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

[GPU/WGF] Use ByteAddressBuffer in precompiled shaders

Browse Source 
Drivers and hardware may provide more optimal paths for untyped buffers
compared to typed.

Also, ByteAddressBuffer may be bound via a root descriptor in Direct3D 12,
greatly simplifying the binding logic.

ByteAddressBuffer also doesn't have the 128 * 2^20 element count limit that
typed and structured buffers have, and doesn't require the structure stride
to be specified at resource creation time in Direct3D 11.
This commit is contained in:
Triang3l
2026-02-09 21:29:18 +03:00
parent c7f61342d7
commit 80dad41454
211 changed files with 121690 additions and 123176 deletions
Download Patch File Download Diff File Expand all files Collapse all files

299
src/xenia/gpu/shaders/resolve.xesli
View File

@@ -44,18 +44,22 @@ push_const_end_xe
#define XE_RESOLVE_PUSH_CONST_BINDING push_const_binding_xe(buffer(0))
#ifndef XE_RESOLVE_CLEAR
#ifdef XE_RESOLVE_COPY_EDRAM_IS_UINT_VECTOR_BUFFER
uint_vector_buffer_declare_xe(xe_resolve_edram, set=0, binding=0, t0,
space0)
#define XE_RESOLVE_COPY_EDRAM_BINDING \
uint_vector_buffer_binding_xe(xe_resolve_edram, buffer(2))
#else
array_buffer_declare_xe(XE_RESOLVE_SOURCE_TYPE, xe_resolve_edram, set=0,
binding=0, t0, space0)
#define XE_RESOLVE_COPY_EDRAM_BINDING \
array_buffer_binding_xe(XE_RESOLVE_SOURCE_TYPE, xe_resolve_edram, \
buffer(2))
#if XE_RESOLVE_COPY_EDRAM_BYTE_BUFFER_ALIGNMENT == 4
#define XE_RESOLVE_COPY_EDRAM_BYTE_BUFFER_DECLARE_MACRO \
byte_buffer_align4_declare_xe
#elif XE_RESOLVE_COPY_EDRAM_BYTE_BUFFER_ALIGNMENT == 8
#define XE_RESOLVE_COPY_EDRAM_BYTE_BUFFER_DECLARE_MACRO \
byte_buffer_align8_declare_xe
#elif XE_RESOLVE_COPY_EDRAM_BYTE_BUFFER_ALIGNMENT == 16
#define XE_RESOLVE_COPY_EDRAM_BYTE_BUFFER_DECLARE_MACRO \
byte_buffer_align16_declare_xe
#endif
#ifdef XE_RESOLVE_COPY_EDRAM_BYTE_BUFFER_DECLARE_MACRO
XE_RESOLVE_COPY_EDRAM_BYTE_BUFFER_DECLARE_MACRO(xe_resolve_edram, set=0,
binding=0, t0, space0)
#endif
#define XE_RESOLVE_COPY_EDRAM_BINDING \
byte_buffer_binding_xe(xe_resolve_edram, buffer(2))
#endif
struct XeResolveInfo {
@@ -153,8 +157,8 @@ XeResolveInfo XeResolveGetInfo(param_push_consts_xe) {
return resolve_info;
}
uint XeResolveEdramPixelStrideInts(XeResolveInfo resolve_info) {
return 1u << (resolve_info.edram_format_ints_log2 +
uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
return 4u << (resolve_info.edram_format_ints_log2 +
uint(resolve_info.edram_msaa_samples >= kXenosMsaaSamples_4X));
}
@@ -231,9 +235,9 @@ uint XeResolveEdramPixelStrideInts(XeResolveInfo resolve_info) {
// Offset to the first sample to participate in averaging (or the sample to be
// copied if not averaging).
uint XeResolveColorCopySourcePixelAddressIntsYHalfPixelOffsetFilling(
uint XeResolveColorCopySourcePixelAddressBytesYHalfPixelOffsetFilling(
XeResolveInfo resolve_info, uint2_xe pixel_index) {
return XeEdramOffsetInts(
return XeEdramOffsetBytes(
uint2_xe(pixel_index.x,
max(pixel_index.y,
resolve_info.half_pixel_offset_fill_source.y)) +
@@ -457,116 +461,114 @@ uint XeResolveEdramPixelStrideInts(XeResolveInfo resolve_info) {
}
}
#ifdef XE_RESOLVE_COPY_EDRAM_IS_UINT_VECTOR_BUFFER
#if XE_RESOLVE_COPY_EDRAM_BYTE_BUFFER_ALIGNMENT == 4
void XeResolveLoad2RGBAUnswappedPixelSamplesFromRaw(
param_uint_vector_buffer_xe(xe_resolve_edram)
param_next_after_uint_vector_buffer_xe
uint sample_address_ints, uint pixel_stride_ints, uint format_ints_log2,
uint format, out_param_xe(float4_xe, pixel_0),
param_byte_buffer_xe(xe_resolve_edram) param_next_after_byte_buffer_xe
uint sample_address_bytes, uint pixel_stride_bytes,
uint format_ints_log2, uint format, out_param_xe(float4_xe, pixel_0),
out_param_xe(float4_xe, pixel_1)) {
dont_flatten_xe if (format_ints_log2 != 0u) {
uint4_xe packed;
dont_flatten_xe if (pixel_stride_ints == 2u) {
packed = uint_vector_buffer_load4_xe(
xe_resolve_edram, sample_address_ints);
dont_flatten_xe if (pixel_stride_bytes == 8u) {
packed = byte_buffer_align4_load16u_xe(
xe_resolve_edram, sample_address_bytes);
} else {
packed.xy = uint_vector_buffer_load2_xe(
xe_resolve_edram, sample_address_ints);
packed.zw = uint_vector_buffer_load2_xe(
xe_resolve_edram, sample_address_ints + pixel_stride_ints);
packed.xy = byte_buffer_align4_load8u_xe(
xe_resolve_edram, sample_address_bytes);
packed.zw = byte_buffer_align4_load8u_xe(
xe_resolve_edram, sample_address_bytes + pixel_stride_bytes);
}
XeResolveUnpack64bpp2Samples(packed, format, pixel_0, pixel_1);
} else {
uint2_xe packed;
dont_flatten_xe if (pixel_stride_ints == 1u) {
packed = uint_vector_buffer_load2_xe(
xe_resolve_edram, sample_address_ints);
dont_flatten_xe if (pixel_stride_bytes == 4u) {
packed = byte_buffer_align4_load8u_xe(
xe_resolve_edram, sample_address_bytes);
} else {
packed.x = uint_vector_buffer_load1_xe(
xe_resolve_edram, sample_address_ints);
packed.y = uint_vector_buffer_load1_xe(
xe_resolve_edram, sample_address_ints + pixel_stride_ints);
packed.x = byte_buffer_align4_load4_xe(
xe_resolve_edram, sample_address_bytes);
packed.y = byte_buffer_align4_load4_xe(
xe_resolve_edram, sample_address_bytes + pixel_stride_bytes);
}
XeResolveUnpack32bpp2Samples(packed, format, pixel_0, pixel_1);
}
}
void XeResolveLoad4RGBAUnswappedPixelSamplesFromRaw(
param_uint_vector_buffer_xe(xe_resolve_edram)
param_next_after_uint_vector_buffer_xe
uint sample_address_ints, uint pixel_stride_ints, uint format_ints_log2,
uint format, out_param_xe(float4_xe, pixel_0),
param_byte_buffer_xe(xe_resolve_edram) param_next_after_byte_buffer_xe
uint sample_address_bytes, uint pixel_stride_bytes,
uint format_ints_log2, uint format, out_param_xe(float4_xe, pixel_0),
out_param_xe(float4_xe, pixel_1), out_param_xe(float4_xe, pixel_2),
out_param_xe(float4_xe, pixel_3)) {
dont_flatten_xe if (format_ints_log2 != 0u) {
uint4_xe packed_01, packed_23;
dont_flatten_xe if (pixel_stride_ints == 2u) {
packed_01 = uint_vector_buffer_load4_xe(
xe_resolve_edram, sample_address_ints);
packed_23 = uint_vector_buffer_load4_xe(
xe_resolve_edram, sample_address_ints + 4u);
dont_flatten_xe if (pixel_stride_bytes == 8u) {
packed_01 = byte_buffer_align4_load16u_xe(
xe_resolve_edram, sample_address_bytes);
packed_23 = byte_buffer_align4_load16u_xe(
xe_resolve_edram, sample_address_bytes + 16u);
} else {
packed_01.xy = uint_vector_buffer_load2_xe(
xe_resolve_edram, sample_address_ints);
packed_01.zw = uint_vector_buffer_load2_xe(
xe_resolve_edram, sample_address_ints + pixel_stride_ints);
packed_23.xy = uint_vector_buffer_load2_xe(
xe_resolve_edram, sample_address_ints + 2u * pixel_stride_ints);
packed_23.zw = uint_vector_buffer_load2_xe(
xe_resolve_edram, sample_address_ints + 3u * pixel_stride_ints);
packed_01.xy = byte_buffer_align4_load8u_xe(
xe_resolve_edram, sample_address_bytes);
packed_01.zw = byte_buffer_align4_load8u_xe(
xe_resolve_edram, sample_address_bytes + pixel_stride_bytes);
packed_23.xy = byte_buffer_align4_load8u_xe(
xe_resolve_edram, sample_address_bytes + 2u * pixel_stride_bytes);
packed_23.zw = byte_buffer_align4_load8u_xe(
xe_resolve_edram, sample_address_bytes + 3u * pixel_stride_bytes);
}
XeResolveUnpack64bpp4Samples(packed_01, packed_23, format, pixel_0,
pixel_1, pixel_2, pixel_3);
} else {
uint4_xe packed;
dont_flatten_xe if (pixel_stride_ints == 1u) {
packed = uint_vector_buffer_load4_xe(
xe_resolve_edram, sample_address_ints);
dont_flatten_xe if (pixel_stride_bytes == 4u) {
packed = byte_buffer_align4_load16u_xe(
xe_resolve_edram, sample_address_bytes);
} else {
packed.x = uint_vector_buffer_load1_xe(
xe_resolve_edram, sample_address_ints);
packed.y = uint_vector_buffer_load1_xe(
xe_resolve_edram, sample_address_ints + pixel_stride_ints);
packed.z = uint_vector_buffer_load1_xe(
xe_resolve_edram, sample_address_ints + 2u * pixel_stride_ints);
packed.w = uint_vector_buffer_load1_xe(
xe_resolve_edram, sample_address_ints + 3u * pixel_stride_ints);
packed.x = byte_buffer_align4_load4_xe(
xe_resolve_edram, sample_address_bytes);
packed.y = byte_buffer_align4_load4_xe(
xe_resolve_edram, sample_address_bytes + pixel_stride_bytes);
packed.z = byte_buffer_align4_load4_xe(
xe_resolve_edram, sample_address_bytes + 2u * pixel_stride_bytes);
packed.w = byte_buffer_align4_load4_xe(
xe_resolve_edram, sample_address_bytes + 3u * pixel_stride_bytes);
}
XeResolveUnpack32bpp4Samples(packed, format, pixel_0, pixel_1, pixel_2,
pixel_3);
}
}
// For red/blue swapping for 64bpp, pre-add 1 to sample_address_ints.
// For red/blue swapping for 64bpp, pre-add 4 to sample_address_bytes.
void XeResolveLoad8RedPixelSamplesFromRaw(
param_uint_vector_buffer_xe(xe_resolve_edram)
param_next_after_uint_vector_buffer_xe
uint sample_address_ints, uint pixel_stride_ints, uint format_ints_log2,
uint format, bool swap_32bpp, out_param_xe(float4_xe, pixels_0123),
param_byte_buffer_xe(xe_resolve_edram) param_next_after_byte_buffer_xe
uint sample_address_bytes, uint pixel_stride_bytes,
uint format_ints_log2, uint format, bool swap_32bpp,
out_param_xe(float4_xe, pixels_0123),
out_param_xe(float4_xe, pixels_4567)) {
uint4_xe packed_0123, packed_4567;
dont_flatten_xe if (pixel_stride_ints == 1u) {
packed_0123 = uint_vector_buffer_load4_xe(
xe_resolve_edram, sample_address_ints);
packed_4567 = uint_vector_buffer_load4_xe(
xe_resolve_edram, sample_address_ints + 4u);
dont_flatten_xe if (pixel_stride_bytes == 4u) {
packed_0123 = byte_buffer_align4_load16u_xe(
xe_resolve_edram, sample_address_bytes);
packed_4567 = byte_buffer_align4_load16u_xe(
xe_resolve_edram, sample_address_bytes + 16u);
} else {
packed_0123.x = uint_vector_buffer_load1_xe(
xe_resolve_edram, sample_address_ints);
packed_0123.y = uint_vector_buffer_load1_xe(
xe_resolve_edram, sample_address_ints + pixel_stride_ints);
packed_0123.z = uint_vector_buffer_load1_xe(
xe_resolve_edram, sample_address_ints + 2u * pixel_stride_ints);
packed_0123.w = uint_vector_buffer_load1_xe(
xe_resolve_edram, sample_address_ints + 3u * pixel_stride_ints);
packed_4567.x = uint_vector_buffer_load1_xe(
xe_resolve_edram, sample_address_ints + 4u * pixel_stride_ints);
packed_4567.y = uint_vector_buffer_load1_xe(
xe_resolve_edram, sample_address_ints + 5u * pixel_stride_ints);
packed_4567.z = uint_vector_buffer_load1_xe(
xe_resolve_edram, sample_address_ints + 6u * pixel_stride_ints);
packed_4567.w = uint_vector_buffer_load1_xe(
xe_resolve_edram, sample_address_ints + 7u * pixel_stride_ints);
packed_0123.x = byte_buffer_align4_load4_xe(
xe_resolve_edram, sample_address_bytes);
packed_0123.y = byte_buffer_align4_load4_xe(
xe_resolve_edram, sample_address_bytes + pixel_stride_bytes);
packed_0123.z = byte_buffer_align4_load4_xe(
xe_resolve_edram, sample_address_bytes + 2u * pixel_stride_bytes);
packed_0123.w = byte_buffer_align4_load4_xe(
xe_resolve_edram, sample_address_bytes + 3u * pixel_stride_bytes);
packed_4567.x = byte_buffer_align4_load4_xe(
xe_resolve_edram, sample_address_bytes + 4u * pixel_stride_bytes);
packed_4567.y = byte_buffer_align4_load4_xe(
xe_resolve_edram, sample_address_bytes + 5u * pixel_stride_bytes);
packed_4567.z = byte_buffer_align4_load4_xe(
xe_resolve_edram, sample_address_bytes + 6u * pixel_stride_bytes);
packed_4567.w = byte_buffer_align4_load4_xe(
xe_resolve_edram, sample_address_bytes + 7u * pixel_stride_bytes);
}
dont_flatten_xe if (format_ints_log2 != 0u) {
XeResolveUnpack64bpp8RedUnswappedSamples(packed_0123, packed_4567,
@@ -579,48 +581,46 @@ uint XeResolveEdramPixelStrideInts(XeResolveInfo resolve_info) {
}
void XeResolveLoad2RGBAColors(
param_uint_vector_buffer_xe(xe_resolve_edram)
param_next_after_uint_vector_buffer_xe
XeResolveInfo resolve_info, uint address_ints,
param_byte_buffer_xe(xe_resolve_edram) param_next_after_byte_buffer_xe
XeResolveInfo resolve_info, uint address_bytes,
out_param_xe(float4_xe, pixel_0), out_param_xe(float4_xe, pixel_1)) {
uint pixel_stride_ints = XeResolveEdramPixelStrideInts(resolve_info);
uint pixel_stride_bytes = XeResolveEdramPixelStrideBytes(resolve_info);
XeResolveLoad2RGBAUnswappedPixelSamplesFromRaw(
pass_uint_vector_buffer_xe(xe_resolve_edram)
pass_next_after_uint_vector_buffer_xe
address_ints, pixel_stride_ints, resolve_info.edram_format_ints_log2,
resolve_info.edram_format, pixel_0, pixel_1);
pass_byte_buffer_xe(xe_resolve_edram) pass_next_after_byte_buffer_xe
address_bytes, pixel_stride_bytes,
resolve_info.edram_format_ints_log2, resolve_info.edram_format,
pixel_0, pixel_1);
float exp_bias = resolve_info.dest_exp_bias_factor;
dont_flatten_xe
if (resolve_info.sample_select >= kXenosCopySampleSelect_01) {
uint tile_row_stride_ints = 80u * resolve_info.resolution_scale.x;
uint tile_row_stride_bytes = 4u * 80u * resolve_info.resolution_scale.x;
// TODO(Triang3l): Gamma-correct resolve for 8_8_8_8_GAMMA.
exp_bias *= 0.5f;
float4_xe msaa_resolve_pixel_0, msaa_resolve_pixel_1;
XeResolveLoad2RGBAUnswappedPixelSamplesFromRaw(
pass_uint_vector_buffer_xe(xe_resolve_edram)
pass_next_after_uint_vector_buffer_xe
address_ints + tile_row_stride_ints, pixel_stride_ints,
pass_byte_buffer_xe(xe_resolve_edram) pass_next_after_byte_buffer_xe
address_bytes + tile_row_stride_bytes, pixel_stride_bytes,
resolve_info.edram_format_ints_log2, resolve_info.edram_format,
msaa_resolve_pixel_0, msaa_resolve_pixel_1);
pixel_0 += msaa_resolve_pixel_0;
pixel_1 += msaa_resolve_pixel_1;
dont_flatten_xe
if (resolve_info.sample_select >= kXenosCopySampleSelect_0123) {
uint sample_stride_ints = 1u << resolve_info.edram_format_ints_log2;
uint sample_stride_bytes = 4u << resolve_info.edram_format_ints_log2;
exp_bias *= 0.5f;
XeResolveLoad2RGBAUnswappedPixelSamplesFromRaw(
pass_uint_vector_buffer_xe(xe_resolve_edram)
pass_next_after_uint_vector_buffer_xe
address_ints + sample_stride_ints, pixel_stride_ints,
pass_byte_buffer_xe(xe_resolve_edram)
pass_next_after_byte_buffer_xe
address_bytes + sample_stride_bytes, pixel_stride_bytes,
resolve_info.edram_format_ints_log2, resolve_info.edram_format,
msaa_resolve_pixel_0, msaa_resolve_pixel_1);
pixel_0 += msaa_resolve_pixel_0;
pixel_1 += msaa_resolve_pixel_1;
XeResolveLoad2RGBAUnswappedPixelSamplesFromRaw(
pass_uint_vector_buffer_xe(xe_resolve_edram)
pass_next_after_uint_vector_buffer_xe
address_ints + tile_row_stride_ints + sample_stride_ints,
pixel_stride_ints, resolve_info.edram_format_ints_log2,
pass_byte_buffer_xe(xe_resolve_edram)
pass_next_after_byte_buffer_xe
address_bytes + tile_row_stride_bytes + sample_stride_bytes,
pixel_stride_bytes, resolve_info.edram_format_ints_log2,
resolve_info.edram_format, msaa_resolve_pixel_0,
msaa_resolve_pixel_1);
pixel_0 += msaa_resolve_pixel_0;
@@ -636,21 +636,20 @@ uint XeResolveEdramPixelStrideInts(XeResolveInfo resolve_info) {
}
void XeResolveLoad4RGBAColors(
param_uint_vector_buffer_xe(xe_resolve_edram)
param_next_after_uint_vector_buffer_xe
XeResolveInfo resolve_info, uint address_ints,
param_byte_buffer_xe(xe_resolve_edram) param_next_after_byte_buffer_xe
XeResolveInfo resolve_info, uint address_bytes,
out_param_xe(float4_xe, pixel_0), out_param_xe(float4_xe, pixel_1),
out_param_xe(float4_xe, pixel_2), out_param_xe(float4_xe, pixel_3)) {
uint pixel_stride_ints = XeResolveEdramPixelStrideInts(resolve_info);
uint pixel_stride_bytes = XeResolveEdramPixelStrideBytes(resolve_info);
XeResolveLoad4RGBAUnswappedPixelSamplesFromRaw(
pass_uint_vector_buffer_xe(xe_resolve_edram)
pass_next_after_uint_vector_buffer_xe
address_ints, pixel_stride_ints, resolve_info.edram_format_ints_log2,
resolve_info.edram_format, pixel_0, pixel_1, pixel_2, pixel_3);
pass_byte_buffer_xe(xe_resolve_edram) pass_next_after_byte_buffer_xe
address_bytes, pixel_stride_bytes,
resolve_info.edram_format_ints_log2, resolve_info.edram_format,
pixel_0, pixel_1, pixel_2, pixel_3);
float exp_bias = resolve_info.dest_exp_bias_factor;
dont_flatten_xe
if (resolve_info.sample_select >= kXenosCopySampleSelect_01) {
uint tile_row_stride_ints = 80u * resolve_info.resolution_scale.x;
uint tile_row_stride_bytes = 4u * 80u * resolve_info.resolution_scale.x;
// TODO(Triang3l): Gamma-correct resolve for 8_8_8_8_GAMMA.
exp_bias *= 0.5f;
float4_xe msaa_resolve_pixel_0;
@@ -658,9 +657,8 @@ uint XeResolveEdramPixelStrideInts(XeResolveInfo resolve_info) {
float4_xe msaa_resolve_pixel_2;
float4_xe msaa_resolve_pixel_3;
XeResolveLoad4RGBAUnswappedPixelSamplesFromRaw(
pass_uint_vector_buffer_xe(xe_resolve_edram)
pass_next_after_uint_vector_buffer_xe
address_ints + tile_row_stride_ints, pixel_stride_ints,
pass_byte_buffer_xe(xe_resolve_edram) pass_next_after_byte_buffer_xe
address_bytes + tile_row_stride_bytes, pixel_stride_bytes,
resolve_info.edram_format_ints_log2, resolve_info.edram_format,
msaa_resolve_pixel_0, msaa_resolve_pixel_1, msaa_resolve_pixel_2,
msaa_resolve_pixel_3);
@@ -670,12 +668,12 @@ uint XeResolveEdramPixelStrideInts(XeResolveInfo resolve_info) {
pixel_3 += msaa_resolve_pixel_3;
dont_flatten_xe
if (resolve_info.sample_select >= kXenosCopySampleSelect_0123) {
uint sample_stride_ints = 1u << resolve_info.edram_format_ints_log2;
uint sample_stride_bytes = 4u << resolve_info.edram_format_ints_log2;
exp_bias *= 0.5f;
XeResolveLoad4RGBAUnswappedPixelSamplesFromRaw(
pass_uint_vector_buffer_xe(xe_resolve_edram)
pass_next_after_uint_vector_buffer_xe
address_ints + sample_stride_ints, pixel_stride_ints,
pass_byte_buffer_xe(xe_resolve_edram)
pass_next_after_byte_buffer_xe
address_bytes + sample_stride_bytes, pixel_stride_bytes,
resolve_info.edram_format_ints_log2, resolve_info.edram_format,
msaa_resolve_pixel_0, msaa_resolve_pixel_1, msaa_resolve_pixel_2,
msaa_resolve_pixel_3);
@@ -684,10 +682,10 @@ uint XeResolveEdramPixelStrideInts(XeResolveInfo resolve_info) {
pixel_2 += msaa_resolve_pixel_2;
pixel_3 += msaa_resolve_pixel_3;
XeResolveLoad4RGBAUnswappedPixelSamplesFromRaw(
pass_uint_vector_buffer_xe(xe_resolve_edram)
pass_next_after_uint_vector_buffer_xe
address_ints + tile_row_stride_ints + sample_stride_ints,
pixel_stride_ints, resolve_info.edram_format_ints_log2,
pass_byte_buffer_xe(xe_resolve_edram)
pass_next_after_byte_buffer_xe
address_bytes + tile_row_stride_bytes + sample_stride_bytes,
pixel_stride_bytes, resolve_info.edram_format_ints_log2,
resolve_info.edram_format, msaa_resolve_pixel_0,
msaa_resolve_pixel_1, msaa_resolve_pixel_2, msaa_resolve_pixel_3);
pixel_0 += msaa_resolve_pixel_0;
@@ -709,33 +707,30 @@ uint XeResolveEdramPixelStrideInts(XeResolveInfo resolve_info) {
}
void XeResolveLoad8RedColors(
param_uint_vector_buffer_xe(xe_resolve_edram)
param_next_after_uint_vector_buffer_xe
XeResolveInfo resolve_info, uint address_ints,
param_byte_buffer_xe(xe_resolve_edram) param_next_after_byte_buffer_xe
XeResolveInfo resolve_info, uint address_bytes,
out_param_xe(float4_xe, pixels_0123),
out_param_xe(float4_xe, pixels_4567)) {
uint pixel_stride_ints = XeResolveEdramPixelStrideInts(resolve_info);
if (resolve_info.dest_swap) {
uint pixel_stride_bytes = XeResolveEdramPixelStrideBytes(resolve_info);
if (resolve_info.dest_swap && resolve_info.edram_format_ints_log2 != 0u) {
// Likely want to load the blue part from the right half for 64bpp.
address_ints += resolve_info.edram_format_ints_log2;
address_bytes += 4u;
}
XeResolveLoad8RedPixelSamplesFromRaw(
pass_uint_vector_buffer_xe(xe_resolve_edram)
pass_next_after_uint_vector_buffer_xe
address_ints, pixel_stride_ints, resolve_info.edram_format_ints_log2,
resolve_info.edram_format, resolve_info.dest_swap, pixels_0123,
pixels_4567);
pass_byte_buffer_xe(xe_resolve_edram) pass_next_after_byte_buffer_xe
address_bytes, pixel_stride_bytes,
resolve_info.edram_format_ints_log2, resolve_info.edram_format,
resolve_info.dest_swap, pixels_0123, pixels_4567);
float exp_bias = resolve_info.dest_exp_bias_factor;
dont_flatten_xe
if (resolve_info.sample_select >= kXenosCopySampleSelect_01) {
uint tile_row_stride_ints = 80u * resolve_info.resolution_scale.x;
uint tile_row_stride_bytes = 4u * 80u * resolve_info.resolution_scale.x;
// TODO(Triang3l): Gamma-correct resolve for 8_8_8_8_GAMMA.
exp_bias *= 0.5f;
float4_xe msaa_resolve_pixels_0123, msaa_resolve_pixels_4567;
XeResolveLoad8RedPixelSamplesFromRaw(
pass_uint_vector_buffer_xe(xe_resolve_edram)
pass_next_after_uint_vector_buffer_xe
address_ints + tile_row_stride_ints, pixel_stride_ints,
pass_byte_buffer_xe(xe_resolve_edram) pass_next_after_byte_buffer_xe
address_bytes + tile_row_stride_bytes, pixel_stride_bytes,
resolve_info.edram_format_ints_log2, resolve_info.edram_format,
resolve_info.dest_swap, msaa_resolve_pixels_0123,
msaa_resolve_pixels_4567);
@@ -743,22 +738,22 @@ uint XeResolveEdramPixelStrideInts(XeResolveInfo resolve_info) {
pixels_4567 += msaa_resolve_pixels_4567;
dont_flatten_xe
if (resolve_info.sample_select >= kXenosCopySampleSelect_0123) {
uint sample_stride_ints = 1u << resolve_info.edram_format_ints_log2;
uint sample_stride_bytes = 4u << resolve_info.edram_format_ints_log2;
exp_bias *= 0.5f;
XeResolveLoad8RedPixelSamplesFromRaw(
pass_uint_vector_buffer_xe(xe_resolve_edram)
pass_next_after_uint_vector_buffer_xe
address_ints + sample_stride_ints, pixel_stride_ints,
pass_byte_buffer_xe(xe_resolve_edram)
pass_next_after_byte_buffer_xe
address_bytes + sample_stride_bytes, pixel_stride_bytes,
resolve_info.edram_format_ints_log2, resolve_info.edram_format,
resolve_info.dest_swap, msaa_resolve_pixels_0123,
msaa_resolve_pixels_4567);
pixels_0123 += msaa_resolve_pixels_0123;
pixels_4567 += msaa_resolve_pixels_4567;
XeResolveLoad8RedPixelSamplesFromRaw(
pass_uint_vector_buffer_xe(xe_resolve_edram)
pass_next_after_uint_vector_buffer_xe
address_ints + tile_row_stride_ints + sample_stride_ints,
pixel_stride_ints, resolve_info.edram_format_ints_log2,
pass_byte_buffer_xe(xe_resolve_edram)
pass_next_after_byte_buffer_xe
address_bytes + tile_row_stride_bytes + sample_stride_bytes,
pixel_stride_bytes, resolve_info.edram_format_ints_log2,
resolve_info.edram_format, resolve_info.dest_swap,
msaa_resolve_pixels_0123, msaa_resolve_pixels_4567);
pixels_0123 += msaa_resolve_pixels_0123;
@@ -768,7 +763,7 @@ uint XeResolveEdramPixelStrideInts(XeResolveInfo resolve_info) {
pixels_0123 *= exp_bias;
pixels_4567 *= exp_bias;
}
#endif // XE_RESOLVE_COPY_EDRAM_IS_UINT_VECTOR_BUFFER
#endif // XE_RESOLVE_COPY_EDRAM_BYTE_BUFFER_ALIGNMENT == 4
uint4_xe XeResolveSwapRedBlue_8_8_8_8(uint4_xe pixels) {
return (pixels & ~0xFF00FFu) | ((pixels & 0xFFu) << 16u) |