ShuriZma
/
xenia
mirror of https://github.com/xenia-project/xenia.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

[D3D12] 32bpb and 128bpb untiling, fix 3D tiled offset

This commit is contained in:
Triang3l 2018-08-06 23:31:09 +03:00
parent cfd3821b83
commit 83cf482a50
7 changed files with 155 additions and 89 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
src/xenia/gpu/d3d12/shaders/texture_address.hlsli
View File

@ -5,13 +5,13 @@
// consecutive blocks along X.
// https://github.com/gildor2/UModel/blob/de8fbd3bc922427ea056b7340202dcdcc19ccff5/Unreal/UnTexture.cpp#L495
uint4 XeTextureTiledOffset2D(uint2 p, uint width, uint log2_bpb) {
uint4 XeTextureTiledOffset2D(uint2 p, uint width, uint bpb_log2) {
uint4 x4 = uint4(0u, 1u, 2u, 3u) + p.xxxx;
// Top bits of coordinates.
uint4 macro =
((x4 >> 5u) + (p.y >> 5u) * ((width + 31u) >> 5u)) << (log2_bpb + 7u);
((x4 >> 5u) + (p.y >> 5u) * ((width + 31u) >> 5u)) << (bpb_log2 + 7u);
// Lower bits of coordinates (result is 6-bit value).
uint4 micro = ((x4 & 7u) + ((p.y & 0xEu) << 2u)) << log2_bpb;
uint4 micro = ((x4 & 7u) + ((p.y & 0xEu) << 2u)) << bpb_log2;
// Mix micro/macro + add few remaining x/y bits.
uint4 offset =
macro + ((micro & ~0xFu) << 1u) + (micro & 0xFu) + ((p.y & 1u) << 4u);
@ -26,27 +26,27 @@ uint4 XeTextureTiledOffset2D(uint2 p, uint width, uint log2_bpb) {
// Reverse-engineered from an executable.
// The base/micro/macro names were chosen pretty much at random and don't have
// the same meaning as in TiledOffset2D.
uint4 XeTextureTiledOffset3D(uint3 p, uint2 width_height, uint log2_bpb) {
uint4 XeTextureTiledOffset3D(uint3 p, uint2 width_height, uint bpb_log2) {
uint4 x4 = uint4(0u, 1u, 2u, 3u) + p.xxxx;
uint2 aligned_size = (width_height + 31u) & ~31u;
uint base = ((p.z >> 2u) * ((aligned_size.x * aligned_size.y) >> 4u) +
(p.y >> 4u)) * (aligned_size.x >> 5u);
uint base = ((p.z >> 2u) * (aligned_size.y >> 4u) + (p.y >> 4u)) *
(aligned_size.x >> 5u);
uint4 micro = (((p.z >> 2u) + (p.y >> 3u)) & 1u).xxxx;
micro += (((micro << 1u) + (x4 >> 3u)) & 3u) << 1u;
uint4 macro = (((x4 & 7u) + ((p.y & 6u) << 2u)) << (log2_bpb + 6u)) >> 6u;
macro = (((((((x4 >> 5u) + base) << (log2_bpb + 6u)) & 0xFFFFFFFu) << 1u) +
uint4 macro = (((x4 & 7u) + ((p.y & 6u) << 2u)) << (bpb_log2 + 6u)) >> 6u;
macro = (((((((x4 >> 5u) + base) << (bpb_log2 + 6u)) & 0xFFFFFFFu) << 1u) +
(macro & ~15u)) << 1u) + (macro & 15u) +
((p.z & 3u) << (log2_bpb + 6u)) + ((p.y & 1u) << 4u);
((p.z & 3u) << (bpb_log2 + 6u)) + ((p.y & 1u) << 4u);
return ((((((((macro >> 6u) & 7u) + ((micro & 1u) << 3u)) << 3u) +
(micro & ~1u)) << 2u) + (macro & ~511u)) << 3u) + (macro & 63u);
}
uint XeTextureGuestLinearOffset(uint3 p, uint height, uint pitch, uint bpb) {
return p.x * bpb + ((p.z * ((height + 31u) & ~31u) + pitch) * p.y);
return p.x * bpb + (p.z * ((height + 31u) & ~31u) + p.y) * pitch;
}
uint XeTextureHostLinearOffset(uint3 p, uint height, uint pitch, uint bpb) {
return p.x * bpb + ((p.z * height + pitch) * p.y);
return p.x * bpb + (p.z * height + p.y) * pitch;
}
#endif // XENIA_GPU_D3D12_SHADERS_TEXTURE_ADDRESS_HLSLI_

22
src/xenia/gpu/d3d12/shaders/texture_copy.hlsli
View File

@ -25,4 +25,26 @@ cbuffer xe_texture_copy_constants : register(b0) {
ByteAddressBuffer xe_texture_copy_source : register(t0);
RWByteAddressBuffer xe_texture_copy_dest : register(u0);
// bpb and bpb_log2 are separate because bpb may be not a power of 2 (like 96).
uint4 XeTextureCopyGuestBlockOffsets(uint3 block_index, uint bpb,
uint bpb_log2) {
uint3 block_index_guest = block_index + xe_texture_copy_guest_mip_offset;
uint4 block_offsets_guest;
[branch] if (xe_texture_copy_guest_pitch == XeTextureCopyGuestPitchTiled) {
[branch] if (xe_texture_copy_is_3d) {
block_offsets_guest = XeTextureTiledOffset3D(
block_index_guest, xe_texture_copy_size.xy, bpb_log2);
} else {
block_offsets_guest = XeTextureTiledOffset2D(
block_index_guest.xy, xe_texture_copy_size.x, bpb_log2);
}
} else {
block_offsets_guest =
uint4(0u, 1u, 2u, 3u) * bpb + XeTextureGuestLinearOffset(
block_index_guest, xe_texture_copy_size.y,
xe_texture_copy_guest_pitch, 16u);
}
return block_offsets_guest + xe_texture_copy_guest_base;
}
#endif // XENIA_GPU_D3D12_SHADERS_TEXTURE_COPY_HLSLI_

29
src/xenia/gpu/d3d12/shaders/texture_load_128bpb.cs.hlsl Normal file
View File

@ -0,0 +1,29 @@
#include "texture_copy.hlsli"
[numthreads(8, 32, 1)]
void main(uint3 xe_thread_id : SV_DispatchThreadID) {
// 1 thread = 4 uint4 blocks.
uint3 block_index = xe_thread_id;
block_index.x <<= 2u;
[branch] if (any(block_index >= xe_texture_copy_size)) {
return;
}
uint4 block_offsets_guest =
XeTextureCopyGuestBlockOffsets(block_index, 16u, 4u);
uint4 block_0 = xe_texture_copy_source.Load4(block_offsets_guest.x);
uint4 block_1 = xe_texture_copy_source.Load4(block_offsets_guest.y);
uint4 block_2 = xe_texture_copy_source.Load4(block_offsets_guest.z);
uint4 block_3 = xe_texture_copy_source.Load4(block_offsets_guest.w);
block_0 = XeByteSwap(block_0, xe_texture_copy_endianness);
block_1 = XeByteSwap(block_1, xe_texture_copy_endianness);
block_2 = XeByteSwap(block_2, xe_texture_copy_endianness);
block_3 = XeByteSwap(block_3, xe_texture_copy_endianness);
uint block_offset_host = XeTextureHostLinearOffset(
block_index, xe_texture_copy_size.y, xe_texture_copy_host_pitch, 16u) +
xe_texture_copy_host_base;
uint4 block_offsets_host = uint4(0u, 16u, 32u, 48u) + block_offset_host;
xe_texture_copy_dest.Store4(block_offsets_host.x, block_0);
xe_texture_copy_dest.Store4(block_offsets_host.y, block_1);
xe_texture_copy_dest.Store4(block_offsets_host.z, block_2);
xe_texture_copy_dest.Store4(block_offsets_host.w, block_3);
}

22
src/xenia/gpu/d3d12/shaders/texture_load_32bpb.cs.hlsl Normal file
View File

@ -0,0 +1,22 @@
#include "texture_copy.hlsli"
[numthreads(8, 32, 1)]
void main(uint3 xe_thread_id : SV_DispatchThreadID) {
// 1 thread = 4 uint blocks.
uint3 block_index = xe_thread_id;
block_index.x <<= 2u;
[branch] if (any(block_index >= xe_texture_copy_size)) {
return;
}
uint4 block_offsets_guest =
XeTextureCopyGuestBlockOffsets(block_index, 4u, 2u);
uint4 blocks = uint4(xe_texture_copy_source.Load(block_offsets_guest.x),
xe_texture_copy_source.Load(block_offsets_guest.y),
xe_texture_copy_source.Load(block_offsets_guest.z),
xe_texture_copy_source.Load(block_offsets_guest.w));
blocks = XeByteSwap(blocks, xe_texture_copy_endianness);
uint block_offset_host = XeTextureHostLinearOffset(
block_index, xe_texture_copy_size.y, xe_texture_copy_host_pitch, 4u) +
xe_texture_copy_host_base;
xe_texture_copy_dest.Store4(block_offset_host, blocks);
}

18
src/xenia/gpu/d3d12/shaders/texture_load_64bpb.cs.hlsl
View File

@ -8,22 +8,8 @@ void main(uint3 xe_thread_id : SV_DispatchThreadID) {
[branch] if (any(block_index >= xe_texture_copy_size)) {
return;
}
uint3 block_index_guest = block_index + xe_texture_copy_guest_mip_offset;
uint4 block_offsets_guest;
[branch] if (xe_texture_copy_guest_pitch == XeTextureCopyGuestPitchTiled) {
[branch] if (xe_texture_copy_is_3d) {
block_offsets_guest = XeTextureTiledOffset3D(
block_index_guest, xe_texture_copy_size.xy, 3u);
} else {
block_offsets_guest = XeTextureTiledOffset2D(
block_index_guest.xy, xe_texture_copy_size.x, 3u);
}
} else {
block_offsets_guest = uint4(0u, 8u, 16u, 24u) + XeTextureGuestLinearOffset(
block_index_guest, xe_texture_copy_size.y, xe_texture_copy_guest_pitch,
8u);
}
block_offsets_guest += xe_texture_copy_guest_base;
uint4 block_offsets_guest =
XeTextureCopyGuestBlockOffsets(block_index, 8u, 3u);
uint4 blocks_01 = uint4(xe_texture_copy_source.Load2(block_offsets_guest.x),
xe_texture_copy_source.Load2(block_offsets_guest.y));
uint4 blocks_23 = uint4(xe_texture_copy_source.Load2(block_offsets_guest.z),

53
src/xenia/gpu/d3d12/texture_cache.cc
View File

@ -24,6 +24,8 @@ namespace gpu {
namespace d3d12 {
// Generated with `xb buildhlsl`.
#include "xenia/gpu/d3d12/shaders/bin/texture_load_128bpb_cs.h"
#include "xenia/gpu/d3d12/shaders/bin/texture_load_32bpb_cs.h"
#include "xenia/gpu/d3d12/shaders/bin/texture_load_64bpb_cs.h"
const TextureCache::HostFormat TextureCache::host_formats_[64] = {
@ -33,8 +35,8 @@ const TextureCache::HostFormat TextureCache::host_formats_[64] = {
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_1_5_5_5
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_5_6_5
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_6_5_5
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_8_8_8_8
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_2_10_10_10
{DXGI_FORMAT_R8G8B8A8_UNORM, CopyMode::k32bpb}, // k_8_8_8_8
{DXGI_FORMAT_R10G10B10A2_UNORM, CopyMode::k32bpb}, // k_2_10_10_10
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_8_A
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_8_B
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_8_8
@ -45,27 +47,27 @@ const TextureCache::HostFormat TextureCache::host_formats_[64] = {
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_4_4_4_4
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_10_11_11
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_11_11_10
{DXGI_FORMAT_BC1_UNORM, CopyMode::k64Bpb}, // k_DXT1
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_DXT2_3
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_DXT4_5
{DXGI_FORMAT_BC1_UNORM, CopyMode::k64bpb}, // k_DXT1
{DXGI_FORMAT_BC2_UNORM, CopyMode::k128bpb}, // k_DXT2_3
{DXGI_FORMAT_BC3_UNORM, CopyMode::k128bpb}, // k_DXT4_5
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_DXV
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_24_8
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_24_8_FLOAT
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_16
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_16_16
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_16_16_16_16
{DXGI_FORMAT_R16G16_UNORM, CopyMode::k32bpb}, // k_16_16
{DXGI_FORMAT_R16G16B16A16_UNORM, CopyMode::k64bpb}, // k_16_16_16_16
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_16_EXPAND
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_16_16_EXPAND
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_16_16_16_16_EXPAND
{DXGI_FORMAT_R16G16_UNORM, CopyMode::k32bpb}, // k_16_16_EXPAND
{DXGI_FORMAT_R16G16B16A16_UNORM, CopyMode::k64bpb}, // k_16_16_16_16_EXPAND
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_16_FLOAT
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_16_16_FLOAT
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_16_16_16_16_FLOAT
{DXGI_FORMAT_R16G16_FLOAT, CopyMode::k32bpb}, // k_16_16_FLOAT
{DXGI_FORMAT_R16G16B16A16_FLOAT, CopyMode::k64bpb}, // k_16_16_16_16_FLOAT
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_32
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_32_32
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_32_32_32_32
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_32_FLOAT
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_32_32_FLOAT
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_32_32_32_32_FLOAT
{DXGI_FORMAT_R32_FLOAT, CopyMode::k32bpb}, // k_32_FLOAT
{DXGI_FORMAT_R32G32_FLOAT, CopyMode::k64bpb}, // k_32_32_FLOAT
{DXGI_FORMAT_R32G32B32A32_FLOAT, CopyMode::k128bpb}, // k_32_32_32_32_FLOAT
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_32_AS_8
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_32_AS_8_8
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_16_MPEG
@ -77,11 +79,12 @@ const TextureCache::HostFormat TextureCache::host_formats_[64] = {
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_16_MPEG_INTERLACED
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_16_16_MPEG_INTERLACED
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_DXN
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_8_8_8_8_AS_16_16_16_16
{DXGI_FORMAT_BC1_UNORM, CopyMode::k64Bpb}, // k_DXT1_AS_16_16_16_16
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_DXT2_3_AS_16_16_16_16
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_DXT4_5_AS_16_16_16_16
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_2_10_10_10_AS_16_16_16_16
{DXGI_FORMAT_R8G8B8A8_UNORM, CopyMode::k32bpb}, // k_8_8_8_8_AS_16_16_16_16
{DXGI_FORMAT_BC1_UNORM, CopyMode::k64bpb}, // k_DXT1_AS_16_16_16_16
{DXGI_FORMAT_BC2_UNORM, CopyMode::k128bpb}, // k_DXT2_3_AS_16_16_16_16
{DXGI_FORMAT_BC3_UNORM, CopyMode::k128bpb}, // k_DXT4_5_AS_16_16_16_16
{DXGI_FORMAT_R10G10B10A2_UNORM,
CopyMode::k32bpb}, // k_2_10_10_10_AS_16_16_16_16
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_10_11_11_AS_16_16_16_16
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_11_11_10_AS_16_16_16_16
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_32_32_32_FLOAT
@ -89,7 +92,7 @@ const TextureCache::HostFormat TextureCache::host_formats_[64] = {
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_DXT5A
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_CTX1
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_DXT3A_AS_1_1_1_1
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_8_8_8_8_GAMMA
{DXGI_FORMAT_R8G8B8A8_UNORM, CopyMode::k32bpb}, // k_8_8_8_8_GAMMA
{DXGI_FORMAT_UNKNOWN, CopyMode::kUnknown}, // k_2_10_10_10_FLOAT
};
@ -97,7 +100,10 @@ const char* const TextureCache::dimension_names_[4] = {"1D", "2D", "3D",
"cube"};
const TextureCache::CopyModeInfo TextureCache::copy_mode_info_[] = {
{texture_load_64bpb_cs, sizeof(texture_load_64bpb_cs)}};
{texture_load_32bpb_cs, sizeof(texture_load_32bpb_cs)},
{texture_load_64bpb_cs, sizeof(texture_load_64bpb_cs)},
{texture_load_128bpb_cs, sizeof(texture_load_128bpb_cs)},
};
TextureCache::TextureCache(D3D12CommandProcessor* command_processor,
RegisterFile* register_file,
@ -810,9 +816,8 @@ bool TextureCache::LoadTextureData(Texture* texture) {
command_processor_->ReleaseScratchGPUBuffer(copy_buffer, copy_buffer_state);
// TODO(Triang3l): Uncomment when done testing untiling shaders.
/* texture->base_in_sync = true;
texture->mips_in_sync = true; */
texture->base_in_sync = true;
texture->mips_in_sync = true;
LogTextureAction(texture, "Loaded");
return true;

4
src/xenia/gpu/d3d12/texture_cache.h
View File

@ -86,7 +86,9 @@ class TextureCache {
};
enum class CopyMode {
k64Bpb,
k32bpb,
k64bpb,
k128bpb,
kCount,