[GPU] XeSL resolve shaders + host depth store width fix

2022-06-19 17:50:21 +03:00
parent 166be463be
commit 9b83d3d0f4
117 changed files with 108477 additions and 55993 deletions
--- a/src/xenia/gpu/shaders/resolve_full_32bpp.xesli
+++ b/src/xenia/gpu/shaders/resolve_full_32bpp.xesli
@@ -0,0 +1,58 @@
+/**
+ ******************************************************************************
+ * 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 "endian.xesli"
+#include "pixel_formats.xesli"
+#define XE_RESOLVE_SOURCE_IS_UINT_VECTOR_BUFFER
+#include "resolve.xesli"
+
+xesl_writeTypedStorageBuffer_declare(xesl_uint4, xe_resolve_dest, set=0,
+                                     binding=0, u0, space0)
+#define xesl_localSize_x 8
+#define xesl_localSize_y 8
+#define xesl_localSize_z 1
+xesl_entry_bindings_begin_compute
+  XE_RESOLVE_PUSH_CONSTANTS_BINDING
+  xesl_entry_binding_next
+  xesl_writeTypedStorageBuffer_binding(xesl_uint4, xe_resolve_dest,
+                                       buffer(1))
+  xesl_entry_binding_next
+  XE_RESOLVE_SOURCE_BINDING
+xesl_entry_bindings_end_inputs_begin_compute
+  xesl_entry_input_globalInvocationID
+xesl_entry_inputs_end_code_begin_compute
+  // 1 thread = 4 host pixels.
+  // 1 pixel per thread is 160% as slow on Nvidia Pascal, 8 pixels per thread is
+  // 115% as slow.
+  XeResolveInfo resolve_info =
+      XeResolveGetInfo(xesl_function_call_pushConstants);
+  xesl_uint2 pixel_index = xesl_GlobalInvocationID.xy << xesl_uint2(2u, 0u);
+  // Group height can't cross resolve granularity, Y overflow check not needed.
+  xesl_dont_flatten
+  if (pixel_index.x >= resolve_info.width_div_8_scaled << 3u) {
+    return;
+  }
+  xesl_float4 pixel_0, pixel_1, pixel_2, pixel_3;
+  XeResolveLoad4RGBAColors(
+      xesl_function_call_uintVectorBuffer(xe_resolve_source)
+      xesl_function_call_next_after_uintVectorBuffer
+      resolve_info,
+      XeResolveColorCopySourcePixelAddressIntsYDuplicating(resolve_info,
+                                                           pixel_index),
+      pixel_0, pixel_1, pixel_2, pixel_3);
+  if (resolve_info.duplicate_second_host_pixel.x && pixel_index.x == 0u) {
+    pixel_0 = pixel_1;
+  }
+  xesl_writeTypedStorageBufferStore(
+      xe_resolve_dest,
+      XeResolveDestPixelAddress(resolve_info, pixel_index, 2u) >> 4u,
+      XeEndianSwap32(XePack32bpp4Pixels(pixel_0, pixel_1, pixel_2, pixel_3,
+                                        resolve_info.dest_format),
+                     resolve_info.dest_endian_128));
+xesl_entry_code_end_compute