Cycles: Skip the compilation of the dedicated SSE2 kernel on x86-64, we can assume SSE2 here, so just re-use the regular one. Saves 500kb in the blender binary.

Reviewed by: brecht
Differential Revision: https://developer.blender.org/D199

6 files changed, 26 insertions, 15 deletions

diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index 998d1a3540f..81499bbfda8 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/intern/cycles/kernel/CMakeLists.txt
@@ -192,10 +192,8 @@ endif()
 include_directories(${INC})
 include_directories(SYSTEM ${INC_SYS})
 
-if(WITH_CYCLES_OPTIMIZED_KERNEL)
-	set_source_files_properties(kernel_sse2.cpp PROPERTIES COMPILE_FLAGS "${CYCLES_SSE2_KERNEL_FLAGS}")
-	set_source_files_properties(kernel_sse3.cpp PROPERTIES COMPILE_FLAGS "${CYCLES_SSE3_KERNEL_FLAGS}")
-endif()
+set_source_files_properties(kernel_sse2.cpp PROPERTIES COMPILE_FLAGS "${CYCLES_SSE2_KERNEL_FLAGS}")
+set_source_files_properties(kernel_sse3.cpp PROPERTIES COMPILE_FLAGS "${CYCLES_SSE3_KERNEL_FLAGS}")
 
 if(WITH_CYCLES_OPTIMIZED_KERNEL_SSE41)
 	set_source_files_properties(kernel_sse41.cpp PROPERTIES COMPILE_FLAGS "${CYCLES_SSE41_KERNEL_FLAGS}")
diff --git a/intern/cycles/kernel/kernel.cpp b/intern/cycles/kernel/kernel.cpp
index 3e2727fde9a..3fe1e80890b 100644
--- a/intern/cycles/kernel/kernel.cpp
+++ b/intern/cycles/kernel/kernel.cpp
@@ -84,6 +84,11 @@ void kernel_tex_copy(KernelGlobals *kg, const char *name, device_ptr mem, size_t
 		assert(0);
 }
 
+/* On x86-64, we can assume SSE2, so avoid the extra kernel and compile this one with SSE2 intrinsics */
+#if defined(__x86_64__) || defined(_M_X64)
+#define __KERNEL_SSE2__
+#endif
+
 /* Path Tracing */
 
 void kernel_cpu_path_trace(KernelGlobals *kg, float *buffer, unsigned int *rng_state, int sample, int x, int y, int offset, int stride)
diff --git a/intern/cycles/kernel/kernel.h b/intern/cycles/kernel/kernel.h
index 105a3887da0..b6db92f26e9 100644
--- a/intern/cycles/kernel/kernel.h
+++ b/intern/cycles/kernel/kernel.h
@@ -17,9 +17,10 @@
 #ifndef __KERNEL_H__
 #define __KERNEL_H__
 
-/* CPU Kernel Interfae */
+/* CPU Kernel Interface */
 
 #include "util_types.h"
+#include "util_optimization.h"
 
 CCL_NAMESPACE_BEGIN
 
@@ -43,7 +44,7 @@ void kernel_cpu_convert_to_half_float(KernelGlobals *kg, uchar4 *rgba, float *bu
 void kernel_cpu_shader(KernelGlobals *kg, uint4 *input, float4 *output,
 	int type, int i);
 
-#ifdef WITH_OPTIMIZED_KERNEL
+#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE2
 void kernel_cpu_sse2_path_trace(KernelGlobals *kg, float *buffer, unsigned int *rng_state,
 	int sample, int x, int y, int offset, int stride);
 void kernel_cpu_sse2_convert_to_byte(KernelGlobals *kg, uchar4 *rgba, float *buffer,
@@ -52,7 +53,9 @@ void kernel_cpu_sse2_convert_to_half_float(KernelGlobals *kg, uchar4 *rgba, floa
 	float sample_scale, int x, int y, int offset, int stride);
 void kernel_cpu_sse2_shader(KernelGlobals *kg, uint4 *input, float4 *output,
 	int type, int i);
+#endif
 
+#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE3
 void kernel_cpu_sse3_path_trace(KernelGlobals *kg, float *buffer, unsigned int *rng_state,
 	int sample, int x, int y, int offset, int stride);
 void kernel_cpu_sse3_convert_to_byte(KernelGlobals *kg, uchar4 *rgba, float *buffer,
@@ -61,7 +64,9 @@ void kernel_cpu_sse3_convert_to_half_float(KernelGlobals *kg, uchar4 *rgba, floa
 	float sample_scale, int x, int y, int offset, int stride);
 void kernel_cpu_sse3_shader(KernelGlobals *kg, uint4 *input, float4 *output,
 	int type, int i);
+#endif
 
+#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE41
 void kernel_cpu_sse41_path_trace(KernelGlobals *kg, float *buffer, unsigned int *rng_state,
 	int sample, int x, int y, int offset, int stride);
 void kernel_cpu_sse41_convert_to_byte(KernelGlobals *kg, uchar4 *rgba, float *buffer,
diff --git a/intern/cycles/kernel/kernel_sse2.cpp b/intern/cycles/kernel/kernel_sse2.cpp
index 9c69e519dca..953c3e4f9c9 100644
--- a/intern/cycles/kernel/kernel_sse2.cpp
+++ b/intern/cycles/kernel/kernel_sse2.cpp
@@ -17,8 +17,10 @@
 /* Optimized CPU kernel entry points. This file is compiled with SSE2
  * optimization flags and nearly all functions inlined, while kernel.cpp
  * is compiled without for other CPU's. */
-
-#ifdef WITH_OPTIMIZED_KERNEL
+ 
+#include "util_optimization.h"
+ 
+#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE2
 
 /* SSE optimization disabled for now on 32 bit, see bug #36316 */
 #if !(defined(__GNUC__) && (defined(i386) || defined(_M_IX86)))
@@ -70,4 +72,3 @@ void kernel_cpu_sse2_shader(KernelGlobals *kg, uint4 *input, float4 *output, int
 CCL_NAMESPACE_END
 
 #endif
-
diff --git a/intern/cycles/kernel/kernel_sse3.cpp b/intern/cycles/kernel/kernel_sse3.cpp
index 05877a41b4a..2a36c974191 100644
--- a/intern/cycles/kernel/kernel_sse3.cpp
+++ b/intern/cycles/kernel/kernel_sse3.cpp
@@ -17,8 +17,10 @@
 /* Optimized CPU kernel entry points. This file is compiled with SSE3/SSSE3
  * optimization flags and nearly all functions inlined, while kernel.cpp
  * is compiled without for other CPU's. */
-
-#ifdef WITH_OPTIMIZED_KERNEL
+ 
+#include "util_optimization.h"
+ 
+#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE3
 
 /* SSE optimization disabled for now on 32 bit, see bug #36316 */
 #if !(defined(__GNUC__) && (defined(i386) || defined(_M_IX86)))
@@ -72,4 +74,3 @@ void kernel_cpu_sse3_shader(KernelGlobals *kg, uint4 *input, float4 *output, int
 CCL_NAMESPACE_END
 
 #endif
-
diff --git a/intern/cycles/kernel/kernel_sse41.cpp b/intern/cycles/kernel/kernel_sse41.cpp
index 0c68fd3651b..6583feaeb45 100644
--- a/intern/cycles/kernel/kernel_sse41.cpp
+++ b/intern/cycles/kernel/kernel_sse41.cpp
@@ -17,8 +17,10 @@
 /* Optimized CPU kernel entry points. This file is compiled with SSE3/SSSE3
  * optimization flags and nearly all functions inlined, while kernel.cpp
  * is compiled without for other CPU's. */
-
-#ifdef WITH_OPTIMIZED_KERNEL
+ 
+#include "util_optimization.h"
+ 
+#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE41
 
 /* SSE optimization disabled for now on 32 bit, see bug #36316 */
 #if !(defined(__GNUC__) && (defined(i386) || defined(_M_IX86)))
@@ -73,4 +75,3 @@ void kernel_cpu_sse41_shader(KernelGlobals *kg, uint4 *input, float4 *output, in
 CCL_NAMESPACE_END
 
 #endif
-