 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
algo/x17/x17-4way.c
View File

@@ -418,11 +418,11 @@ int scanhash_x17_16x32( struct work *work, uint32_t max_nonce,
const bool bench = opt_benchmark;
// convert LE32 to LE64
edata[0] = v128_swap64_32( casti_m128i( pdata, 0 ) );
edata[1] = v128_swap64_32( casti_m128i( pdata, 1 ) );
edata[2] = v128_swap64_32( casti_m128i( pdata, 2 ) );
edata[3] = v128_swap64_32( casti_m128i( pdata, 3 ) );
edata[4] = v128_swap64_32( casti_m128i( pdata, 4 ) );
edata[0] = v128_swap64_32( casti_v128u32( pdata, 0 ) );
edata[1] = v128_swap64_32( casti_v128u32( pdata, 1 ) );
edata[2] = v128_swap64_32( casti_v128u32( pdata, 2 ) );
edata[3] = v128_swap64_32( casti_v128u32( pdata, 3 ) );
edata[4] = v128_swap64_32( casti_v128u32( pdata, 4 ) );
mm512_intrlv80_8x64( vdata, edata );
blake512_8way_prehash_le( &blake512_8way_ctx, x17_16way_midstate, vdata );
@@ -681,11 +681,11 @@ int scanhash_x17_8x64( struct work *work, uint32_t max_nonce,
const bool bench = opt_benchmark;
// convert LE32 to LE64
edata[0] = v128_swap64_32( casti_m128i( pdata, 0 ) );
edata[1] = v128_swap64_32( casti_m128i( pdata, 1 ) );
edata[2] = v128_swap64_32( casti_m128i( pdata, 2 ) );
edata[3] = v128_swap64_32( casti_m128i( pdata, 3 ) );
edata[4] = v128_swap64_32( casti_m128i( pdata, 4 ) );
edata[0] = v128_swap64_32( casti_v128u32( pdata, 0 ) );
edata[1] = v128_swap64_32( casti_v128u32( pdata, 1 ) );
edata[2] = v128_swap64_32( casti_v128u32( pdata, 2 ) );
edata[3] = v128_swap64_32( casti_v128u32( pdata, 3 ) );
edata[4] = v128_swap64_32( casti_v128u32( pdata, 4 ) );
mm512_intrlv80_8x64( vdata, edata );
*noncev = _mm512_add_epi32( *noncev, _mm512_set_epi32(
@@ -895,11 +895,11 @@ int scanhash_x17_4x64( struct work *work, uint32_t max_nonce,
const bool bench = opt_benchmark;
// convert LE32 to LE64
edata[0] = v128_swap64_32( casti_m128i( pdata, 0 ) );
edata[1] = v128_swap64_32( casti_m128i( pdata, 1 ) );
edata[2] = v128_swap64_32( casti_m128i( pdata, 2 ) );
edata[3] = v128_swap64_32( casti_m128i( pdata, 3 ) );
edata[4] = v128_swap64_32( casti_m128i( pdata, 4 ) );
edata[0] = v128_swap64_32( casti_v128u32( pdata, 0 ) );
edata[1] = v128_swap64_32( casti_v128u32( pdata, 1 ) );
edata[2] = v128_swap64_32( casti_v128u32( pdata, 2 ) );
edata[3] = v128_swap64_32( casti_v128u32( pdata, 3 ) );
edata[4] = v128_swap64_32( casti_v128u32( pdata, 4 ) );
mm256_intrlv80_4x64( vdata, edata );
*noncev = _mm256_add_epi32( *noncev, _mm256_set_epi32( 0,3,0,2, 0,1,0,0 ) );