[Bf-blender-cvs] [a8cc0d707e8] master: Code refactor: split defines into separate header, changes to SSE type headers.
Brecht Van Lommel
noreply at git.blender.org
Mon Aug 7 14:03:44 CEST 2017
Commit: a8cc0d707e82ac781f44bf6cd7ed1e8974d8ed39
Author: Brecht Van Lommel
Date: Wed Aug 2 02:09:08 2017 +0200
Branches: master
https://developer.blender.org/rBa8cc0d707e82ac781f44bf6cd7ed1e8974d8ed39
Code refactor: split defines into separate header, changes to SSE type headers.
I need to use some macros defined in util_simd.h for float3/float4, to emulate
SSE4 instructions on SSE2. But due to issues with order of header includes this
was not possible, this does some refactoring to make it work.
Differential Revision: https://developer.blender.org/D2764
===================================================================
M intern/cycles/device/device_cpu.cpp
M intern/cycles/kernel/CMakeLists.txt
M intern/cycles/util/CMakeLists.txt
A intern/cycles/util/util_defines.h
M intern/cycles/util/util_optimization.h
M intern/cycles/util/util_simd.h
M intern/cycles/util/util_sseb.h
M intern/cycles/util/util_ssef.h
M intern/cycles/util/util_ssei.h
M intern/cycles/util/util_types.h
===================================================================
diff --git a/intern/cycles/device/device_cpu.cpp b/intern/cycles/device/device_cpu.cpp
index 18112437b45..a00be3eeaab 100644
--- a/intern/cycles/device/device_cpu.cpp
+++ b/intern/cycles/device/device_cpu.cpp
@@ -48,6 +48,7 @@
#include "util/util_logging.h"
#include "util/util_map.h"
#include "util/util_opengl.h"
+#include "util/util_optimization.h"
#include "util/util_progress.h"
#include "util/util_system.h"
#include "util/util_thread.h"
diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index 23e9bd311c4..88c4c4e3282 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/intern/cycles/kernel/CMakeLists.txt
@@ -233,6 +233,7 @@ set(SRC_FILTER_HEADERS
set(SRC_UTIL_HEADERS
../util/util_atomic.h
../util/util_color.h
+ ../util/util_defines.h
../util/util_half.h
../util/util_hash.h
../util/util_math.h
diff --git a/intern/cycles/util/CMakeLists.txt b/intern/cycles/util/CMakeLists.txt
index 43f9a57d099..7f3747a0f58 100644
--- a/intern/cycles/util/CMakeLists.txt
+++ b/intern/cycles/util/CMakeLists.txt
@@ -38,6 +38,7 @@ set(SRC_HEADERS
util_atomic.h
util_boundbox.h
util_debug.h
+ util_defines.h
util_guarded_allocator.cpp
util_foreach.h
util_function.h
diff --git a/intern/cycles/util/util_types.h b/intern/cycles/util/util_defines.h
similarity index 51%
copy from intern/cycles/util/util_types.h
copy to intern/cycles/util/util_defines.h
index a5d1d7152d5..d0d87e74332 100644
--- a/intern/cycles/util/util_types.h
+++ b/intern/cycles/util/util_defines.h
@@ -1,5 +1,6 @@
+
/*
- * Copyright 2011-2013 Blender Foundation
+ * Copyright 2011-2017 Blender Foundation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
@@ -14,12 +15,8 @@
* limitations under the License.
*/
-#ifndef __UTIL_TYPES_H__
-#define __UTIL_TYPES_H__
-
-#ifndef __KERNEL_OPENCL__
-# include <stdlib.h>
-#endif
+#ifndef __UTIL_DEFINES_H__
+#define __UTIL_DEFINES_H__
/* Bitness */
@@ -77,130 +74,6 @@
# endif
#endif /* __KERNEL_GPU__ */
-/* Standard Integer Types */
-
-#ifndef __KERNEL_GPU__
-/* int8_t, uint16_t, and friends */
-# ifndef _WIN32
-# include <stdint.h>
-# endif
-/* SIMD Types */
-# include "util/util_optimization.h"
-#endif /* __KERNEL_GPU__ */
-
-CCL_NAMESPACE_BEGIN
-
-/* Types
- *
- * Define simpler unsigned type names, and integer with defined number of bits.
- * Also vector types, named to be compatible with OpenCL builtin types, while
- * working for CUDA and C++ too. */
-
-/* Shorter Unsigned Names */
-
-#ifndef __KERNEL_OPENCL__
-typedef unsigned char uchar;
-typedef unsigned int uint;
-typedef unsigned short ushort;
-#endif
-
-/* Fixed Bits Types */
-
-#ifdef __KERNEL_OPENCL__
-typedef ulong uint64_t;
-#endif
-
-#ifndef __KERNEL_GPU__
-# ifdef _WIN32
-typedef signed char int8_t;
-typedef unsigned char uint8_t;
-
-typedef signed short int16_t;
-typedef unsigned short uint16_t;
-
-typedef signed int int32_t;
-typedef unsigned int uint32_t;
-
-typedef long long int64_t;
-typedef unsigned long long uint64_t;
-# ifdef __KERNEL_64_BIT__
-typedef int64_t ssize_t;
-# else
-typedef int32_t ssize_t;
-# endif
-# endif /* _WIN32 */
-
-/* Generic Memory Pointer */
-
-typedef uint64_t device_ptr;
-#endif /* __KERNEL_GPU__ */
-
-ccl_device_inline size_t align_up(size_t offset, size_t alignment)
-{
- return (offset + alignment - 1) & ~(alignment - 1);
-}
-
-ccl_device_inline size_t divide_up(size_t x, size_t y)
-{
- return (x + y - 1) / y;
-}
-
-ccl_device_inline size_t round_up(size_t x, size_t multiple)
-{
- return ((x + multiple - 1) / multiple) * multiple;
-}
-
-ccl_device_inline size_t round_down(size_t x, size_t multiple)
-{
- return (x / multiple) * multiple;
-}
-
-/* Interpolation types for textures
- * cuda also use texture space to store other objects */
-enum InterpolationType {
- INTERPOLATION_NONE = -1,
- INTERPOLATION_LINEAR = 0,
- INTERPOLATION_CLOSEST = 1,
- INTERPOLATION_CUBIC = 2,
- INTERPOLATION_SMART = 3,
-
- INTERPOLATION_NUM_TYPES,
-};
-
-/* Texture types
- * Since we store the type in the lower bits of a flat index,
- * the shift and bit mask constant below need to be kept in sync.
- */
-
-enum ImageDataType {
- IMAGE_DATA_TYPE_FLOAT4 = 0,
- IMAGE_DATA_TYPE_BYTE4 = 1,
- IMAGE_DATA_TYPE_HALF4 = 2,
- IMAGE_DATA_TYPE_FLOAT = 3,
- IMAGE_DATA_TYPE_BYTE = 4,
- IMAGE_DATA_TYPE_HALF = 5,
-
- IMAGE_DATA_NUM_TYPES
-};
-
-#define IMAGE_DATA_TYPE_SHIFT 3
-#define IMAGE_DATA_TYPE_MASK 0x7
-
-/* Extension types for textures.
- *
- * Defines how the image is extrapolated past its original bounds.
- */
-enum ExtensionType {
- /* Cause the image to repeat horizontally and vertically. */
- EXTENSION_REPEAT = 0,
- /* Extend by repeating edge pixels of the image. */
- EXTENSION_EXTEND = 1,
- /* Clip to image size and set exterior pixels as transparent. */
- EXTENSION_CLIP = 2,
-
- EXTENSION_NUM_TYPES,
-};
-
/* macros */
/* hints for branch prediction, only use in code that runs a _lot_ */
@@ -250,9 +123,6 @@ template<typename T> static inline T decltype_helper(T x) { return x; }
#define CHECK_TYPE_INLINE(val, type) \
((void)(((type)0) != (val)))
-
-CCL_NAMESPACE_END
-
#ifndef __KERNEL_GPU__
# include <cassert>
# define util_assert(statement) assert(statement)
@@ -260,43 +130,5 @@ CCL_NAMESPACE_END
# define util_assert(statement)
#endif
-/* Vectorized types declaration. */
-#include "util/util_types_uchar2.h"
-#include "util/util_types_uchar3.h"
-#include "util/util_types_uchar4.h"
-
-#include "util/util_types_int2.h"
-#include "util/util_types_int3.h"
-#include "util/util_types_int4.h"
-
-#include "util/util_types_uint2.h"
-#include "util/util_types_uint3.h"
-#include "util/util_types_uint4.h"
-
-#include "util/util_types_float2.h"
-#include "util/util_types_float3.h"
-#include "util/util_types_float4.h"
-
-#include "util/util_types_vector3.h"
-
-/* Vectorized types implementation. */
-#include "util/util_types_uchar2_impl.h"
-#include "util/util_types_uchar3_impl.h"
-#include "util/util_types_uchar4_impl.h"
-
-#include "util/util_types_int2_impl.h"
-#include "util/util_types_int3_impl.h"
-#include "util/util_types_int4_impl.h"
-
-#include "util/util_types_uint2_impl.h"
-#include "util/util_types_uint3_impl.h"
-#include "util/util_types_uint4_impl.h"
-
-#include "util/util_types_float2_impl.h"
-#include "util/util_types_float3_impl.h"
-#include "util/util_types_float4_impl.h"
-
-#include "util/util_types_vector3_impl.h"
-
-#endif /* __UTIL_TYPES_H__ */
+#endif /* __UTIL_DEFINES_H__ */
diff --git a/intern/cycles/util/util_optimization.h b/intern/cycles/util/util_optimization.h
index 6f70a474fe7..0382c0811dd 100644
--- a/intern/cycles/util/util_optimization.h
+++ b/intern/cycles/util/util_optimization.h
@@ -19,16 +19,6 @@
#ifndef __KERNEL_GPU__
-/* quiet unused define warnings */
-#if defined(__KERNEL_SSE2__) || \
- defined(__KERNEL_SSE3__) || \
- defined(__KERNEL_SSSE3__) || \
- defined(__KERNEL_SSE41__) || \
- defined(__KERNEL_AVX__) || \
- defined(__KERNEL_AVX2__)
- /* do nothing */
-#endif
-
/* x86
*
* Compile a regular, SSE2 and SSE3 kernel. */
@@ -73,48 +63,6 @@
#endif /* defined(__x86_64__) || defined(_M_X64) */
-/* SSE Experiment
- *
- * This is disabled code for an experiment to use SSE types globally for types
- * such as float3 and float4. Currently this gives an overall slowdown. */
-
-#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
-
-/* SSE Intrinsics includes
- *
- * We assume __KERNEL_SSEX__ flags to have been defined at this point */
-
-/* SSE intrinsics headers */
-#ifndef FREE_WINDOWS64
-
-#ifdef _MSC_VER
-# include <intrin.h>
-#elif (defined(__x86_64__) || defined(__i386__))
-# include <x86intrin.h>
-#endif
-
-#else
-
-/* MinGW64 has conflicting declarations for these SSE headers in <windows.h>.
- * Since we can't avoid including <windows.h>, better only include that */
-#include "util/util_windows.h"
-
-#endif
-
#endif
#endif /* __UTIL_OPTIMIZATION_H__ */
diff --git a/intern/cycles/util/util_simd.h b/intern/cycles/util/util_simd.h
index 587febe3e52..7d938a0fbca 100644
--- a/intern/cycles/util/util_simd.h
+++ b/intern/cycles/util/util_simd.h
@@ -18,19 +18,38 @@
#ifndef __UTIL_SIMD_TYPES_H__
#define __UTIL_SIMD_TYPES_H__
+#ifndef __KERNEL_GPU__
+
#include <limits>
#include "util/util_debug.h"
-#include "util/util_types.h"
+#include "util/util_defines.h"
+
+/* SSE Intrinsics includes
+ *
+ * We assume __KERNEL_SSEX__ flags to have been defined at this point */
+
+/* SSE intrinsics headers */
+#ifndef FREE_WINDOWS64
+
+#ifdef _MSC_VER
+# include <intrin.h>
+#elif (defined(__x86_64__) || defined(__i386__))
+# include <x86intrin.h>
+#endif
+
+#else
+
+/* MinGW64 has conflicting declarations for these SSE headers in <windows.h>.
+ * Since we can't avoid including <windows.h>, better only include that */
+#include "util/util_windows.h"
+
+#endif
CCL_NAMESPACE_BEGIN
#ifdef __KERNEL_SSE2__
-struct sseb;
-struct ssei;
-struct ssef;
-
extern const __m128 _mm_lookupmask_ps[16];
/* Special Types */
@@ -496,13 +515,19 @@ ccl_device_inline int bitscan(int value)
#endif /* __KERNEL_SSE2__ */
+/* quiet unused define warnings */
+#if defined(__KERNEL_SSE2__) || \
+ defined(__KERNEL_SSE3__) || \
+ defined(__KERNEL_SSSE3__) || \
+ defined(__KERNEL_SSE41__) || \
+ defined(__KERNEL_AVX__) || \
+ defined(__KERNEL_AVX2__)
+ /* do nothing */
+#endif
+
CCL_NAMESPACE_END
-#include "util/util_math.h"
-#include "util/util_sseb.h"
-#include "util/util_ssei.h"
-#include "util/util_ssef.h"
-#include "util/util_avxf.h"
+#endif /* __KERNEL_GPU__ */
#endif /* __UTIL_SIMD_TYPES_H__ */
diff --git a/intern/cycles/util/util_sseb.h b/intern/cycles/util/util_sseb.h
index 6e669701f3b..93c22aafdcd 100644
--- a/intern/cycles/util/util_sseb.h
+++ b/intern/cycles/util/util_sseb.h
@@ -22,6 +22,9 @@ CCL_NAMESPACE_BEGIN
#ifdef __KERNEL_SSE2__
+struct ssei;
+struct ssef;
+
/*! 4-wide SSE
@@ Diff output truncated at 10240 characters. @@
More information about the Bf-blender-cvs
mailing list