Cr8-xyz/blender
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Code cleanup: cycles

Browse Source 
* Reshuffle SSE #ifdefs to try to avoid compilation errors enabling SSE on 32 bit.
* Remove CUDA kernel launch size exception on Mac, is not needed.
* Make OSL file compilation quiet like c/cpp files.
This commit is contained in:
Brecht Van Lommel
2013-06-26 23:29:33 +00:00
parent 372c4f80f4
commit 7902fa57b6
7 changed files with 40 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
intern/cycles/device/device_cuda.cpp
View File

@@ -607,13 +607,8 @@ public:
cuda_assert(cuParamSetSize(cuPathTrace, offset))
/* launch kernel: todo find optimal size, cache config for fermi */
#ifndef __APPLE__
int xthreads = 16;
int ythreads = 16;
#else
int xthreads = 8;
int ythreads = 8;
#endif
int xblocks = (rtile.w + xthreads - 1)/xthreads;
int yblocks = (rtile.h + ythreads - 1)/ythreads;
@@ -676,13 +671,8 @@ public:
cuda_assert(cuParamSetSize(cuFilmConvert, offset))
/* launch kernel: todo find optimal size, cache config for fermi */
#ifndef __APPLE__
int xthreads = 16;
int ythreads = 16;
#else
int xthreads = 8;
int ythreads = 8;
#endif
int xblocks = (task.w + xthreads - 1)/xthreads;
int yblocks = (task.h + ythreads - 1)/ythreads;
@@ -730,11 +720,7 @@ public:
cuda_assert(cuParamSetSize(cuDisplace, offset))
/* launch kernel: todo find optimal size, cache config for fermi */
#ifndef __APPLE__
int xthreads = 16;
#else
int xthreads = 8;
#endif
int xblocks = (task.shader_w + xthreads - 1)/xthreads;
cuda_assert(cuFuncSetCacheConfig(cuDisplace, CU_FUNC_CACHE_PREFER_L1))

1
intern/cycles/kernel/kernel_compat_opencl.h
View File

@@ -108,7 +108,6 @@
/* data lookup defines */
#define kernel_data (*kg->data)
#define kernel_tex_lookup(t, x, offset, size) kernel_tex_lookup_(kg->t, offset, size, x)
#define kernel_tex_fetch(t, index) kg->t[index]
/* define NULL */

2
intern/cycles/kernel/kernel_sse2.cpp
View File

@@ -22,7 +22,7 @@
#ifdef WITH_OPTIMIZED_KERNEL
//#define __KERNEL_SSE2__
#define __KERNEL_SSE2__
#include "kernel.h"
#include "kernel_compat_cpu.h"

2
intern/cycles/kernel/shaders/CMakeLists.txt
View File

@@ -91,7 +91,7 @@ foreach(_file ${SRC_OSL})
string(REPLACE ${CMAKE_SOURCE_DIR} ${CMAKE_BINARY_DIR} _OSO_FILE ${_OSO_FILE})
add_custom_command(
OUTPUT ${_OSO_FILE}
COMMAND ${OSL_COMPILER} -O2 -I"${CMAKE_CURRENT_SOURCE_DIR}" ${_OSL_FILE}
COMMAND ${OSL_COMPILER} -q -O2 -I"${CMAKE_CURRENT_SOURCE_DIR}" ${_OSL_FILE}
DEPENDS ${_OSL_FILE} ${SRC_OSL_HEADERS})
list(APPEND SRC_OSO
${_OSO_FILE}

2
intern/cycles/kernel/shaders/SConscript
View File

@@ -57,7 +57,7 @@ if env['WITH_BF_CYCLES_OSL']:
osl_file = os.path.join(source_dir, f)
oso_file = os.path.join(build_dir, f.replace('.osl', '.oso'))
command = "%s -O2 -I%s -o %s %s" % (osl_compiler, source_dir, oso_file, osl_file)
command = "%s -q -O2 -I%s -o %s %s" % (osl_compiler, source_dir, oso_file, osl_file)
shaders.Command(oso_file, f, command)
shaders.Depends(oso_file, [f] + dependencies)

68
intern/cycles/util/util_types.h
View File

@@ -61,59 +61,49 @@
#ifndef __KERNEL_GPU__
/* not enabled, globally applying it just gives slowdown,
* but useful for testing. */
//#define __KERNEL_SSE__
#ifdef __KERNEL_SSE__
#include <xmmintrin.h> /* SSE 1 */
#include <emmintrin.h> /* SSE 2 */
#include <pmmintrin.h> /* SSE 3 */
#include <tmmintrin.h> /* SSSE 3 */
#include <smmintrin.h> /* SSE 4 */
/* not enabled, globally applying it gives slowdown, only for testing. */
#if 0
#define __KERNEL_SSE__
#ifndef __KERNEL_SSE2__
#define __KERNEL_SSE2__
#endif
#ifndef __KERNEL_SSE3__
#define __KERNEL_SSE3__
#endif
#ifndef __KERNEL_SSSE3__
#define __KERNEL_SSSE3__
#endif
#ifndef __KERNEL_SSE4__
#define __KERNEL_SSE4__
#endif
#endif
#else
/* SSE2 is always available on x86_64 CPUs, so auto enable */
#if defined(__x86_64__) && !defined(__KERNEL_SSE2__)
#define __KERNEL_SSE2__
#endif
#if defined(__x86_64__) || defined(__KERNEL_SSSE3__)
/* SSE intrinsics headers */
#ifndef FREE_WINDOWS64
/* MinGW64 has conflicting declarations for these SSE headers in <windows.h>.
* Since we can't avoid including <windows.h>, better only include that */
#ifdef FREE_WINDOWS64
#include <windows.h>
#else
#ifdef __KERNEL_SSE2__
#include <xmmintrin.h> /* SSE 1 */
#include <emmintrin.h> /* SSE 2 */
#endif
#ifdef __KERNEL_SSE3__
#include <pmmintrin.h> /* SSE 3 */
#endif
#ifdef __KERNEL_SSSE3__
#include <tmmintrin.h> /* SSSE 3 */
#endif
#endif
/* SSE2 is available on x64 and SSE3 CPUs, so enable here as well */
#ifndef __KERNEL_SSE2__
#define __KERNEL_SSE2__
#endif
#else
#endif
/* MinGW64 has conflicting declarations for these SSE headers in <windows.h>.
* Since we can't avoid including <windows.h>, better only include that */
#include <windows.h>
#endif
@@ -553,30 +543,6 @@ template<size_t i0, size_t i1, size_t i2, size_t i3> __device_inline const __m12
}
#endif
#ifndef __KERNEL_GPU__
static inline void *malloc_aligned(size_t size, size_t alignment)
{
void *data = (void*)malloc(size + sizeof(void*) + alignment - 1);
union { void *ptr; size_t offset; } u;
u.ptr = (char*)data + sizeof(void*);
u.offset = (u.offset + alignment - 1) & ~(alignment - 1);
*(((void**)u.ptr) - 1) = data;
return u.ptr;
}
static inline void free_aligned(void *ptr)
{
if(ptr) {
void *data = *(((void**)ptr) - 1);
free(data);
}
}
#endif
CCL_NAMESPACE_END
#endif /* __UTIL_TYPES_H__ */

20
intern/cycles/util/util_vector.h
View File

@@ -30,6 +30,26 @@ CCL_NAMESPACE_BEGIN
using std::vector;
static inline void *malloc_aligned(size_t size, size_t alignment)
{
void *data = (void*)malloc(size + sizeof(void*) + alignment - 1);
union { void *ptr; size_t offset; } u;
u.ptr = (char*)data + sizeof(void*);
u.offset = (u.offset + alignment - 1) & ~(alignment - 1);
*(((void**)u.ptr) - 1) = data;
return u.ptr;
}
static inline void free_aligned(void *ptr)
{
if(ptr) {
void *data = *(((void**)ptr) - 1);
free(data);
}
}
/* Array
*
* Simplified version of vector, serving multiple purposes: