// Copyright (C) 2003 Dolphin Project.
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, version 2.0.
// This program 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 General Public License 2.0 for more details.
// A copy of the GPL 2.0 should have been included with the program.
// If not, see http://www.gnu.org/licenses/
// Official SVN repository and contact information can be found at
// http://code.google.com/p/dolphin-emu/
#if _WIN32
#include <intrin.h>
#endif
#include "OpenCL.h"
#include <xmmintrin.h>
#include "XFBConvert.h"
#include "Common.h"
namespace {
const __m128i _bias1 = _mm_set_epi32(128/2 << 16, 0, 128/2 << 16, 16 << 16);
const __m128i _bias2 = _mm_set_epi32(128/2 << 16, 16 << 16, 128/2 << 16, 0);
__m128i _y[256];
__m128i _u[256];
__m128i _v[256];
__m128i _r1[256];
__m128i _r2[256];
__m128i _g1[256];
__m128i _g2[256];
__m128i _b1[256];
__m128i _b2[256];
} // namespace
#if defined(HAVE_OPENCL) && HAVE_OPENCL
bool Inited = false;
cl_kernel To_kernel;
cl_program To_program;
cl_kernel From_kernel;
cl_program From_program;
const char *__ConvertFromXFB = "int bound(int i) \n \
{ \n \
return (i>255)?255:((i<0)?0:i); \n \
} \n \
\n \
void yuv2rgb(int y, int u, int v, int &r, int &g, int &b) \n \
{ \n \
b = bound((76283*(y - 16) + 132252*(u - 128))>>16); \n \
g = bound((76283*(y - 16) - 53281 *(v - 128) - 25624*(u - 128))>>16); //last one u? \n \
r = bound((76283*(y - 16) + 104595*(v - 128))>>16); \n \
} \n \
\n \
void ConvertFromXFB(u32 *dst, const u8* _pXFB) \n \
{ \n \
const unsigned char *src = _pXFB; \n \
int id = get_global_id(0); \n \
int srcOffset = id * 4; \n \
int dstOffset = id; \n \
u32 numBlocks = (width * height) / 2; \n \
\n \
int Y1 = src[srcOffset]; \n \
int U = src[srcOffset + 1]; \n \
int Y2 = src[srcOffset + 2]; \n \
int V = src[srcOffset + 3]; \n \
\n \
int r, g, b; \n \
yuv2rgb(Y1,U,V, r,g,b); \n \
dst[dstOffset] = 0xFF000000 | (r<<16) | (g<<8) | (b); \n \
yuv2rgb(Y2,U,V, r,g,b); \n \
dst[dstOffset + 1] = 0xFF000000 | (r<<16) | (g<<8) | (b); \n \
} \n";
const char *__ConvertToXFB = "__kernel void ConvertToXFB(__global unsigned int *dst, __global const unsigned char* _pEFB) \n \
{ \n \
const unsigned char *src = _pEFB;\n \
int id = get_global_id(0);\n \
int srcOffset = id * 8; \n \
\n \
int y1 = (((16843 * src[srcOffset]) + (33030 * src[srcOffset + 1]) + (6423 * src[srcOffset + 2])) >> 16) + 16; \n \
int u1 = ((-(9699 * src[srcOffset]) - (19071 * src[srcOffset + 1]) + (28770 * src[srcOffset + 2])) >> 16) + 128;\n \
srcOffset += 4;\n \
\n \
int y2 = (((16843 * src[srcOffset]) + (33030 * src[srcOffset + 1]) + (6423 * src[srcOffset + 2])) >> 16) + 16;\n \
int v2 = (((28770 * src[srcOffset]) - (24117 * src[srcOffset + 1]) - (4653 * src[srcOffset + 2])) >> 16) + 128;\n \
\n \
dst[id] = (v2 << 24) | (y2 << 16) | (u1 << 8) | (y1); \n \
} \n ";
void InitKernels()
{
From_program = OpenCL::CompileProgram(__ConvertFromXFB);
From_kernel = OpenCL::CompileKernel(From_program, "ConvertFromXFB");
To_program = OpenCL::CompileProgram(__ConvertToXFB);
To_kernel = OpenCL::CompileKernel(To_program, "ConvertToXFB");
Inited = true;
}
#endif
void InitXFBConvTables()
{
for (int i = 0; i < 256; i++)
{
_y[i] = _mm_set_epi32(0xFFFFFFF, 76283*(i - 16), 76283*(i - 16), 76283*(i - 16));
_u[i] = _mm_set_epi32( 0, 0, -25624 * (i - 128), 132252 * (i - 128));
_v[i] = _mm_set_epi32( 0, 104595 * (i - 128), -53281 * (i - 128), 0);
_r1[i] = _mm_add_epi32(_mm_set_epi32( 28770 * i / 2, 0, -9699 * i / 2, 16843 * i),
_bias1);
_g1[i] = _mm_set_epi32(-24117 * i / 2, 0, -19071 * i / 2, 33030 * i);
_b1[i] = _mm_set_epi32( -4653 * i / 2, 0, 28770 * i / 2, 6423 * i);
_r2[i] = _mm_add_epi32(_mm_set_epi32( 28770 * i / 2, 16843 * i, -9699 * i / 2, 0),
_bias2);
_g2[i] = _mm_set_epi32(-24117 * i / 2, 33030 * i, -19071 * i / 2, 0);
_b2[i] = _mm_set_epi32( -4653 * i / 2, 6423 * i, 28770 * i / 2, 0);
}
}
void ConvertFromXFB(u32 *dst, const u8* _pXFB, int width, int height)
{
if (((size_t)dst & 0xF) != 0) {
PanicAlert("ConvertFromXFB - unaligned destination");
}
const unsigned char *src = _pXFB;
u32 numBlocks = ((width * height) / 2) / 2;
#if defined(HAVE_OPENCL) && HAVE_OPENCL
if(!Inited)
InitKernels();
int err;
size_t global = 0; // global domain size for our calculation
size_t local = 0; // local domain size for our calculation
printf("width %d, height %d\n", width, height);
// Create the input and output arrays in device memory for our calculation
//
cl_mem _dst = clCreateBuffer(OpenCL::g_context, CL_MEM_WRITE_ONLY, sizeof(unsigned int) * numBlocks, NULL, NULL);
if (!dst)
{
printf("Error: Failed to allocate device memory!\n");
exit(1);
}
cl_mem _src = clCreateBuffer(OpenCL::g_context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(unsigned char) * width * height, (void*)_pXFB, NULL);
if (!src)
{
printf("Error: Failed to allocate device memory!\n");
exit(1);
}
// Set the arguments to our compute kernel
//
err = 0;
err = clSetKernelArg(From_kernel, 0, sizeof(cl_mem), &_dst);
err |= clSetKernelArg(From_kernel, 1, sizeof(cl_mem), &_src);
if (err != CL_SUCCESS)
{
printf("Error: Failed to set kernel arguments! %d\n", err);
exit(1);
}
// Get the maximum work group size for executing the kernel on the device
//
err = clGetKernelWorkGroupInfo(From_kernel, OpenCL::device_id, CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), &local, NULL);
if (err != CL_SUCCESS)
{
printf("Error: Failed to retrieve kernel work group info! %d\n", err);
local = 64;
}
// Execute the kernel over the entire range of our 1d input data set
// using the maximum number of work group items for this device
//
global = numBlocks;
if(global < local)
{
// Global can't be less than local
}
err = clEnqueueNDRangeKernel(OpenCL::g_cmdq, From_kernel, 1, NULL, &global, &local, 0, NULL, NULL);
if (err != CL_SUCCESS)
{
printf("Error: Failed to execute kernel! %d\n", err);
return;
}
// Wait for the command commands to get serviced before reading back results
//
clFinish(OpenCL::g_cmdq);
// Read back the results from the device to verify the output
//
err = clEnqueueReadBuffer( OpenCL::g_cmdq, _dst, CL_TRUE, 0, sizeof(unsigned int) * numBlocks, dst, 0, NULL, NULL );
if (err != CL_SUCCESS)
{
printf("Error: Failed to read output array! %d\n", err);
exit(1);
}
clReleaseMemObject(_dst);
clReleaseMemObject(_src);
#else
for (u32 i = 0; i < numBlocks; i++)
{
__m128i y1 = _y[src[0]];
__m128i u = _u[src[1]];
__m128i y2 = _y[src[2]];
__m128i v = _v[src[3]];
__m128i y1_2 = _y[src[4+0]];
__m128i u_2 = _u[src[4+1]];
__m128i y2_2 = _y[src[4+2]];
__m128i v_2 = _v[src[4+3]];
__m128i c1 = _mm_srai_epi32(_mm_add_epi32(y1, _mm_add_epi32(u, v)), 16);
__m128i c2 = _mm_srai_epi32(_mm_add_epi32(y2, _mm_add_epi32(u, v)), 16);
__m128i c3 = _mm_srai_epi32(_mm_add_epi32(y1_2, _mm_add_epi32(u_2, v_2)), 16);
__m128i c4 = _mm_srai_epi32(_mm_add_epi32(y2_2, _mm_add_epi32(u_2, v_2)), 16);
__m128i four_dest = _mm_packus_epi16(_mm_packs_epi32(c1, c2), _mm_packs_epi32(c3, c4));
_mm_store_si128((__m128i *)dst, four_dest);
dst += 4;
src += 8;
}
#endif
}
void ConvertToXFB(u32 *dst, const u8* _pEFB, int width, int height)
{
const unsigned char *src = _pEFB;
u32 numBlocks = ((width * height) / 2) / 4;
if (((size_t)dst & 0xF) != 0) {
PanicAlert("ConvertToXFB - unaligned XFB");
}
#if defined(HAVE_OPENCL) && HAVE_OPENCL
if(!Inited)
InitKernels();
int err;
size_t global = 0; // global domain size for our calculation
size_t local = 0; // local domain size for our calculation
printf("width %d, height %d\n", width, height);
// Create the input and output arrays in device memory for our calculation
//
cl_mem _dst = clCreateBuffer(OpenCL::g_context, CL_MEM_WRITE_ONLY, sizeof(unsigned int) * numBlocks, NULL, NULL);
if (!dst)
{
printf("Error: Failed to allocate device memory!\n");
exit(1);
}
cl_mem _src = clCreateBuffer(OpenCL::g_context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(unsigned char) * width * height, (void*)_pEFB, NULL);
if (!src)
{
printf("Error: Failed to allocate device memory!\n");
exit(1);
}
// Set the arguments to our compute kernel
//
err = 0;
err = clSetKernelArg(To_kernel, 0, sizeof(cl_mem), &_dst);
err |= clSetKernelArg(To_kernel, 1, sizeof(cl_mem), &_src);
if (err != CL_SUCCESS)
{
printf("Error: Failed to set kernel arguments! %d\n", err);
exit(1);
}
// Get the maximum work group size for executing the kernel on the device
//
err = clGetKernelWorkGroupInfo(To_kernel, OpenCL::device_id, CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), &local, NULL);
if (err != CL_SUCCESS)
{
printf("Error: Failed to retrieve kernel work group info! %d\n", err);
local = 64;
}
// Execute the kernel over the entire range of our 1d input data set
// using the maximum number of work group items for this device
//
global = numBlocks;
if(global < local)
{
// Global can't be less than local
}
err = clEnqueueNDRangeKernel(OpenCL::g_cmdq, To_kernel, 1, NULL, &global, &local, 0, NULL, NULL);
if (err != CL_SUCCESS)
{
printf("Error: Failed to execute kernel! %d\n", err);
return;
}
// Wait for the command commands to get serviced before reading back results
//
clFinish(OpenCL::g_cmdq);
// Read back the results from the device to verify the output
//
err = clEnqueueReadBuffer( OpenCL::g_cmdq, _dst, CL_TRUE, 0, sizeof(unsigned int) * numBlocks, dst, 0, NULL, NULL );
if (err != CL_SUCCESS)
{
printf("Error: Failed to read output array! %d\n", err);
exit(1);
}
clReleaseMemObject(_dst);
clReleaseMemObject(_src);
#else
for (u32 i = 0; i < numBlocks; i++)
{
__m128i yuyv0 = _mm_srai_epi32(
_mm_add_epi32(
_mm_add_epi32(_r1[src[0]], _mm_add_epi32(_g1[src[1]], _b1[src[2]])),
_mm_add_epi32(_r2[src[4]], _mm_add_epi32(_g2[src[5]], _b2[src[6]]))), 16);
src += 8;
__m128i yuyv1 = _mm_srai_epi32(
_mm_add_epi32(
_mm_add_epi32(_r1[src[0]], _mm_add_epi32(_g1[src[1]], _b1[src[2]])),
_mm_add_epi32(_r2[src[4]], _mm_add_epi32(_g2[src[5]], _b2[src[6]]))), 16);
src += 8;
__m128i yuyv2 = _mm_srai_epi32(
_mm_add_epi32(
_mm_add_epi32(_r1[src[0]], _mm_add_epi32(_g1[src[1]], _b1[src[2]])),
_mm_add_epi32(_r2[src[4]], _mm_add_epi32(_g2[src[5]], _b2[src[6]]))), 16);
src += 8;
__m128i yuyv3 = _mm_srai_epi32(
_mm_add_epi32(
_mm_add_epi32(_r1[src[0]], _mm_add_epi32(_g1[src[1]], _b1[src[2]])),
_mm_add_epi32(_r2[src[4]], _mm_add_epi32(_g2[src[5]], _b2[src[6]]))), 16);
src += 8;
__m128i four_dest = _mm_packus_epi16(_mm_packs_epi32(yuyv0, yuyv1), _mm_packs_epi32(yuyv2, yuyv3));
_mm_store_si128((__m128i *)dst, four_dest);
dst += 4;
}
#endif
}