kc3-lang/angle/src/libANGLE/renderer/metal/shaders/visibility.metal

• Show log

  Commit

• Hash : 5b7c5b34
  Author :
  Date : 2020-10-18T17:19:51
  

  Metal: don't precompile default shaders.
  
  Default shaders will be compiled from source at runtime, this is because
  they depend on ANGLE format table and there is currently no way to
  pre-compile metal shaders in a cross-platform manner.
  
  Using default shaders' source instead of pre-compiled form seems to
  reduce the libGLESv2's binary size. However, the startup time will be
  increased due to runtime cost of compilation, thus the compilation now
  will be done asynchronously.
  
  Bug: angleproject:5186
  Change-Id: I0e1987d6c76692d5169255736fbb8e215185c33b
  Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2482405
  Reviewed-by: Jamie Madill <jmadill@chromium.org>
  Commit-Queue: Le Hoang Quyen <le.hoang.q@gmail.com>
  

Download

• src/libANGLE/renderer/metal/shaders/visibility.metal

• //
  // Copyright 2019 The ANGLE Project. All rights reserved.
  // Use of this source code is governed by a BSD-style license that can be
  // found in the LICENSE file.
  //
  
  #include "common.h"
  
  constant bool kCombineWithExistingResult [[function_constant(1000)]];
  
  // Combine the visibility result of current render pass with previous value from previous render
  // pass
  struct CombineVisibilityResultOptions
  {
      // Start offset in the render pass's visibility buffer allocated for the query.
      uint startOffset;
      // How many offsets in the render pass's visibility buffer is used for the query?
      uint numOffsets;
  };
  
  kernel void combineVisibilityResult(uint idx [[thread_position_in_grid]],
                                      constant CombineVisibilityResultOptions &options [[buffer(0)]],
                                      constant ushort4 *renderpassVisibilityResult [[buffer(1)]],
                                      device ushort4 *finalResults [[buffer(2)]])
  {
      if (idx > 0)
      {
          // NOTE(hqle):
          // This is a bit wasteful to use a WARP of multiple threads just for combining one integer.
          // Consider a better approach.
          return;
      }
      ushort4 finalResult16x4;
  
      if (kCombineWithExistingResult)
      {
          finalResult16x4 = finalResults[0];
      }
      else
      {
          finalResult16x4 = ushort4(0, 0, 0, 0);
      }
  
      for (uint i = 0; i < options.numOffsets; ++i)
      {
          uint offset              = options.startOffset + i;
          ushort4 renderpassResult = renderpassVisibilityResult[offset];
  
          // Only boolean result is required, so bitwise OR is enough
          finalResult16x4          = finalResult16x4 | renderpassResult;
      }
      finalResults[0] = finalResult16x4;
  }
  

Download