diff options
author | Brecht Van Lommel <brechtvanlommel@pandora.be> | 2013-02-12 18:58:46 +0400 |
---|---|---|
committer | Brecht Van Lommel <brechtvanlommel@pandora.be> | 2013-02-12 18:58:46 +0400 |
commit | d095bcc8aa9ec60e537ed86b48f1025623cb2b38 (patch) | |
tree | 93b3ab24f502b561803ad9e4e9b84538208651bd /intern/cycles | |
parent | c3c4ef3c6f09f10b29074225e412ae7ce67e8955 (diff) |
Fix cycles not using SSE3 kernel after recent, order with SSE2 should be switched,
pointed out by Chad Fraleigh.
Diffstat (limited to 'intern/cycles')
-rw-r--r-- | intern/cycles/device/device_cpu.cpp | 24 |
1 files changed, 12 insertions, 12 deletions
diff --git a/intern/cycles/device/device_cpu.cpp b/intern/cycles/device/device_cpu.cpp index 1915245bb55..c2ba91faf69 100644 --- a/intern/cycles/device/device_cpu.cpp +++ b/intern/cycles/device/device_cpu.cpp @@ -171,7 +171,7 @@ public: int end_sample = tile.start_sample + tile.num_samples; #ifdef WITH_OPTIMIZED_KERNEL - if(system_cpu_support_sse2()) { + if(system_cpu_support_sse3()) { for(int sample = start_sample; sample < end_sample; sample++) { if (task.get_cancel() || task_pool.cancelled()) { if(task.need_finish_queue == false) @@ -180,7 +180,7 @@ public: for(int y = tile.y; y < tile.y + tile.h; y++) { for(int x = tile.x; x < tile.x + tile.w; x++) { - kernel_cpu_sse2_path_trace(&kg, render_buffer, rng_state, + kernel_cpu_sse3_path_trace(&kg, render_buffer, rng_state, sample, x, y, tile.offset, tile.stride); } } @@ -190,7 +190,7 @@ public: task.update_progress(tile); } } - else if(system_cpu_support_sse3()) { + else if(system_cpu_support_sse2()) { for(int sample = start_sample; sample < end_sample; sample++) { if (task.get_cancel() || task_pool.cancelled()) { if(task.need_finish_queue == false) @@ -199,7 +199,7 @@ public: for(int y = tile.y; y < tile.y + tile.h; y++) { for(int x = tile.x; x < tile.x + tile.w; x++) { - kernel_cpu_sse3_path_trace(&kg, render_buffer, rng_state, + kernel_cpu_sse2_path_trace(&kg, render_buffer, rng_state, sample, x, y, tile.offset, tile.stride); } } @@ -247,16 +247,16 @@ public: void thread_tonemap(DeviceTask& task) { #ifdef WITH_OPTIMIZED_KERNEL - if(system_cpu_support_sse2()) { + if(system_cpu_support_sse3()) { for(int y = task.y; y < task.y + task.h; y++) for(int x = task.x; x < task.x + task.w; x++) - kernel_cpu_sse2_tonemap(&kernel_globals, (uchar4*)task.rgba, (float*)task.buffer, + kernel_cpu_sse3_tonemap(&kernel_globals, (uchar4*)task.rgba, (float*)task.buffer, task.sample, task.resolution, x, y, task.offset, task.stride); } - else if(system_cpu_support_sse3()) { + else if(system_cpu_support_sse2()) { for(int y = task.y; y < task.y + task.h; y++) for(int x = task.x; x < task.x + task.w; x++) - kernel_cpu_sse3_tonemap(&kernel_globals, (uchar4*)task.rgba, (float*)task.buffer, + kernel_cpu_sse2_tonemap(&kernel_globals, (uchar4*)task.rgba, (float*)task.buffer, task.sample, task.resolution, x, y, task.offset, task.stride); } else @@ -278,17 +278,17 @@ public: #endif #ifdef WITH_OPTIMIZED_KERNEL - if(system_cpu_support_sse2()) { + if(system_cpu_support_sse3()) { for(int x = task.shader_x; x < task.shader_x + task.shader_w; x++) { - kernel_cpu_sse2_shader(&kg, (uint4*)task.shader_input, (float4*)task.shader_output, task.shader_eval_type, x); + kernel_cpu_sse3_shader(&kg, (uint4*)task.shader_input, (float4*)task.shader_output, task.shader_eval_type, x); if(task_pool.cancelled()) break; } } - else if(system_cpu_support_sse3()) { + else if(system_cpu_support_sse2()) { for(int x = task.shader_x; x < task.shader_x + task.shader_w; x++) { - kernel_cpu_sse3_shader(&kg, (uint4*)task.shader_input, (float4*)task.shader_output, task.shader_eval_type, x); + kernel_cpu_sse2_shader(&kg, (uint4*)task.shader_input, (float4*)task.shader_output, task.shader_eval_type, x); if(task_pool.cancelled()) break; |