fabi/Xenia-Canary
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[GPU/UI] XeSL readability improvements + float suffix

Browse Source 
Use the _xe suffix instead of the xesl_ prefix for quicker visual
recognition of identifiers, also switch to snake_case for consistency.

Also add the f suffix to float32 literals because the Metal Shading
Language is based on C++.
This commit is contained in:
Triang3l
2025-08-19 21:36:06 +03:00
parent 3b4b04c371
commit 04d5c40d0d
201 changed files with 90552 additions and 92369 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
src/xenia/gpu/shaders/resolve_full_32bpp.xesli
View File

@@ -12,43 +12,42 @@
#define XE_RESOLVE_COPY_EDRAM_IS_UINT_VECTOR_BUFFER
#include "resolve.xesli"
xesl_writeTypedStorageBuffer_declare(xesl_uint4, xe_resolve_dest, set=1,
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
array_buffer_wo_declare_xe(uint4_xe, xe_resolve_dest, set=1, binding=0, u0,
space0)
#define LOCAL_SIZE_X_XE 8
#define LOCAL_SIZE_Y_XE 8
#define LOCAL_SIZE_Z_XE 1
entry_bindings_begin_compute_xe
XE_RESOLVE_PUSH_CONST_BINDING
entry_binding_next_xe
array_buffer_wo_binding_xe(uint4_xe, xe_resolve_dest, buffer(1))
entry_binding_next_xe
XE_RESOLVE_COPY_EDRAM_BINDING
xesl_entry_bindings_end_inputs_begin_compute
xesl_entry_input_globalInvocationID
xesl_entry_inputs_end_code_begin_compute
entry_bindings_end_inputs_begin_compute_xe
entry_in_global_thread_id_xe
entry_inputs_end_code_begin_compute_xe
{
// 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);
XeResolveInfo resolve_info = XeResolveGetInfo(pass_push_consts_xe);
uint2_xe pixel_index = in_global_thread_id_xe.xy << uint2_xe(2u, 0u);
// Group height can't cross resolve granularity, Y overflow check not needed.
xesl_dont_flatten
dont_flatten_xe
if (pixel_index.x >= resolve_info.width_div_8_scaled << 3u) {
return;
}
xesl_float4 pixel_0, pixel_1, pixel_2, pixel_3;
float4_xe pixel_0, pixel_1, pixel_2, pixel_3;
XeResolveLoad4RGBAColors(
xesl_function_call_uintVectorBuffer(xe_resolve_edram)
xesl_function_call_next_after_uintVectorBuffer
pass_uint_vector_buffer_xe(xe_resolve_edram)
pass_next_after_uint_vector_buffer_xe
resolve_info,
XeResolveColorCopySourcePixelAddressIntsYHalfPixelOffsetFilling(
resolve_info, pixel_index),
pixel_0, pixel_1, pixel_2, pixel_3);
xesl_uint4 packed = XePack32bpp4Pixels(pixel_0, pixel_1, pixel_2, pixel_3,
resolve_info.dest_format);
xesl_dont_flatten
uint4_xe packed = XePack32bpp4Pixels(pixel_0, pixel_1, pixel_2, pixel_3,
resolve_info.dest_format);
dont_flatten_xe
if (pixel_index.x == 0u &&
resolve_info.half_pixel_offset_fill_source.x != 0u) {
if (resolve_info.half_pixel_offset_fill_source.x >= 2u) {
@@ -59,8 +58,9 @@ xesl_entry_inputs_end_code_begin_compute
}
packed.x = packed.y;
}
xesl_typedStorageBufferStore(
array_buffer_store_xe(
xe_resolve_dest,
XeResolveDestPixelAddress(resolve_info, pixel_index, 2u) >> 4u,
XeEndianSwap32(packed, resolve_info.dest_endian_128));
xesl_entry_code_end_compute
}
entry_code_end_compute_xe