//
// Copyright 2019 The ANGLE Project Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
//
// BlitResolveStencilNoExport.comp: Blit stencil images or resolve multisampled ones into a buffer.
// This is used where VK_EXT_shader_stencil_export is not available, to output the resolved stencil
// into a temporary buffer, which is then copied into the stencil aspect of the final image.

#version 450 core

#if IsResolve
#extension GL_EXT_samplerless_texture_functions : require
#endif

#define MAKE_SRC_RESOURCE(prefix, type) prefix ## type

#if IsResolve

#define CoordType ivec2
#if SrcIsArray
#define SRC_RESOURCE utexture2DMSArray
#define TEXEL_FETCH(src, coord, sample) texelFetch(src, ivec3(coord, params.srcLayer), sample)
#else
#define SRC_RESOURCE utexture2DMS
#define TEXEL_FETCH(src, coord, sample) texelFetch(src, coord, sample)
#endif

#else

#define CoordType vec2
#if SrcIsArray
#define SRC_RESOURCE utexture2DArray
#define TEXEL_FETCH(src, coord, sample) texture(usampler2DArray(src, blitSampler), vec3(coord * params.invSrcExtent, params.srcLayer))
#else
#define SRC_RESOURCE utexture2D
#define TEXEL_FETCH(src, coord, sample) texture(usampler2D(src, blitSampler), coord * params.invSrcExtent)
#endif

#endif  // IsResolve

layout (local_size_x = 8, local_size_y = 8, local_size_z = 1) in;

layout(push_constant) uniform PushConstants {
    // Translation from source to destination coordinates.
    CoordType offset;
    vec2 stretch;
    vec2 invSrcExtent;
    int srcLayer;
    int srcWidth;
    // Output.
    ivec4 blitArea;
    int destPitch;
    // Flip control.
    bool flipX;
    bool flipY;
} params;

layout (set = 0, binding = 0) buffer dest
{
    // Note: every invocation handles 4 stencil value and output one value here.
    uint destData[];
};

layout(set = 0, binding = 1) uniform SRC_RESOURCE stencil;

#if !IsResolve
layout(set = 0, binding = 2) uniform sampler blitSampler;
#endif

void main()
{
    ivec2 destSubImageCoords = ivec2(gl_GlobalInvocationID.x * 4, gl_GlobalInvocationID.y);
    if (any(lessThanEqual(params.blitArea.zw, destSubImageCoords)))
    {
        return;
    }

    // See comment in BlitResolve.frag regarding how the following coordinate transformation is
    // derived.
    //
    // When blitting with the fragment shader, the input coordinates are within the blit area.
    // With the compute shader however, the input coordinates start from 0,0.  That is, if the
    // fragment shader input is Xf and the compute shader's input is Xc, we have:
    //
    //     Xf = Xc + params.blitOffset
    //
    // So we need to offset the input coordinates here by the blit area offset.
    destSubImageCoords += params.blitArea.xy;
    CoordType srcImageCoords = CoordType(destSubImageCoords);
#if !IsResolve
    srcImageCoords *= params.stretch;
#endif
    srcImageCoords -= params.offset;

    // If flipping, negate the coordinates.
    if (params.flipX)
        srcImageCoords.x = -srcImageCoords.x;
    if (params.flipY)
        srcImageCoords.y = -srcImageCoords.y;

    int xDir = params.flipX ? -1 : 1;

    uint outStencils = 0;

    for (int i = 0; i < 4; ++i)
    {
        // Bounds check on X:
        if (srcImageCoords.x >= 0 && srcImageCoords.x < params.srcWidth)
        {
            // Note: always resolve using sample 0.  GLES3 gives us freedom in choosing how
            // to resolve depth/stencil images.
            uint stencilValue = TEXEL_FETCH(stencil, srcImageCoords, 0).x;

#if IsBigEndian
            outStencils |= (stencilValue & 0xFF) << ((3 - i) * 8);
#else
            outStencils |= (stencilValue & 0xFF) << (i * 8);
#endif
        }

        srcImageCoords.x += xDir;
    }

    destData[gl_GlobalInvocationID.y * params.destPitch + gl_GlobalInvocationID.x] = outStencils;
}