xgiovio/MultiPar
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Use 64-bit integer to XOR data

Browse Source
This commit is contained in:
Yutaka Sawada
2023-06-01 10:49:40 +09:00
committed by GitHub
parent adebd70424
commit e830eac1d6
1 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
source/par2j/lib_opencl.c
View File

@@ -1,5 +1,5 @@
// lib_opencl.c // lib_opencl.c
// Copyright : 2023-05-29 Yutaka Sawada // Copyright : 2023-06-01 Yutaka Sawada
// License : GPL // License : GPL
#ifndef _WIN32_WINNT #ifndef _WIN32_WINNT
@@ -774,7 +774,7 @@ int gpu_multiply_blocks(
unsigned char *buf, // Products go here unsigned char *buf, // Products go here
int len) // Byte length int len) // Byte length
{ {
unsigned int *vram, *src, *dst; unsigned __int64 *vram, *src, *dst;
size_t global_size, local_size; size_t global_size, local_size;
cl_int ret; cl_int ret;
@@ -801,14 +801,14 @@ int gpu_multiply_blocks(
if (ret != CL_SUCCESS) if (ret != CL_SUCCESS)
return (ret << 8) | 12; return (ret << 8) | 12;
// 4バイトごとに XOR する (SSE2 で XOR しても速くならず) // 8バイトごとに XOR する (SSE2 で XOR しても速くならず)
src = vram; src = vram;
dst = (unsigned int *)buf; dst = (unsigned __int64 *)buf;
while (len > 0){ while (len > 0){
*dst ^= *src; *dst ^= *src;
dst++; dst++;
src++; src++;
len -= 4; len -= 8;
} }
// ホスト側でデータを変更しなくても、clEnqueueMapBufferと対で呼び出さないといけない // ホスト側でデータを変更しなくても、clEnqueueMapBufferと対で呼び出さないといけない