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

[GPU] Texture integer scaling fetches; 8_8_8_8_GAMMA resolve

Browse Source 
Integer num_format fetches now get their scale CPU-side instead of trying to guess in the shader. This is authoritative; no cvar. Both translators now use texture_integer_scale_bits after signs/gamma and before exponent bias. This alone fixes black screens and a bunch of rendering bugs across at least several dozen titles.

This new information lives in the updated FormatInfo table, including fixed component widths.

Resolve also has two new fixes.

8_8_8_8_GAMMA EDRAM sources can now decode through the PWL curve while still in linear space, before MSAA resolve or format conversion, then re-encoded for gamma destinations. This is also now default enabled via gamma_decode_pwl_resolve and has improved blowout in at least 4 titles, with no obvious regressions thus far.

Full resolve can also now pack fixed destinations according to copy_dest_number.
This commit is contained in:
goldislead
2026-06-30 20:40:19 -07:00
committed by Radosław Gliński
parent 573cce0fe2
commit d119505289
24 changed files with 669 additions and 156 deletions
Download Patch File Download Diff File Expand all files Collapse all files

180
src/xenia/gpu/shaders/resolve.xesli
View File

@@ -70,6 +70,7 @@ struct XeResolveInfo {
uint edram_base_tiles;
uint edram_format;
uint edram_format_ints_log2;
bool decode_pwl_gamma;
uint2_xe resolution_scale;
uint2_xe half_pixel_offset_fill_source;
uint2_xe edram_offset_scaled;
@@ -81,6 +82,7 @@ struct XeResolveInfo {
bool dest_is_array;
uint dest_slice;
uint dest_format;
uint dest_num_format;
float dest_exp_bias_factor;
bool dest_swap;
uint dest_row_pitch_macro_tiles;
@@ -101,6 +103,7 @@ XeResolveInfo XeResolveGetInfo(param_push_consts_xe) {
resolve_info.edram_base_tiles = (edram_info >> 13u) & ((1u << 11u) - 1u);
resolve_info.edram_format = (edram_info >> 24u) & ((1u << 4u) - 1u);
resolve_info.edram_format_ints_log2 = (edram_info >> 28u) & 1u;
resolve_info.decode_pwl_gamma = (edram_info & (1u << 30u)) != 0u;
#ifdef XE_RESOLVE_RESOLUTION_SCALED
resolve_info.resolution_scale =
(uint_x2_xe(coordinate_info) >> uint2_xe(16u, 19u)) & 7u;
@@ -132,6 +135,7 @@ XeResolveInfo XeResolveGetInfo(param_push_consts_xe) {
resolve_info.dest_is_array = (dest_info & (1u << 3u)) != 0u;
resolve_info.dest_slice = (dest_info >> 4u) & ((1u << 3u) - 1u);
resolve_info.dest_format = (dest_info >> 7u) & ((1u << 6u) - 1u);
resolve_info.dest_num_format = (dest_info >> 13u) & ((1u << 3u) - 1u);
resolve_info.dest_exp_bias_factor = int_bits_to_float_xe(
(int(dest_info) << (32 - (16 + 6)) >> (32 - 6) << 23) +
float_bits_to_int_xe(1.0f));
@@ -580,6 +584,127 @@ uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
}
}
// Gamma RTs store linear 10 bit color through an 8 bit PWL curve.
// Host sRGB is not a substitute. Some titles appear overexposed if full
// 8_8_8_8_GAMMA resolves average the encoded bytes directly (5345085D,
// 45410934), so we decode before MSAA averaging and encode again only
// for compatible 8_8_8_8 destinations.
float XePWLGammaToLinear(float gamma) {
gamma = saturate_xe(gamma);
float scale;
float offset;
if (gamma >= 96.0f / 255.0f) {
if (gamma >= 192.0f / 255.0f) {
scale = 8.0f / 1024.0f;
offset = -1024.0f;
} else {
scale = 4.0f / 1024.0f;
offset = -256.0f;
}
} else {
if (gamma >= 64.0f / 255.0f) {
scale = 2.0f / 1024.0f;
offset = -64.0f;
} else {
scale = 1.0f / 1024.0f;
offset = 0.0f;
}
}
float linear_value = gamma * (255.0f * 1024.0f) * scale + offset;
linear_value += trunc(linear_value * scale);
return linear_value * (1.0f / 1023.0f);
}
float3_xe XePWLGammaToLinear3(float3_xe gamma) {
return float3_xe(XePWLGammaToLinear(gamma.x),
XePWLGammaToLinear(gamma.y),
XePWLGammaToLinear(gamma.z));
}
float4_xe XePWLGammaToLinear4(float4_xe gamma) {
return float4_xe(XePWLGammaToLinear(gamma.x),
XePWLGammaToLinear(gamma.y),
XePWLGammaToLinear(gamma.z),
XePWLGammaToLinear(gamma.w));
}
float XeLinearToPWLGamma(float linear_value) {
linear_value = saturate_xe(linear_value);
float scale;
float offset;
if (linear_value >= 128.0f / 1023.0f) {
if (linear_value >= 512.0f / 1023.0f) {
scale = 1023.0f / 8.0f;
offset = 128.0f / 255.0f;
} else {
scale = 1023.0f / 4.0f;
offset = 64.0f / 255.0f;
}
} else {
if (linear_value >= 64.0f / 1023.0f) {
scale = 1023.0f / 2.0f;
offset = 32.0f / 255.0f;
} else {
scale = 1023.0f;
offset = 0.0f;
}
}
return trunc(linear_value * scale) * (1.0f / 255.0f) + offset;
}
float3_xe XeLinearToPWLGamma3(float3_xe linear_value) {
return float3_xe(XeLinearToPWLGamma(linear_value.x),
XeLinearToPWLGamma(linear_value.y),
XeLinearToPWLGamma(linear_value.z));
}
bool XeResolveSourceUsesPWLGamma(XeResolveInfo resolve_info) {
return resolve_info.decode_pwl_gamma &&
resolve_info.edram_format ==
kXenosColorRenderTargetFormat_8_8_8_8_GAMMA;
}
bool XeResolveDestStoresPWLGamma(XeResolveInfo resolve_info) {
// Resolve constants don't carry a destination gamma bit. Treat an
// unsigned 8_8_8_8 destination as PWL gamma storage.
bool dest_number_is_unorm =
resolve_info.dest_num_format ==
kXenosSurfaceNumberFormat_UnsignedRepeatingFraction;
bool dest_is_8888 =
resolve_info.dest_format == kXenosFormat_8_8_8_8 ||
resolve_info.dest_format == kXenosFormat_8_8_8_8_A ||
resolve_info.dest_format ==
kXenosFormat_8_8_8_8_AS_16_16_16_16;
return dest_number_is_unorm && dest_is_8888;
}
void XeResolvePWLGammaToLinearRGB(inout_param_xe(float4_xe, pixel)) {
pixel.rgb = XePWLGammaToLinear3(pixel.rgb);
}
void XeResolveLinearToPWLGammaRGB(inout_param_xe(float4_xe, pixel)) {
pixel.rgb = XeLinearToPWLGamma3(pixel.rgb);
}
void XeResolveDecodePWLGammaSource(
XeResolveInfo resolve_info, inout_param_xe(float4_xe, pixel)) {
// Source gamma is RGB only. 8_8_8_8_GAMMA still stores alpha as ordinary
// fixed data, so alpha needs to stay with normal resolve.
dont_flatten_xe if (XeResolveSourceUsesPWLGamma(resolve_info)) {
XeResolvePWLGammaToLinearRGB(pixel);
}
}
void XeResolveEncodePWLGammaDest(
XeResolveInfo resolve_info, inout_param_xe(float4_xe, pixel)) {
// Only re-encode when the source was PWL gamma and the destination is the
// 8_8_8_8 UNORM storage we treat as the same PWL byte stream.
dont_flatten_xe if (XeResolveSourceUsesPWLGamma(resolve_info) &&
XeResolveDestStoresPWLGamma(resolve_info)) {
XeResolveLinearToPWLGammaRGB(pixel);
}
}
void XeResolveLoad2RGBAColors(
param_byte_buffer_xe(xe_resolve_edram) param_next_after_byte_buffer_xe
XeResolveInfo resolve_info, uint address_bytes,
@@ -590,11 +715,12 @@ uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
address_bytes, pixel_stride_bytes,
resolve_info.edram_format_ints_log2, resolve_info.edram_format,
pixel_0, pixel_1);
XeResolveDecodePWLGammaSource(resolve_info, pixel_0);
XeResolveDecodePWLGammaSource(resolve_info, 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_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(
@@ -602,6 +728,8 @@ uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
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);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_0);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_1);
pixel_0 += msaa_resolve_pixel_0;
pixel_1 += msaa_resolve_pixel_1;
dont_flatten_xe
@@ -614,6 +742,8 @@ uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
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);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_0);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_1);
pixel_0 += msaa_resolve_pixel_0;
pixel_1 += msaa_resolve_pixel_1;
XeResolveLoad2RGBAUnswappedPixelSamplesFromRaw(
@@ -623,12 +753,16 @@ uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
pixel_stride_bytes, resolve_info.edram_format_ints_log2,
resolve_info.edram_format, msaa_resolve_pixel_0,
msaa_resolve_pixel_1);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_0);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_1);
pixel_0 += msaa_resolve_pixel_0;
pixel_1 += msaa_resolve_pixel_1;
}
}
pixel_0 *= exp_bias;
pixel_1 *= exp_bias;
XeResolveEncodePWLGammaDest(resolve_info, pixel_0);
XeResolveEncodePWLGammaDest(resolve_info, pixel_1);
dont_flatten_xe if (resolve_info.dest_swap) {
pixel_0 = pixel_0.bgra;
pixel_1 = pixel_1.bgra;
@@ -646,11 +780,14 @@ uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
address_bytes, pixel_stride_bytes,
resolve_info.edram_format_ints_log2, resolve_info.edram_format,
pixel_0, pixel_1, pixel_2, pixel_3);
XeResolveDecodePWLGammaSource(resolve_info, pixel_0);
XeResolveDecodePWLGammaSource(resolve_info, pixel_1);
XeResolveDecodePWLGammaSource(resolve_info, pixel_2);
XeResolveDecodePWLGammaSource(resolve_info, 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_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;
float4_xe msaa_resolve_pixel_1;
@@ -662,6 +799,10 @@ uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
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);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_0);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_1);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_2);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_3);
pixel_0 += msaa_resolve_pixel_0;
pixel_1 += msaa_resolve_pixel_1;
pixel_2 += msaa_resolve_pixel_2;
@@ -677,6 +818,10 @@ uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
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);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_0);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_1);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_2);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_3);
pixel_0 += msaa_resolve_pixel_0;
pixel_1 += msaa_resolve_pixel_1;
pixel_2 += msaa_resolve_pixel_2;
@@ -688,6 +833,10 @@ uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
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);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_0);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_1);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_2);
XeResolveDecodePWLGammaSource(resolve_info, msaa_resolve_pixel_3);
pixel_0 += msaa_resolve_pixel_0;
pixel_1 += msaa_resolve_pixel_1;
pixel_2 += msaa_resolve_pixel_2;
@@ -698,6 +847,10 @@ uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
pixel_1 *= exp_bias;
pixel_2 *= exp_bias;
pixel_3 *= exp_bias;
XeResolveEncodePWLGammaDest(resolve_info, pixel_0);
XeResolveEncodePWLGammaDest(resolve_info, pixel_1);
XeResolveEncodePWLGammaDest(resolve_info, pixel_2);
XeResolveEncodePWLGammaDest(resolve_info, pixel_3);
dont_flatten_xe if (resolve_info.dest_swap) {
pixel_0 = pixel_0.bgra;
pixel_1 = pixel_1.bgra;
@@ -721,11 +874,14 @@ uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
address_bytes, pixel_stride_bytes,
resolve_info.edram_format_ints_log2, resolve_info.edram_format,
resolve_info.dest_swap, pixels_0123, pixels_4567);
dont_flatten_xe if (XeResolveSourceUsesPWLGamma(resolve_info)) {
pixels_0123 = XePWLGammaToLinear4(pixels_0123);
pixels_4567 = XePWLGammaToLinear4(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_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(
@@ -734,6 +890,12 @@ uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
resolve_info.edram_format_ints_log2, resolve_info.edram_format,
resolve_info.dest_swap, msaa_resolve_pixels_0123,
msaa_resolve_pixels_4567);
dont_flatten_xe if (XeResolveSourceUsesPWLGamma(resolve_info)) {
msaa_resolve_pixels_0123 =
XePWLGammaToLinear4(msaa_resolve_pixels_0123);
msaa_resolve_pixels_4567 =
XePWLGammaToLinear4(msaa_resolve_pixels_4567);
}
pixels_0123 += msaa_resolve_pixels_0123;
pixels_4567 += msaa_resolve_pixels_4567;
dont_flatten_xe
@@ -747,6 +909,12 @@ uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
resolve_info.edram_format_ints_log2, resolve_info.edram_format,
resolve_info.dest_swap, msaa_resolve_pixels_0123,
msaa_resolve_pixels_4567);
dont_flatten_xe if (XeResolveSourceUsesPWLGamma(resolve_info)) {
msaa_resolve_pixels_0123 =
XePWLGammaToLinear4(msaa_resolve_pixels_0123);
msaa_resolve_pixels_4567 =
XePWLGammaToLinear4(msaa_resolve_pixels_4567);
}
pixels_0123 += msaa_resolve_pixels_0123;
pixels_4567 += msaa_resolve_pixels_4567;
XeResolveLoad8RedPixelSamplesFromRaw(
@@ -756,6 +924,12 @@ uint XeResolveEdramPixelStrideBytes(XeResolveInfo resolve_info) {
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);
dont_flatten_xe if (XeResolveSourceUsesPWLGamma(resolve_info)) {
msaa_resolve_pixels_0123 =
XePWLGammaToLinear4(msaa_resolve_pixels_0123);
msaa_resolve_pixels_4567 =
XePWLGammaToLinear4(msaa_resolve_pixels_4567);
}
pixels_0123 += msaa_resolve_pixels_0123;
pixels_4567 += msaa_resolve_pixels_4567;
}