/**
 ******************************************************************************
 * Xenia : Xbox 360 Emulator Research Project                                 *
 ******************************************************************************
 * Copyright 2022 Ben Vanik. All rights reserved.                             *
 * Released under the BSD license - see LICENSE in the root for more details. *
 ******************************************************************************
 */

#include "edram.xesli"
#include "host_depth_store.xesli"

xesl_entry
  xesl_writeTypedStorageBuffer(xesl_uint4, xe_host_depth_store_dest, set=0,
                               binding=0, u0, space0)
  xesl_entry_binding_next
  xesl_texture(xesl_texture2DMS, xe_host_depth_store_source, set=1, binding=0,
               t0, space0)
xesl_entry_bindings_end_local_size(8, 8, 1)
  xesl_input_global_invocation_id
xesl_entry_signature_end
  // 1 thread = 8 samples (4x0.5 pixels, resolve granularity is 8 pixels).
  // Group height can't cross resolve granularity, Y overflow check not needed.
  xesl_dont_flatten
  if ((xesl_GlobalInvocationID.x >> 1u) >= XeHostDepthStoreScaledWidthDiv8()) {
    return;
  }
  xesl_int2 pixel_index =
      xesl_int2(XeHostDepthStoreScaledOrigin() +
                xesl_uint2(xesl_GlobalInvocationID.x << 2u,
                           xesl_GlobalInvocationID.y >> 1u));
  // For simplicity, passing samples directly, not pixels, to XeEdramOffsetInts.
  uint edram_address_int4s =
      XeEdramOffsetInts(
          (xesl_uint2(pixel_index) << 1u) | (xesl_GlobalInvocationID.xy & 1u),
          0u, XeHostDepthStorePitchTiles(), kXenosMsaaSamples_1X, false, 0u, 0u,
          XeHostDepthStoreResolutionScale())
      >> 2u;
  // Render target horizontal sample in bit 0, vertical sample in bit 1.
  int source_sample_left = int((xesl_GlobalInvocationID.y & 1u) << 1u);
  int source_sample_right = source_sample_left + 1;
  xesl_writeTypedStorageBufferStore(
      xe_host_depth_store_dest, edram_address_int4s,
      xesl_floatBitsToUint(xesl_float4(
          xesl_texelFetch2DMS(xe_host_depth_store_source, pixel_index,
                              source_sample_left).r,
          xesl_texelFetch2DMS(xe_host_depth_store_source, pixel_index,
                              source_sample_right).r,
          xesl_texelFetch2DMS(xe_host_depth_store_source,
                              pixel_index + xesl_int2(1, 0),
                              source_sample_left).r,
          xesl_texelFetch2DMS(xe_host_depth_store_source,
                              pixel_index + xesl_int2(1, 0),
                              source_sample_right).r)));
  xesl_writeTypedStorageBufferStore(
      xe_host_depth_store_dest, edram_address_int4s + 1u,
      xesl_floatBitsToUint(xesl_float4(
          xesl_texelFetch2DMS(xe_host_depth_store_source,
                              pixel_index + xesl_int2(2, 0),
                              source_sample_left).r,
          xesl_texelFetch2DMS(xe_host_depth_store_source,
                              pixel_index + xesl_int2(2, 0),
                              source_sample_right).r,
          xesl_texelFetch2DMS(xe_host_depth_store_source,
                              pixel_index + xesl_int2(3, 0),
                              source_sample_left).r,
          xesl_texelFetch2DMS(xe_host_depth_store_source,
                              pixel_index + xesl_int2(3, 0),
                              source_sample_right).r)));
xesl_entry_end