gpu, cmaa: copy RGBA8 texture via glCopyTexSubImage2D() instead of imageStore().

gpu/command_buffer/service/gles2_cmd_apply_framebuffer_attachment_cmaa_intel.cc

 // Copyright 2016 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "gpu/command_buffer/service/gles2_cmd_apply_framebuffer_attachment_cmaa_intel.h"
 
 #include "base/logging.h"
 #include "gpu/command_buffer/service/framebuffer_manager.h"
 #include "gpu/command_buffer/service/gles2_cmd_decoder.h"
 #include "ui/gl/gl_context.h"
@@ skipping 10 matching lines @@
     : initialized_(false),
       textures_initialized_(false),
       is_in_gamma_correct_mode_(false),
       supports_usampler_(true),
       supports_r8_image_(true),
       supports_r8_read_format_(true),
       is_gles31_compatible_(false),
       frame_id_(0),
       width_(0),
       height_(0),
-      copy_to_framebuffer_shader_(0),
-      copy_to_image_shader_(0),
       edges0_shader_(0),
       edges1_shader_(0),
       edges_combine_shader_(0),
       process_and_apply_shader_(0),
       debug_display_edges_shader_(0),
       cmaa_framebuffer_(0),
       copy_framebuffer_(0),
       rgba8_texture_(0),
       working_color_texture_(0),
       edges0_texture_(0),
       edges1_texture_(0),
       mini4_edge_texture_(0),
       mini4_edge_depth_texture_(0),
       edges1_shader_result_texture_float4_slot1_(0),
       edges1_shader_result_texture_(0),
       edges_combine_shader_result_texture_float4_slot1_(0),
       process_and_apply_shader_result_texture_float4_slot1_(0),
-      edges_combine_shader_result_texture_slot2_(0),
-      copy_to_image_shader_outTexture_(0) {}
+      edges_combine_shader_result_texture_slot2_(0) {}
 
 ApplyFramebufferAttachmentCMAAINTELResourceManager::
     ~ApplyFramebufferAttachmentCMAAINTELResourceManager() {
   Destroy();
 }
 
 void ApplyFramebufferAttachmentCMAAINTELResourceManager::Initialize(
     gles2::GLES2Decoder* decoder) {
   DCHECK(decoder);
   is_gles31_compatible_ =
       decoder->GetGLContext()->GetVersionInfo()->IsAtLeastGLES(3, 1);
 
-  copy_to_image_shader_ = CreateProgram("", vert_str_, copy_frag_str_);
-  copy_to_framebuffer_shader_ =
-      CreateProgram("#define OUT_FBO 1\n", vert_str_, copy_frag_str_);
-
   // Check if RGBA8UI is supported as an FBO colour target with depth.
   // If not supported, GLSL needs to convert the data to/from float so there is
   // a small extra cost.
   {
     GLuint rgba8ui_texture = 0, depth_texture = 0;
     glGenTextures(1, &rgba8ui_texture);
     glBindTexture(GL_TEXTURE_2D, rgba8ui_texture);
     glTexStorage2DEXT(GL_TEXTURE_2D, 1, GL_RGBA8UI, 4, 4);
 
     glGenTextures(1, &depth_texture);
@@ skipping 105 matching lines @@
   edges1_shader_result_texture_float4_slot1_ =
       glGetUniformLocation(edges0_shader_, "g_resultTextureFlt4Slot1");
   edges1_shader_result_texture_ =
       glGetUniformLocation(edges1_shader_, "g_resultTexture");
   edges_combine_shader_result_texture_float4_slot1_ =
       glGetUniformLocation(edges_combine_shader_, "g_resultTextureFlt4Slot1");
   edges_combine_shader_result_texture_slot2_ =
       glGetUniformLocation(edges_combine_shader_, "g_resultTextureSlot2");
   process_and_apply_shader_result_texture_float4_slot1_ = glGetUniformLocation(
       process_and_apply_shader_, "g_resultTextureFlt4Slot1");
-  copy_to_image_shader_outTexture_ =
-      glGetUniformLocation(copy_to_image_shader_, "outTexture");
 
   initialized_ = true;
 }
 
 void ApplyFramebufferAttachmentCMAAINTELResourceManager::Destroy() {
   if (!initialized_)
     return;
 
   ReleaseTextures();
 
-  glDeleteProgram(copy_to_image_shader_);
-  glDeleteProgram(copy_to_framebuffer_shader_);
   glDeleteProgram(process_and_apply_shader_);
   glDeleteProgram(edges_combine_shader_);
   glDeleteProgram(edges1_shader_);
   glDeleteProgram(edges0_shader_);
   glDeleteProgram(debug_display_edges_shader_);
 
   initialized_ = false;
 }
 
 // Apply CMAA(Conservative Morphological Anti-Aliasing) algorithm to the
 // color attachments of currently bound draw framebuffer.
 // Reference GL_INTEL_framebuffer_CMAA for details.
 void ApplyFramebufferAttachmentCMAAINTELResourceManager::
     ApplyFramebufferAttachmentCMAAINTEL(gles2::GLES2Decoder* decoder,
                                         gles2::Framebuffer* framebuffer) {
   DCHECK(decoder);
   DCHECK(initialized_);
   if (!framebuffer)
     return;
 
-  GLuint last_framebuffer = framebuffer->service_id();
+  glDisable(GL_SCISSOR_TEST);
+  glDisable(GL_STENCIL_TEST);
+  glDisable(GL_CULL_FACE);
+  glDisable(GL_BLEND);
 
   // Process each color attachment of the current draw framebuffer.
   uint32_t max_draw_buffers = decoder->GetContextGroup()->max_draw_buffers();
   for (uint32_t i = 0; i < max_draw_buffers; i++) {
     const gles2::Framebuffer::Attachment* attachment =
         framebuffer->GetAttachment(GL_COLOR_ATTACHMENT0 + i);
     if (attachment && attachment->IsTextureAttachment()) {
       // Get the texture info.
       GLuint source_texture_client_id = attachment->object_name();
       GLuint source_texture = 0;
       if (!decoder->GetServiceTextureId(source_texture_client_id,
                                         &source_texture))
         continue;
       GLsizei width = attachment->width();
       GLsizei height = attachment->height();
       GLenum internal_format = attachment->internal_format();
 
       // Resize internal structures - only if needed.
       OnSize(width, height);
 
       // CMAA internally expects GL_RGBA8 textures.
       // Process using a GL_RGBA8 copy if this is not the case.
       bool do_copy = internal_format != GL_RGBA8;
 
       // Copy source_texture to rgba8_texture_
       if (do_copy) {
-        CopyTexture(source_texture, rgba8_texture_, false);
+        glBindFramebufferEXT(GL_FRAMEBUFFER, copy_framebuffer_);
+        glFramebufferTexture2DEXT(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0,
+                                  GL_TEXTURE_2D, source_texture, 0);
+        CopyTexture(rgba8_texture_);
       }
 
       // CMAA Effect
-      glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, last_framebuffer);
       if (do_copy) {
         ApplyCMAAEffectTexture(rgba8_texture_, rgba8_texture_);
       } else {
         ApplyCMAAEffectTexture(source_texture, source_texture);
       }
 
       // Copy rgba8_texture_ to source_texture
       if (do_copy) {
-        // Move source_texture to the first color attachment of the copy fbo.
-        glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, last_framebuffer);
-        glFramebufferTexture2DEXT(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + i,
-                                  GL_TEXTURE_2D, 0, 0);
-        glBindFramebufferEXT(GL_FRAMEBUFFER, copy_framebuffer_);
-        glFramebufferTexture2DEXT(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0,
-                                  GL_TEXTURE_2D, source_texture, 0);
-
-        CopyTexture(rgba8_texture_, source_texture, true);
-
-        // Restore color attachments
         glBindFramebufferEXT(GL_FRAMEBUFFER, copy_framebuffer_);
         glFramebufferTexture2DEXT(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0,
                                   GL_TEXTURE_2D, rgba8_texture_, 0);
-        glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, last_framebuffer);
-        glFramebufferTexture2DEXT(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + i,
-                                  GL_TEXTURE_2D, source_texture, 0);
+        CopyTexture(source_texture);
       }
+      decoder->RestoreTextureState(source_texture);
     }
   }
 
   // Restore state
   decoder->RestoreAllAttributes();
   decoder->RestoreTextureUnitBindings(0);
   decoder->RestoreTextureUnitBindings(1);
   decoder->RestoreActiveTexture();
   decoder->RestoreProgramBindings();
   decoder->RestoreBufferBindings();
@@ skipping 263 matching lines @@
     glDeleteTextures(1, &edges0_texture_);
     glDeleteTextures(1, &edges1_texture_);
     glDeleteTextures(1, &mini4_edge_texture_);
     glDeleteTextures(1, &mini4_edge_depth_texture_);
     glDeleteTextures(1, &working_color_texture_);
   }
   textures_initialized_ = false;
 }
 
 void ApplyFramebufferAttachmentCMAAINTELResourceManager::CopyTexture(
-    GLint source,
-    GLint dest,
-    bool via_fbo) {
-  glViewport(0, 0, width_, height_);
+    GLint dest) {
   glActiveTexture(GL_TEXTURE0);
-  glBindTexture(GL_TEXTURE_2D, source);
+  glBindTexture(GL_TEXTURE_2D, dest);
+  glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
+  glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
+  glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+  glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
 
-  if (!via_fbo) {
-    glUseProgram(copy_to_image_shader_);
-    if (!is_gles31_compatible_) {
-      glUniform1i(copy_to_image_shader_outTexture_, 0);
-    }
-    glBindImageTextureEXT(0, dest, 0, GL_FALSE, 0, GL_WRITE_ONLY, GL_RGBA8);
-  } else {
-    glDisable(GL_DEPTH_TEST);
-    glDisable(GL_STENCIL_TEST);
-    glDisable(GL_CULL_FACE);
-    glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
-    glDepthMask(GL_FALSE);
-    glDisable(GL_BLEND);
-    glUseProgram(copy_to_framebuffer_shader_);
-  }
-
-  glDrawArrays(GL_TRIANGLES, 0, 3);
-  glUseProgram(0);
-  glBindTexture(GL_TEXTURE_2D, 0);
+  glCopyTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 0, 0, width_, height_);

[dshwang, 2016/09/02 15:38:11]

Use glCopyTexSubImage2D, because we know the source and dest texture is always
RGBA8 or RGBA or RGB, and target is always GL_TEXTURE_2D.
In addition, it speeds up. It's always fater then glDrawArrays in all platforms.

Linux (Intel IvyBridge i7-3520M): 16 us vs 6 us -> 260% faster
Android (Nexus 5): 331.8 us vs 252.7 us -> 31% faster
See https://src.chromium.org/viewvc/chrome?revision=289160&view=revision

 }
 
 GLuint ApplyFramebufferAttachmentCMAAINTELResourceManager::CreateProgram(
     const char* defines,
     const char* vs_source,
     const char* fs_source) {
   GLuint program = glCreateProgram();
 
   GLuint vs = CreateShader(GL_VERTEX_SHADER, defines, vs_source);
   GLuint fs = CreateShader(GL_FRAGMENT_SHADER, defines, fs_source);
@@ skipping 1271 matching lines @@
       CombineEdges();
     \n#endif\n
     \n#if defined BLUR_EDGES\n
       BlurEdges();
     \n#endif\n
     \n#if defined DISPLAY_EDGES\n
       DisplayEdges();
     \n#endif\n
     }
   );
-
-const char
-  ApplyFramebufferAttachmentCMAAINTELResourceManager::copy_frag_str_[] =
-    SHADER(
-      precision highp float;
-      layout(binding = 0) uniform highp sampler2D inTexture;
-      layout(location = 0) out vec4 outColor;
-      \n#ifdef GL_ES\n
-      layout(binding = 0, rgba8) restrict writeonly uniform highp
-                                                     image2D outTexture;
-      \n#else\n
-      layout(rgba8) restrict writeonly uniform highp image2D outTexture;
-      \n#endif\n
-
-      void main() {
-        ivec2 screenPosI = ivec2( gl_FragCoord.xy );
-        vec4 pixel = texelFetch(inTexture, screenPosI, 0);
-      \n#ifdef OUT_FBO\n
-        outColor = pixel;
-      \n#else\n
-        imageStore(outTexture, screenPosI, pixel);
-      \n#endif\n
-      }
-    );
 /* clang-format on */
 
 }  // namespace gles2
 }  // namespace gpu