[Bf-blender-cvs] [944b632] master: Cycles: Log whch optimizations are used for CPU kernels
Sergey Sharybin
noreply at git.blender.org
Wed Jan 6 16:28:16 CET 2016
Commit: 944b6322e6e8fbd6b496702851cd2dedfdaec91b
Author: Sergey Sharybin
Date: Wed Jan 6 20:25:19 2016 +0500
Branches: master
https://developer.blender.org/rB944b6322e6e8fbd6b496702851cd2dedfdaec91b
Cycles: Log whch optimizations are used for CPU kernels
Not fully thread-safe, but is rather harmless. Just some messages
might be logged several times.
===================================================================
M intern/cycles/device/device_cpu.cpp
M intern/cycles/util/util_logging.h
===================================================================
diff --git a/intern/cycles/device/device_cpu.cpp b/intern/cycles/device/device_cpu.cpp
index 1b07ab6..1d57089 100644
--- a/intern/cycles/device/device_cpu.cpp
+++ b/intern/cycles/device/device_cpu.cpp
@@ -181,6 +181,8 @@ public:
void thread_path_trace(DeviceTask& task)
{
+ static bool cpu_type_logged = false;
+
if(task_pool.canceled()) {
if(task.need_finish_queue == false)
return;
@@ -197,31 +199,44 @@ public:
void(*path_trace_kernel)(KernelGlobals*, float*, unsigned int*, int, int, int, int, int);
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_AVX2
- if(system_cpu_support_avx2())
+ if(system_cpu_support_avx2()) {
path_trace_kernel = kernel_cpu_avx2_path_trace;
+ VLOG_ONCE(1, cpu_type_logged) << "Path tracing using AVX2 kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_AVX
- if(system_cpu_support_avx())
+ if(system_cpu_support_avx()) {
path_trace_kernel = kernel_cpu_avx_path_trace;
+ VLOG_ONCE(1, cpu_type_logged) << "Path tracing using AVX kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE41
- if(system_cpu_support_sse41())
+ if(system_cpu_support_sse41()) {
path_trace_kernel = kernel_cpu_sse41_path_trace;
+ VLOG_ONCE(1, cpu_type_logged) << "Path tracing using SSE4.1 kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE3
- if(system_cpu_support_sse3())
+ if(system_cpu_support_sse3()) {
path_trace_kernel = kernel_cpu_sse3_path_trace;
+ VLOG_ONCE(1, cpu_type_logged) << "Path tracing using SSE3 kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE2
- if(system_cpu_support_sse2())
+ if(system_cpu_support_sse2()) {
path_trace_kernel = kernel_cpu_sse2_path_trace;
+ VLOG_ONCE(1, cpu_type_logged) << "Path tracing using SSE2 kernel.";
+ }
else
#endif
+ {
path_trace_kernel = kernel_cpu_path_trace;
+ VLOG_ONCE(1, cpu_type_logged) << "Path tracing using regular kernel.";
+ }
while(task.acquire_tile(this, tile)) {
float *render_buffer = (float*)tile.buffer;
@@ -262,36 +277,50 @@ public:
void thread_film_convert(DeviceTask& task)
{
+ static bool cpu_type_logged = false;
float sample_scale = 1.0f/(task.sample + 1);
if(task.rgba_half) {
void(*convert_to_half_float_kernel)(KernelGlobals *, uchar4 *, float *, float, int, int, int, int);
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_AVX2
- if(system_cpu_support_avx2())
+ if(system_cpu_support_avx2()) {
convert_to_half_float_kernel = kernel_cpu_avx2_convert_to_half_float;
+ VLOG_ONCE(1, cpu_type_logged) << "Converting to half float using AVX2 kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_AVX
- if(system_cpu_support_avx())
+ if(system_cpu_support_avx()) {
convert_to_half_float_kernel = kernel_cpu_avx_convert_to_half_float;
+ VLOG_ONCE(1, cpu_type_logged) << "Converting to half float using AVX kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE41
- if(system_cpu_support_sse41())
+ if(system_cpu_support_sse41()) {
convert_to_half_float_kernel = kernel_cpu_sse41_convert_to_half_float;
+ VLOG_ONCE(1, cpu_type_logged) << "Converting to half float using SSE4.1 kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE3
- if(system_cpu_support_sse3())
+ if(system_cpu_support_sse3()) {
convert_to_half_float_kernel = kernel_cpu_sse3_convert_to_half_float;
+ VLOG_ONCE(1, cpu_type_logged) << "Converting to half float using SSE3 kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE2
- if(system_cpu_support_sse2())
+ if(system_cpu_support_sse2()) {
convert_to_half_float_kernel = kernel_cpu_sse2_convert_to_half_float;
+ VLOG_ONCE(1, cpu_type_logged) << "Converting to half float using SSE2 kernel.";
+ }
else
#endif
+ {
convert_to_half_float_kernel = kernel_cpu_convert_to_half_float;
+ VLOG_ONCE(1, cpu_type_logged) << "Converting to half float using regular kernel.";
+ }
for(int y = task.y; y < task.y + task.h; y++)
for(int x = task.x; x < task.x + task.w; x++)
@@ -301,31 +330,44 @@ public:
else {
void(*convert_to_byte_kernel)(KernelGlobals *, uchar4 *, float *, float, int, int, int, int);
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_AVX2
- if(system_cpu_support_avx2())
+ if(system_cpu_support_avx2()) {
convert_to_byte_kernel = kernel_cpu_avx2_convert_to_byte;
+ VLOG_ONCE(1, cpu_type_logged) << "Converting to byte using AVX2 kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_AVX
- if(system_cpu_support_avx())
+ if(system_cpu_support_avx()) {
convert_to_byte_kernel = kernel_cpu_avx_convert_to_byte;
+ VLOG_ONCE(1, cpu_type_logged) << "Converting to byte using AVX kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE41
- if(system_cpu_support_sse41())
+ if(system_cpu_support_sse41()) {
convert_to_byte_kernel = kernel_cpu_sse41_convert_to_byte;
+ VLOG_ONCE(1, cpu_type_logged) << "Converting to byte using SSE4.1 kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE3
- if(system_cpu_support_sse3())
+ if(system_cpu_support_sse3()) {
convert_to_byte_kernel = kernel_cpu_sse3_convert_to_byte;
+ VLOG_ONCE(1, cpu_type_logged) << "Converting to byte using SSE3 kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE2
- if(system_cpu_support_sse2())
+ if(system_cpu_support_sse2()) {
convert_to_byte_kernel = kernel_cpu_sse2_convert_to_byte;
+ VLOG_ONCE(1, cpu_type_logged) << "Converting to byte using SSE2 kernel.";
+ }
else
#endif
+ {
convert_to_byte_kernel = kernel_cpu_convert_to_byte;
+ VLOG_ONCE(1, cpu_type_logged) << "Converting to byte using regular kernel.";
+ }
for(int y = task.y; y < task.y + task.h; y++)
for(int x = task.x; x < task.x + task.w; x++)
@@ -338,6 +380,7 @@ public:
void thread_shader(DeviceTask& task)
{
KernelGlobals kg = kernel_globals;
+ static bool cpu_type_logged = false;
#ifdef WITH_OSL
OSLShader::thread_init(&kg, &kernel_globals, &osl_globals);
@@ -345,31 +388,44 @@ public:
void(*shader_kernel)(KernelGlobals*, uint4*, float4*, float*, int, int, int, int);
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_AVX2
- if(system_cpu_support_avx2())
+ if(system_cpu_support_avx2()) {
shader_kernel = kernel_cpu_avx2_shader;
+ VLOG_ONCE(1, cpu_type_logged) << "Shading using AVX2 kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_AVX
- if(system_cpu_support_avx())
+ if(system_cpu_support_avx()) {
shader_kernel = kernel_cpu_avx_shader;
+ VLOG_ONCE(1, cpu_type_logged) << "Shading using AVX kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE41
- if(system_cpu_support_sse41())
+ if(system_cpu_support_sse41()) {
shader_kernel = kernel_cpu_sse41_shader;
+ VLOG_ONCE(1, cpu_type_logged) << "Shading using SSE4.1 kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE3
- if(system_cpu_support_sse3())
+ if(system_cpu_support_sse3()) {
shader_kernel = kernel_cpu_sse3_shader;
+ VLOG_ONCE(1, cpu_type_logged) << "Shading using SSE3 kernel.";
+ }
else
#endif
#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE2
- if(system_cpu_support_sse2())
+ if(system_cpu_support_sse2()) {
shader_kernel = kernel_cpu_sse2_shader;
+ VLOG_ONCE(1, cpu_type_logged) << "Shading using SSE2 kernel.";
+ }
else
#endif
+ {
shader_kernel = kernel_cpu_shader;
+ VLOG_ONCE(1, cpu_type_logged) << "Shading using regular kernel.";
+ }
for(int sample = 0; sample < task.num_samples; sample++) {
for(int x = task.shader_x; x < task.shader_x + task.shader_w; x++)
diff --git a/intern/cycles/util/util_logging.h b/intern/cycles/util/util_logging.h
index 7fc42ac..2aa9c25 100644
--- a/intern/cycles/util/util_logging.h
+++ b/intern/cycles/util/util_logging.h
@@ -43,6 +43,8 @@ public:
#endif
+#define VLOG_ONCE(level, flag) if(!flag) flag = true, VLOG(level)
+
struct float3;
void util_logging_init(const char *argv0);
More information about the Bf-blender-cvs
mailing list