debuggerx/eden
1
0
Fork 0
mirror of https://git.eden-emu.dev/eden-emu/eden.git synced 2026-08-30 02:16:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

[video_core] Implementation on ASTC HDR formats + Vulkan fixes (#4190)

Browse Source 
Based on the register of the ASTC implementation from Ameerj (yuzu), currently our conversion from ASTC on Desktop tends to duplicates the unpacking of ASTC converted textures, leading to bad performance; since there are some graphical effects that actually access on the band of 32bits (RGAF32) than the common 16bits (RGAF16) from our ASTC path decoding method, which commonly uses ARGB8 (degradation -> LDR) transformation to reduce the memory cost per conversion; meanwhile it could actually provide initially better performance; the engine tended to get stalled when this writted stage effect got skipped; mostly because CPU tends to look for an unimplemented texture/ format query (from which doesn't exist on our MaxwellToVK); unlike what's commonly believed HDR was a format to write enviroment effects even before their appearance on video output (since Switch doesn't have HDR, only SDR on certain engines).

The test result on this PR showed games like Astral Chain and other heavy reliant on ASTC decoding improved their performance and stability by around 15%; meanwhile could it be tied to a hardware gain, it's clearly a gain for direct access on what's required on most games. This PR also contains few Vulkan bug fix I encountered along this implementation.

Special Thanks:

-> Mr. Smoly Gidolard (@gidoly)

Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4190
This commit is contained in:
CamilleLaVey
2026-07-09 05:29:50 +02:00
committed by crueter
parent 8225151a44
commit 3a6ac6e418
6 changed files with 685 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/video_core/host_shaders/astc_decoder.comp
+276 -6
View File
@@ -1,3 +1,6 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
@@ -40,7 +43,11 @@ layout(binding = BINDING_INPUT_BUFFER, std430) readonly restrict buffer InputBuf
uvec4 astc_data[];
};
#ifdef VULKAN
layout(binding = BINDING_OUTPUT_IMAGE) uniform writeonly restrict image2DArray dest_image;
#else
layout(binding = BINDING_OUTPUT_IMAGE, rgba8) uniform writeonly restrict image2DArray dest_image;
#endif
const uint GOB_SIZE_X_SHIFT = 6;
const uint GOB_SIZE_Y_SHIFT = 3;
@@ -589,6 +596,158 @@ ivec4 BlueContract(int a, int r, int g, int b) {
return ivec4(a, (r + b) >> 1, (g + b) >> 1, b);
}
bool IsHDRColorEndpointMode(uint cem) {
return cem == 2u || cem == 3u || cem == 7u || cem == 11u || cem == 14u || cem == 15u;
}
int SignExtend(int value, uint nbits) {
int sign_bit = 1 << (nbits - 1u);
return (value ^ sign_bit) - sign_bit;
}
void DecodeHDREndpointMode7(uint v0, uint v1, uint v2, uint v3, out ivec3 e0, out ivec3 e1) {
uint modeval = ((v0 & 0xC0u) >> 6u) | ((v1 & 0x80u) >> 5u) | ((v2 & 0x80u) >> 4u);
uint majcomp;
uint mode;
if ((modeval & 0xCu) != 0xCu) {
majcomp = modeval >> 2u;
mode = modeval & 3u;
} else if (modeval != 0xFu) {
majcomp = modeval & 3u;
mode = 4u;
} else {
majcomp = 0u;
mode = 5u;
}
int red = int(v0 & 0x3Fu);
int green = int(v1 & 0x1Fu);
int blue = int(v2 & 0x1Fu);
int scale = int(v3 & 0x1Fu);
uint x0 = (v1 >> 6u) & 1u;
uint x1 = (v1 >> 5u) & 1u;
uint x2 = (v2 >> 6u) & 1u;
uint x3 = (v2 >> 5u) & 1u;
uint x4 = (v3 >> 7u) & 1u;
uint x5 = (v3 >> 6u) & 1u;
uint x6 = (v3 >> 5u) & 1u;
uint ohm = 1u << mode;
if ((ohm & 0x30u) != 0u) green |= int(x0 << 6u);
if ((ohm & 0x3Au) != 0u) green |= int(x1 << 5u);
if ((ohm & 0x30u) != 0u) blue |= int(x2 << 6u);
if ((ohm & 0x3Au) != 0u) blue |= int(x3 << 5u);
if ((ohm & 0x3Du) != 0u) scale |= int(x6 << 5u);
if ((ohm & 0x2Du) != 0u) scale |= int(x5 << 6u);
if ((ohm & 0x04u) != 0u) scale |= int(x4 << 7u);
if ((ohm & 0x3Bu) != 0u) red |= int(x4 << 6u);
if ((ohm & 0x04u) != 0u) red |= int(x3 << 6u);
if ((ohm & 0x10u) != 0u) red |= int(x5 << 7u);
if ((ohm & 0x0Fu) != 0u) red |= int(x2 << 7u);
if ((ohm & 0x05u) != 0u) red |= int(x1 << 8u);
if ((ohm & 0x0Au) != 0u) red |= int(x0 << 8u);
if ((ohm & 0x05u) != 0u) red |= int(x0 << 9u);
if ((ohm & 0x02u) != 0u) red |= int(x6 << 9u);
if ((ohm & 0x01u) != 0u) red |= int(x3 << 10u);
if ((ohm & 0x02u) != 0u) red |= int(x5 << 10u);
int shamts[6] = int[](1, 1, 2, 3, 4, 5);
int shamt = shamts[mode];
red <<= shamt;
green <<= shamt;
blue <<= shamt;
scale <<= shamt;
if (mode != 5u) {
green = red - green;
blue = red - blue;
}
if (majcomp == 1u) {
int t = red; red = green; green = t;
}
if (majcomp == 2u) {
int t = red; red = blue; blue = t;
}
e1 = ivec3(clamp(red, 0, 0xFFF), clamp(green, 0, 0xFFF), clamp(blue, 0, 0xFFF));
e0 = ivec3(clamp(red - scale, 0, 0xFFF), clamp(green - scale, 0, 0xFFF),
clamp(blue - scale, 0, 0xFFF));
}
void DecodeHDREndpointMode11(uint v0, uint v1, uint v2, uint v3, uint v4, uint v5, out ivec3 e0,
out ivec3 e1) {
uint majcomp = ((v4 & 0x80u) >> 7u) | ((v5 & 0x80u) >> 6u);
if (majcomp == 3u) {
e0 = ivec3(int(v0 << 4u), int(v2 << 4u), int((v4 & 0x7Fu) << 5u));
e1 = ivec3(int(v1 << 4u), int(v3 << 4u), int((v5 & 0x7Fu) << 5u));
return;
}
uint mode = ((v1 & 0x80u) >> 7u) | ((v2 & 0x80u) >> 6u) | ((v3 & 0x80u) >> 5u);
int va = int(v0 | ((v1 & 0x40u) << 2u));
int vb0 = int(v2 & 0x3Fu);
int vb1 = int(v3 & 0x3Fu);
int vc = int(v1 & 0x3Fu);
int vd0 = int(v4 & 0x7Fu);
int vd1 = int(v5 & 0x7Fu);
int dbitstab[8] = int[](7, 6, 7, 6, 5, 6, 5, 6);
vd0 = SignExtend(vd0, uint(dbitstab[mode]));
vd1 = SignExtend(vd1, uint(dbitstab[mode]));
uint x0 = (v2 >> 6u) & 1u;
uint x1 = (v3 >> 6u) & 1u;
uint x2 = (v4 >> 6u) & 1u;
uint x3 = (v5 >> 6u) & 1u;
uint x4 = (v4 >> 5u) & 1u;
uint x5 = (v5 >> 5u) & 1u;
uint ohm = 1u << mode;
if ((ohm & 0xA4u) != 0u) va |= int(x0 << 9u);
if ((ohm & 0x08u) != 0u) va |= int(x2 << 9u);
if ((ohm & 0x50u) != 0u) va |= int(x4 << 9u);
if ((ohm & 0x50u) != 0u) va |= int(x5 << 10u);
if ((ohm & 0xA0u) != 0u) va |= int(x1 << 10u);
if ((ohm & 0xC0u) != 0u) va |= int(x2 << 11u);
if ((ohm & 0x04u) != 0u) vc |= int(x1 << 6u);
if ((ohm & 0xE8u) != 0u) vc |= int(x3 << 6u);
if ((ohm & 0x20u) != 0u) vc |= int(x2 << 7u);
if ((ohm & 0x5Bu) != 0u) vb0 |= int(x0 << 6u);
if ((ohm & 0x5Bu) != 0u) vb1 |= int(x1 << 6u);
if ((ohm & 0x12u) != 0u) vb0 |= int(x2 << 7u);
if ((ohm & 0x12u) != 0u) vb1 |= int(x3 << 7u);
int shamt = (int(mode) >> 1) ^ 3;
va <<= shamt;
vb0 <<= shamt;
vb1 <<= shamt;
vc <<= shamt;
vd0 <<= shamt;
vd1 <<= shamt;
int r1 = clamp(va, 0, 0xFFF);
int g1 = clamp(va - vb0, 0, 0xFFF);
int b1 = clamp(va - vb1, 0, 0xFFF);
int r0 = clamp(va - vc, 0, 0xFFF);
int g0 = clamp(va - vb0 - vc - vd0, 0, 0xFFF);
int b0 = clamp(va - vb1 - vc - vd1, 0, 0xFFF);
if (majcomp == 1u) {
int t;
t = r0; r0 = g0; g0 = t;
t = r1; r1 = g1; g1 = t;
} else if (majcomp == 2u) {
int t;
t = r0; r0 = b0; b0 = t;
t = r1; r1 = b1; b1 = t;
}
e0 = ivec3(r0, g0, b0);
e1 = ivec3(r1, g1, b1);
}
void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode, uint color_values[32],
inout uint colvals_index) {
#define READ_UINT_VALUES(N) \
@@ -715,8 +874,86 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode, ui
}
break;
}
case 2: {
READ_UINT_VALUES(2)
uint y0, y1;
if (V[0].y >= V[0].x) {
y0 = V[0].x << 4u;
y1 = V[0].y << 4u;
} else {
y0 = (V[0].y << 4u) + 8u;
y1 = (V[0].x << 4u) - 8u;
}
ep1 = uvec4(0x780u, y0, y0, y0);
ep2 = uvec4(0x780u, y1, y1, y1);
break;
}
case 3: {
READ_UINT_VALUES(2)
uint y0, d;
if ((V[0].x & 0x80u) != 0u) {
y0 = ((V[0].y & 0xE0u) << 4u) | ((V[0].x & 0x7Fu) << 2u);
d = (V[0].y & 0x1Fu) << 2u;
} else {
y0 = ((V[0].y & 0xF0u) << 4u) | ((V[0].x & 0x7Fu) << 1u);
d = (V[0].y & 0x0Fu) << 1u;
}
const uint y1 = min(y0 + d, 0xFFFu);
ep1 = uvec4(0x780u, y0, y0, y0);
ep2 = uvec4(0x780u, y1, y1, y1);
break;
}
case 7: {
READ_UINT_VALUES(4)
ivec3 e0, e1;
DecodeHDREndpointMode7(V[0].x, V[0].y, V[0].z, V[0].w, e0, e1);
ep1 = uvec4(0x780u, uint(e0.x), uint(e0.y), uint(e0.z));
ep2 = uvec4(0x780u, uint(e1.x), uint(e1.y), uint(e1.z));
break;
}
case 11: {
READ_UINT_VALUES(6)
ivec3 e0, e1;
DecodeHDREndpointMode11(V[0].x, V[0].y, V[0].z, V[0].w, V[1].x, V[1].y, e0, e1);
ep1 = uvec4(0x780u, uint(e0.x), uint(e0.y), uint(e0.z));
ep2 = uvec4(0x780u, uint(e1.x), uint(e1.y), uint(e1.z));
break;
}
case 14: {
READ_UINT_VALUES(8)
ivec3 e0, e1;
DecodeHDREndpointMode11(V[0].x, V[0].y, V[0].z, V[0].w, V[1].x, V[1].y, e0, e1);
ep1 = uvec4(V[1].z, uint(e0.x), uint(e0.y), uint(e0.z));
ep2 = uvec4(V[1].w, uint(e1.x), uint(e1.y), uint(e1.z));
break;
}
case 15: {
READ_UINT_VALUES(8)
ivec3 e0, e1;
DecodeHDREndpointMode11(V[0].x, V[0].y, V[0].z, V[0].w, V[1].x, V[1].y, e0, e1);
const uint mode = ((V[1].z >> 7u) & 1u) | ((V[1].w >> 6u) & 2u);
int a6 = int(V[1].z & 0x7Fu);
int a7 = int(V[1].w & 0x7Fu);
int alpha0, alpha1;
if (mode == 3u) {
alpha0 = a6 << 5;
alpha1 = a7 << 5;
} else {
a6 |= (a7 << int(mode + 1u)) & 0x780;
a7 &= int(0x3Fu >> mode);
a7 ^= int(0x20u >> mode);
a7 -= int(0x20u >> mode);
a6 <<= int(4u - mode);
a7 <<= int(4u - mode);
a7 += a6;
alpha0 = a6;
alpha1 = clamp(a7, 0, 0xFFF);
}
ep1 = uvec4(uint(alpha0), uint(e0.x), uint(e0.y), uint(e0.z));
ep2 = uvec4(uint(alpha1), uint(e1.x), uint(e1.y), uint(e1.z));
break;
}
default: {
// HDR mode, or more likely a bug computing the color_endpoint_mode
ep1 = uvec4(0xFF, 0xFF, 0, 0);
ep2 = uvec4(0xFF, 0xFF, 0, 0);
break;
@@ -1112,13 +1349,46 @@ void DecompressBlock(ivec3 coord) {
if (num_partitions > 1) {
local_partition = Select2DPartition(partition_index, i, j, num_partitions);
}
const uvec4 C0 = ReplicateByteTo16(endpoints0[local_partition]);
const uvec4 C1 = ReplicateByteTo16(endpoints1[local_partition]);
const uint local_cem = color_endpoint_mode[local_partition];
const uvec4 weight_vec = GetUnquantizedWeightVector(j, i, size_params, plane_index, dual_plane);
const vec4 Cf =
vec4((C0 * (uvec4(64) - weight_vec) + C1 * weight_vec + uvec4(32)) / 64);
const vec4 p = (Cf / 65535.0f);
vec4 p;
if (IsHDRColorEndpointMode(local_cem)) {
const uvec4 C0 = endpoints0[local_partition] << 4u;
const uvec4 C1 = endpoints1[local_partition] << 4u;
const uvec4 C = (C0 * (uvec4(64) - weight_vec) + C1 * weight_vec + uvec4(32)) / 64u;
const uvec4 E = (C & uvec4(0xF800u)) >> 11u;
const uvec4 M = C & uvec4(0x7FFu);
const uvec4 Mt_lo = 3u * M;
const uvec4 Mt_mid = 4u * M - 512u;
const uvec4 Mt_hi = 5u * M - 2048u;
const uvec4 Mt =
mix(Mt_lo, mix(Mt_mid, Mt_hi, greaterThanEqual(M, uvec4(1536u))),
greaterThanEqual(M, uvec4(512u)));
const uvec4 Cf = (E << 10u) + (Mt >> 3u);
const uvec4 half_bits = mix(Cf, uvec4(0x7BFFu), greaterThanEqual(Cf, uvec4(0x7C00u)));
p = vec4(unpackHalf2x16(half_bits.x).x, unpackHalf2x16(half_bits.y).x,
unpackHalf2x16(half_bits.z).x, unpackHalf2x16(half_bits.w).x);
if (local_cem == 14u) {
const uint a0 = ReplicateByteTo16(uvec4(endpoints0[local_partition].x)).x;
const uint a1 = ReplicateByteTo16(uvec4(endpoints1[local_partition].x)).x;
const uint Ca = (a0 * (64u - weight_vec.x) + a1 * weight_vec.x + 32u) / 64u;
p.x = float(Ca) / 65535.0f;
}
} else {
const uvec4 C0 = ReplicateByteTo16(endpoints0[local_partition]);
const uvec4 C1 = ReplicateByteTo16(endpoints1[local_partition]);
const vec4 Cf =
vec4((C0 * (uvec4(64) - weight_vec) + C1 * weight_vec + uvec4(32)) / 64);
p = Cf / 65535.0f;
}
#ifdef VULKAN
imageStore(dest_image, coord + ivec3(i, j, 0), p.gbar);
#else
imageStore(dest_image, coord + ivec3(i, j, 0), clamp(p, 0.0f, 1.0f).gbar);
#endif
}
}
}