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

Huge set of performance improvements, combined with an architecture specific build and clang-cl users have reported absurd gains over master for some gains, in the range 50%-90%

Browse Source 
But for normal msvc builds i would put it at around 30-50%
Added per-xexmodule caching of information per instruction, can be used to remember what code needs compiling at start up
Record what guest addresses wrote mmio and backpropagate that to future runs, eliminating dependence on exception trapping. this makes many games like h3 actually tolerable to run under a debugger
fixed a number of errors where temporaries were being passed by reference/pointer
Can now be compiled with clang-cl 14.0.1, requires -Werror off though and some other solution/project changes.
Added macros wrapping compiler extensions like noinline, forceinline, __expect, and cold.
Removed the "global lock" in guest code completely. It does not properly emulate the behavior of mfmsrd/mtmsr and it seriously cripples amd cpus. Removing this yielded around a 3x speedup in Halo Reach for me.
Disabled the microprofiler for now. The microprofiler has a huge performance cost associated with it. Developers can re-enable it in the base/profiling header if they really need it
Disable the trace writer in release builds. despite just returning after checking if the file was open the trace functions were consuming about 0.60% cpu time total
Add IsValidReg, GetRegisterInfo is a huge (about 45k) branching function and using that to check if a register was valid consumed a significant chunk of time
Optimized RingBuffer::ReadAndSwap and RingBuffer::read_count. This gave us the largest overall boost in performance. The memcpies were unnecessary and one of them was always a no-op
Added simplification rules for multiplicative patterns like (x+x), (x<<1)+x
For the most frequently called win32 functions i added code to call their underlying NT implementations, which lets us skip a lot of MS code we don't care about/isnt relevant to our usecases
^this can be toggled off in the platform_win header
handle indirect call true with constant function pointer, was occurring in h3
lookup host format swizzle in denser array
by default, don't check if a gpu register is unknown, instead just check if its out of range. controlled by a cvar
^looking up whether its known or not took approx 0.3% cpu time
Changed some things in /cpu to make the project UNITYBUILD friendly
The timer thread was spinning way too much and consuming a ton of cpu, changed it to use a blocking wait instead
tagged some conditions as XE_UNLIKELY/LIKELY based on profiler feedback (will only affect clang builds)
Shifted around some code in CommandProcessor::WriteRegister based on how frequently it was executed
added support for docdecaduple precision floating point so that we can represent our performance gains numerically
tons of other stuff im probably forgetting
This commit is contained in:
chss95cs@gmail.com
2022-08-13 12:59:00 -07:00
parent 2f59487bf3
commit cb85fe401c
49 changed files with 1462 additions and 483 deletions
Download Patch File Download Diff File Expand all files Collapse all files

58
src/xenia/gpu/command_processor.cc
View File

@@ -16,6 +16,7 @@
#include "third_party/fmt/include/fmt/format.h"
#include "xenia/base/byte_stream.h"
#include "xenia/base/cvar.h"
#include "xenia/base/logging.h"
#include "xenia/base/math.h"
#include "xenia/base/profiling.h"
@@ -28,6 +29,10 @@
#include "xenia/kernel/kernel_state.h"
#include "xenia/kernel/user_module.h"
DEFINE_bool(log_unknown_register_writes, false,
"Log writes to unknown registers from "
"CommandProcessor::WriteRegister. Has significant performance hit.",
"GPU");
namespace xe {
namespace gpu {
@@ -329,19 +334,9 @@ void CommandProcessor::UpdateWritePointer(uint32_t value) {
write_ptr_index_ = value;
write_ptr_index_event_->Set();
}
void CommandProcessor::WriteRegister(uint32_t index, uint32_t value) {
void CommandProcessor::HandleSpecialRegisterWrite(uint32_t index,
uint32_t value) {
RegisterFile& regs = *register_file_;
if (index >= RegisterFile::kRegisterCount) {
XELOGW("CommandProcessor::WriteRegister index out of bounds: {}", index);
return;
}
regs.values[index].u32 = value;
if (!regs.GetRegisterInfo(index)) {
XELOGW("GPU: Write to unknown register ({:04X} = {:08X})", index, value);
}
// Scratch register writeback.
if (index >= XE_GPU_REG_SCRATCH_REG0 && index <= XE_GPU_REG_SCRATCH_REG7) {
uint32_t scratch_reg = index - XE_GPU_REG_SCRATCH_REG0;
@@ -469,6 +464,43 @@ void CommandProcessor::WriteRegister(uint32_t index, uint32_t value) {
}
}
}
void CommandProcessor::WriteRegister(uint32_t index, uint32_t value) {
if (XE_UNLIKELY(cvars::log_unknown_register_writes)) {
// chrispy: rearrange check order, place set after checks
if (XE_UNLIKELY(!register_file_->IsValidRegister(index))) {
XELOGW("GPU: Write to unknown register ({:04X} = {:08X})", index, value);
check_reg_out_of_bounds:
if (XE_UNLIKELY(index >= RegisterFile::kRegisterCount)) {
XELOGW("CommandProcessor::WriteRegister index out of bounds: {}",
index);
return;
}
}
} else {
goto check_reg_out_of_bounds;
}
register_file_->values[index].u32 = value;
// regs with extra logic on write: XE_GPU_REG_COHER_STATUS_HOST
// XE_GPU_REG_DC_LUT_RW_INDEX
// XE_GPU_REG_DC_LUT_SEQ_COLOR XE_GPU_REG_DC_LUT_PWL_DATA
// XE_GPU_REG_DC_LUT_30_COLOR
// quick pre-test
// todo: figure out just how unlikely this is. if very (it ought to be, theres
// a ton of registers other than these) make this predicate branchless and
// mark with unlikely, then make HandleSpecialRegisterWrite noinline yep, its
// very unlikely. these ORS here are meant to be bitwise ors, so that we do
// not do branching evaluation of the conditions (we will almost always take
// all of the branches)
if (XE_UNLIKELY(
(index - XE_GPU_REG_SCRATCH_REG0 < 8) |
(index == XE_GPU_REG_COHER_STATUS_HOST) |
((index - XE_GPU_REG_DC_LUT_RW_INDEX) <=
(XE_GPU_REG_DC_LUT_30_COLOR - XE_GPU_REG_DC_LUT_RW_INDEX)))) {
HandleSpecialRegisterWrite(index, value);
}
}
void CommandProcessor::MakeCoherent() {
SCOPE_profile_cpu_f("gpu");
@@ -570,7 +602,7 @@ void CommandProcessor::ExecuteIndirectBuffer(uint32_t ptr, uint32_t count) {
// Return up a level if we encounter a bad packet.
XELOGE("**** INDIRECT RINGBUFFER: Failed to execute packet.");
assert_always();
//break;
// break;
}
} while (reader.read_count());

4
src/xenia/gpu/command_processor.h
View File

@@ -150,7 +150,9 @@ class CommandProcessor {
void WorkerThreadMain();
virtual bool SetupContext() = 0;
virtual void ShutdownContext() = 0;
// rarely needed, most register writes have no special logic here
XE_NOINLINE
void HandleSpecialRegisterWrite(uint32_t index, uint32_t value);
virtual void WriteRegister(uint32_t index, uint32_t value);
const reg::DC_LUT_30_COLOR* gamma_ramp_256_entry_table() const {

2
src/xenia/gpu/d3d12/d3d12_command_processor.cc
View File

@@ -712,7 +712,7 @@ void D3D12CommandProcessor::SetViewport(const D3D12_VIEWPORT& viewport) {
ff_viewport_update_needed_ |= ff_viewport_.Height != viewport.Height;
ff_viewport_update_needed_ |= ff_viewport_.MinDepth != viewport.MinDepth;
ff_viewport_update_needed_ |= ff_viewport_.MaxDepth != viewport.MaxDepth;
if (ff_viewport_update_needed_) {
if (XE_UNLIKELY(ff_viewport_update_needed_)) {
ff_viewport_ = viewport;
deferred_command_list_.RSSetViewport(ff_viewport_);
ff_viewport_update_needed_ = false;

26
src/xenia/gpu/d3d12/d3d12_render_target_cache.cc
View File

@@ -4799,18 +4799,16 @@ void D3D12RenderTargetCache::PerformTransfersAndResolveClears(
if (!current_transfers.empty()) {
are_current_command_list_render_targets_valid_ = false;
if (dest_rt_key.is_depth) {
command_list.D3DOMSetRenderTargets(
0, nullptr, FALSE, &dest_d3d12_rt.descriptor_draw().GetHandle());
auto handle = dest_d3d12_rt.descriptor_draw().GetHandle();
command_list.D3DOMSetRenderTargets(0, nullptr, FALSE, &handle);
if (!use_stencil_reference_output_) {
command_processor_.SetStencilReference(UINT8_MAX);
}
} else {
command_list.D3DOMSetRenderTargets(
1,
&(dest_d3d12_rt.descriptor_load_separate().IsValid()
? dest_d3d12_rt.descriptor_load_separate().GetHandle()
: dest_d3d12_rt.descriptor_draw().GetHandle()),
FALSE, nullptr);
auto handle = dest_d3d12_rt.descriptor_load_separate().IsValid()
? dest_d3d12_rt.descriptor_load_separate().GetHandle()
: dest_d3d12_rt.descriptor_draw().GetHandle();
command_list.D3DOMSetRenderTargets(1, &handle, FALSE, nullptr);
}
uint32_t dest_pitch_tiles = dest_rt_key.GetPitchTiles();
@@ -5425,12 +5423,12 @@ void D3D12RenderTargetCache::PerformTransfersAndResolveClears(
dest_d3d12_rt.SetResourceState(D3D12_RESOURCE_STATE_RENDER_TARGET),
D3D12_RESOURCE_STATE_RENDER_TARGET);
if (clear_via_drawing) {
command_list.D3DOMSetRenderTargets(
1,
&(dest_d3d12_rt.descriptor_load_separate().IsValid()
? dest_d3d12_rt.descriptor_load_separate().GetHandle()
: dest_d3d12_rt.descriptor_draw().GetHandle()),
FALSE, nullptr);
auto handle =
(dest_d3d12_rt.descriptor_load_separate().IsValid()
? dest_d3d12_rt.descriptor_load_separate().GetHandle()
: dest_d3d12_rt.descriptor_draw().GetHandle());
command_list.D3DOMSetRenderTargets(1, &handle, FALSE, nullptr);
are_current_command_list_render_targets_valid_ = true;
D3D12_VIEWPORT clear_viewport;
clear_viewport.TopLeftX = float(clear_rect.left);

329
src/xenia/gpu/d3d12/d3d12_texture_cache.cc
View File

@@ -78,314 +78,24 @@ namespace shaders {
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/texture_load_r5g6b5_b5g6r5_scaled_cs.h"
} // namespace shaders
const D3D12TextureCache::HostFormat D3D12TextureCache::host_formats_[64] = {
// k_1_REVERSE
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_1
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_8
{DXGI_FORMAT_R8_TYPELESS, DXGI_FORMAT_R8_UNORM, kLoadShaderIndex8bpb,
DXGI_FORMAT_R8_SNORM, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_1_5_5_5
// Red and blue swapped in the load shader for simplicity.
{DXGI_FORMAT_B5G5R5A1_UNORM, DXGI_FORMAT_B5G5R5A1_UNORM,
kLoadShaderIndexR5G5B5A1ToB5G5R5A1, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_5_6_5
// Red and blue swapped in the load shader for simplicity.
{DXGI_FORMAT_B5G6R5_UNORM, DXGI_FORMAT_B5G6R5_UNORM,
kLoadShaderIndexR5G6B5ToB5G6R5, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_6_5_5
// On the host, green bits in blue, blue bits in green.
{DXGI_FORMAT_B5G6R5_UNORM, DXGI_FORMAT_B5G6R5_UNORM,
kLoadShaderIndexR5G5B6ToB5G6R5WithRBGASwizzle, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, XE_GPU_MAKE_TEXTURE_SWIZZLE(R, B, G, G)},
// k_8_8_8_8
{DXGI_FORMAT_R8G8B8A8_TYPELESS, DXGI_FORMAT_R8G8B8A8_UNORM,
kLoadShaderIndex32bpb, DXGI_FORMAT_R8G8B8A8_SNORM, kLoadShaderIndexUnknown,
false, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_2_10_10_10
{DXGI_FORMAT_R10G10B10A2_TYPELESS, DXGI_FORMAT_R10G10B10A2_UNORM,
kLoadShaderIndex32bpb, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_8_A
{DXGI_FORMAT_R8_TYPELESS, DXGI_FORMAT_R8_UNORM, kLoadShaderIndex8bpb,
DXGI_FORMAT_R8_SNORM, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_8_B
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_8_8
{DXGI_FORMAT_R8G8_TYPELESS, DXGI_FORMAT_R8G8_UNORM, kLoadShaderIndex16bpb,
DXGI_FORMAT_R8G8_SNORM, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_Cr_Y1_Cb_Y0_REP
// Red and blue swapped in the load shader for simplicity.
// TODO(Triang3l): The DXGI_FORMAT_R8G8B8A8_U/SNORM conversion is usable for
// the signed version, separate unsigned and signed load shaders completely
// (as one doesn't need decompression for this format, while another does).
{DXGI_FORMAT_G8R8_G8B8_UNORM, DXGI_FORMAT_G8R8_G8B8_UNORM,
kLoadShaderIndexGBGR8ToGRGB8, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
true, DXGI_FORMAT_R8G8B8A8_UNORM, kLoadShaderIndexGBGR8ToRGB8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_Y1_Cr_Y0_Cb_REP
// Red and blue swapped in the load shader for simplicity.
// TODO(Triang3l): The DXGI_FORMAT_R8G8B8A8_U/SNORM conversion is usable for
// the signed version, separate unsigned and signed load shaders completely
// (as one doesn't need decompression for this format, while another does).
{DXGI_FORMAT_R8G8_B8G8_UNORM, DXGI_FORMAT_R8G8_B8G8_UNORM,
kLoadShaderIndexBGRG8ToRGBG8, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
true, DXGI_FORMAT_R8G8B8A8_UNORM, kLoadShaderIndexBGRG8ToRGB8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_16_16_EDRAM
// Not usable as a texture, also has -32...32 range.
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_8_8_8_8_A
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_4_4_4_4
// Red and blue swapped in the load shader for simplicity.
{DXGI_FORMAT_B4G4R4A4_UNORM, DXGI_FORMAT_B4G4R4A4_UNORM,
kLoadShaderIndexRGBA4ToBGRA4, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
false, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_10_11_11
{DXGI_FORMAT_R16G16B16A16_TYPELESS, DXGI_FORMAT_R16G16B16A16_UNORM,
kLoadShaderIndexR11G11B10ToRGBA16, DXGI_FORMAT_R16G16B16A16_SNORM,
kLoadShaderIndexR11G11B10ToRGBA16SNorm, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_11_11_10
{DXGI_FORMAT_R16G16B16A16_TYPELESS, DXGI_FORMAT_R16G16B16A16_UNORM,
kLoadShaderIndexR10G11B11ToRGBA16, DXGI_FORMAT_R16G16B16A16_SNORM,
kLoadShaderIndexR10G11B11ToRGBA16SNorm, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_DXT1
{DXGI_FORMAT_BC1_UNORM, DXGI_FORMAT_BC1_UNORM, kLoadShaderIndex64bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true,
DXGI_FORMAT_R8G8B8A8_UNORM, kLoadShaderIndexDXT1ToRGBA8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_DXT2_3
{DXGI_FORMAT_BC2_UNORM, DXGI_FORMAT_BC2_UNORM, kLoadShaderIndex128bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true,
DXGI_FORMAT_R8G8B8A8_UNORM, kLoadShaderIndexDXT3ToRGBA8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_DXT4_5
{DXGI_FORMAT_BC3_UNORM, DXGI_FORMAT_BC3_UNORM, kLoadShaderIndex128bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true,
DXGI_FORMAT_R8G8B8A8_UNORM, kLoadShaderIndexDXT5ToRGBA8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_16_16_16_16_EDRAM
// Not usable as a texture, also has -32...32 range.
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// R32_FLOAT for depth because shaders would require an additional SRV to
// sample stencil, which we don't provide.
// k_24_8
{DXGI_FORMAT_R32_FLOAT, DXGI_FORMAT_R32_FLOAT, kLoadShaderIndexDepthUnorm,
DXGI_FORMAT_R32_FLOAT, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_24_8_FLOAT
{DXGI_FORMAT_R32_FLOAT, DXGI_FORMAT_R32_FLOAT, kLoadShaderIndexDepthFloat,
DXGI_FORMAT_R32_FLOAT, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_16
{DXGI_FORMAT_R16_TYPELESS, DXGI_FORMAT_R16_UNORM, kLoadShaderIndex16bpb,
DXGI_FORMAT_R16_SNORM, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_16_16
{DXGI_FORMAT_R16G16_TYPELESS, DXGI_FORMAT_R16G16_UNORM,
kLoadShaderIndex32bpb, DXGI_FORMAT_R16G16_SNORM, kLoadShaderIndexUnknown,
false, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_16_16_16_16
{DXGI_FORMAT_R16G16B16A16_TYPELESS, DXGI_FORMAT_R16G16B16A16_UNORM,
kLoadShaderIndex64bpb, DXGI_FORMAT_R16G16B16A16_SNORM,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_16_EXPAND
{DXGI_FORMAT_R16_FLOAT, DXGI_FORMAT_R16_FLOAT, kLoadShaderIndex16bpb,
DXGI_FORMAT_R16_FLOAT, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_16_16_EXPAND
{DXGI_FORMAT_R16G16_FLOAT, DXGI_FORMAT_R16G16_FLOAT, kLoadShaderIndex32bpb,
DXGI_FORMAT_R16G16_FLOAT, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_16_16_16_16_EXPAND
{DXGI_FORMAT_R16G16B16A16_FLOAT, DXGI_FORMAT_R16G16B16A16_FLOAT,
kLoadShaderIndex64bpb, DXGI_FORMAT_R16G16B16A16_FLOAT,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_16_FLOAT
{DXGI_FORMAT_R16_FLOAT, DXGI_FORMAT_R16_FLOAT, kLoadShaderIndex16bpb,
DXGI_FORMAT_R16_FLOAT, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_16_16_FLOAT
{DXGI_FORMAT_R16G16_FLOAT, DXGI_FORMAT_R16G16_FLOAT, kLoadShaderIndex32bpb,
DXGI_FORMAT_R16G16_FLOAT, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_16_16_16_16_FLOAT
{DXGI_FORMAT_R16G16B16A16_FLOAT, DXGI_FORMAT_R16G16B16A16_FLOAT,
kLoadShaderIndex64bpb, DXGI_FORMAT_R16G16B16A16_FLOAT,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_32
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_32_32
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_32_32_32_32
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_32_FLOAT
{DXGI_FORMAT_R32_FLOAT, DXGI_FORMAT_R32_FLOAT, kLoadShaderIndex32bpb,
DXGI_FORMAT_R32_FLOAT, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_32_32_FLOAT
{DXGI_FORMAT_R32G32_FLOAT, DXGI_FORMAT_R32G32_FLOAT, kLoadShaderIndex64bpb,
DXGI_FORMAT_R32G32_FLOAT, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_32_32_32_32_FLOAT
{DXGI_FORMAT_R32G32B32A32_FLOAT, DXGI_FORMAT_R32G32B32A32_FLOAT,
kLoadShaderIndex128bpb, DXGI_FORMAT_R32G32B32A32_FLOAT,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_32_AS_8
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_32_AS_8_8
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_16_MPEG
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_16_16_MPEG
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_8_INTERLACED
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_32_AS_8_INTERLACED
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_32_AS_8_8_INTERLACED
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_16_INTERLACED
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_16_MPEG_INTERLACED
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_16_16_MPEG_INTERLACED
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_DXN
{DXGI_FORMAT_BC5_UNORM, DXGI_FORMAT_BC5_UNORM, kLoadShaderIndex128bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true, DXGI_FORMAT_R8G8_UNORM,
kLoadShaderIndexDXNToRG8, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_8_8_8_8_AS_16_16_16_16
{DXGI_FORMAT_R8G8B8A8_TYPELESS, DXGI_FORMAT_R8G8B8A8_UNORM,
kLoadShaderIndex32bpb, DXGI_FORMAT_R8G8B8A8_SNORM, kLoadShaderIndexUnknown,
false, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_DXT1_AS_16_16_16_16
{DXGI_FORMAT_BC1_UNORM, DXGI_FORMAT_BC1_UNORM, kLoadShaderIndex64bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true,
DXGI_FORMAT_R8G8B8A8_UNORM, kLoadShaderIndexDXT1ToRGBA8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_DXT2_3_AS_16_16_16_16
{DXGI_FORMAT_BC2_UNORM, DXGI_FORMAT_BC2_UNORM, kLoadShaderIndex128bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true,
DXGI_FORMAT_R8G8B8A8_UNORM, kLoadShaderIndexDXT3ToRGBA8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_DXT4_5_AS_16_16_16_16
{DXGI_FORMAT_BC3_UNORM, DXGI_FORMAT_BC3_UNORM, kLoadShaderIndex128bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true,
DXGI_FORMAT_R8G8B8A8_UNORM, kLoadShaderIndexDXT5ToRGBA8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_2_10_10_10_AS_16_16_16_16
{DXGI_FORMAT_R10G10B10A2_UNORM, DXGI_FORMAT_R10G10B10A2_UNORM,
kLoadShaderIndex32bpb, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_10_11_11_AS_16_16_16_16
{DXGI_FORMAT_R16G16B16A16_TYPELESS, DXGI_FORMAT_R16G16B16A16_UNORM,
kLoadShaderIndexR11G11B10ToRGBA16, DXGI_FORMAT_R16G16B16A16_SNORM,
kLoadShaderIndexR11G11B10ToRGBA16SNorm, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_11_11_10_AS_16_16_16_16
{DXGI_FORMAT_R16G16B16A16_TYPELESS, DXGI_FORMAT_R16G16B16A16_UNORM,
kLoadShaderIndexR10G11B11ToRGBA16, DXGI_FORMAT_R16G16B16A16_SNORM,
kLoadShaderIndexR10G11B11ToRGBA16SNorm, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_32_32_32_FLOAT
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_DXT3A
// R8_UNORM has the same size as BC2, but doesn't have the 4x4 size
// alignment requirement.
{DXGI_FORMAT_R8_UNORM, DXGI_FORMAT_R8_UNORM, kLoadShaderIndexDXT3A,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_DXT5A
{DXGI_FORMAT_BC4_UNORM, DXGI_FORMAT_BC4_UNORM, kLoadShaderIndex64bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true, DXGI_FORMAT_R8_UNORM,
kLoadShaderIndexDXT5AToR8, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_CTX1
{DXGI_FORMAT_R8G8_UNORM, DXGI_FORMAT_R8G8_UNORM, kLoadShaderIndexCTX1,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_DXT3A_AS_1_1_1_1
{DXGI_FORMAT_B4G4R4A4_UNORM, DXGI_FORMAT_B4G4R4A4_UNORM,
kLoadShaderIndexDXT3AAs1111ToBGRA4, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_8_8_8_8_GAMMA_EDRAM
// Not usable as a texture.
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_2_10_10_10_FLOAT_EDRAM
// Not usable as a texture.
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
};
/*
chrispy: we're getting cache misses in GetHostFormatSwizzle, use a
denser array todo: not all 65536 possible swizzles are used, this could
probably be one cache line
*/
using SwizzleArray = std::array<unsigned short, 64>;
static constexpr SwizzleArray build_xenos_swizzle_for_format() {
SwizzleArray result{0};
for (int i = 0; i < 64; ++i) {
result[i] =
static_cast<uint16_t>(D3D12TextureCache::host_formats_[i].swizzle);
}
return result;
}
alignas(64) constexpr SwizzleArray xenos_swizzle_for_format =
build_xenos_swizzle_for_format();
D3D12TextureCache::D3D12TextureCache(const RegisterFile& register_file,
D3D12SharedMemory& shared_memory,
@@ -1544,7 +1254,8 @@ bool D3D12TextureCache::IsScaledResolveSupportedForFormat(
}
uint32_t D3D12TextureCache::GetHostFormatSwizzle(TextureKey key) const {
return host_formats_[uint32_t(key.format)].swizzle;
// return host_formats_[uint32_t(key.format)].swizzle;
return xenos_swizzle_for_format[uint32_t(key.format)];
}
uint32_t D3D12TextureCache::GetMaxHostTextureWidthHeight(

371
src/xenia/gpu/d3d12/d3d12_texture_cache.h
View File

@@ -160,29 +160,6 @@ class D3D12TextureCache final : public TextureCache {
ID3D12Resource* RequestSwapTexture(
D3D12_SHADER_RESOURCE_VIEW_DESC& srv_desc_out,
xenos::TextureFormat& format_out);
protected:
bool IsSignedVersionSeparateForFormat(TextureKey key) const override;
bool IsScaledResolveSupportedForFormat(TextureKey key) const override;
uint32_t GetHostFormatSwizzle(TextureKey key) const override;
uint32_t GetMaxHostTextureWidthHeight(
xenos::DataDimension dimension) const override;
uint32_t GetMaxHostTextureDepthOrArraySize(
xenos::DataDimension dimension) const override;
std::unique_ptr<Texture> CreateTexture(TextureKey key) override;
// This binds pipelines, allocates descriptors, and copies!
bool LoadTextureDataFromResidentMemoryImpl(Texture& texture, bool load_base,
bool load_mips) override;
void UpdateTextureBindingsImpl(uint32_t fetch_constant_mask) override;
private:
static constexpr uint32_t kLoadGuestXThreadsPerGroupLog2 = 2;
static constexpr uint32_t kLoadGuestYBlocksPerGroupLog2 = 5;
struct HostFormat {
// Format info for the regular case.
// DXGI format (typeless when different signedness or number representation
@@ -223,6 +200,352 @@ class D3D12TextureCache final : public TextureCache {
// Mapping of Xenos swizzle components to DXGI format components.
uint32_t swizzle;
};
static constexpr HostFormat host_formats_[64]{
// k_1_REVERSE
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_1
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_8
{DXGI_FORMAT_R8_TYPELESS, DXGI_FORMAT_R8_UNORM, kLoadShaderIndex8bpb,
DXGI_FORMAT_R8_SNORM, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_1_5_5_5
// Red and blue swapped in the load shader for simplicity.
{DXGI_FORMAT_B5G5R5A1_UNORM, DXGI_FORMAT_B5G5R5A1_UNORM,
kLoadShaderIndexR5G5B5A1ToB5G5R5A1, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_5_6_5
// Red and blue swapped in the load shader for simplicity.
{DXGI_FORMAT_B5G6R5_UNORM, DXGI_FORMAT_B5G6R5_UNORM,
kLoadShaderIndexR5G6B5ToB5G6R5, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_6_5_5
// On the host, green bits in blue, blue bits in green.
{DXGI_FORMAT_B5G6R5_UNORM, DXGI_FORMAT_B5G6R5_UNORM,
kLoadShaderIndexR5G5B6ToB5G6R5WithRBGASwizzle, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, XE_GPU_MAKE_TEXTURE_SWIZZLE(R, B, G, G)},
// k_8_8_8_8
{DXGI_FORMAT_R8G8B8A8_TYPELESS, DXGI_FORMAT_R8G8B8A8_UNORM,
kLoadShaderIndex32bpb, DXGI_FORMAT_R8G8B8A8_SNORM,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_2_10_10_10
{DXGI_FORMAT_R10G10B10A2_TYPELESS, DXGI_FORMAT_R10G10B10A2_UNORM,
kLoadShaderIndex32bpb, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
false, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_8_A
{DXGI_FORMAT_R8_TYPELESS, DXGI_FORMAT_R8_UNORM, kLoadShaderIndex8bpb,
DXGI_FORMAT_R8_SNORM, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_8_B
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_8_8
{DXGI_FORMAT_R8G8_TYPELESS, DXGI_FORMAT_R8G8_UNORM, kLoadShaderIndex16bpb,
DXGI_FORMAT_R8G8_SNORM, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_Cr_Y1_Cb_Y0_REP
// Red and blue swapped in the load shader for simplicity.
// TODO(Triang3l): The DXGI_FORMAT_R8G8B8A8_U/SNORM conversion is
// usable for
// the signed version, separate unsigned and signed load shaders
// completely
// (as one doesn't need decompression for this format, while another
// does).
{DXGI_FORMAT_G8R8_G8B8_UNORM, DXGI_FORMAT_G8R8_G8B8_UNORM,
kLoadShaderIndexGBGR8ToGRGB8, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, true, DXGI_FORMAT_R8G8B8A8_UNORM,
kLoadShaderIndexGBGR8ToRGB8, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_Y1_Cr_Y0_Cb_REP
// Red and blue swapped in the load shader for simplicity.
// TODO(Triang3l): The DXGI_FORMAT_R8G8B8A8_U/SNORM conversion is
// usable for
// the signed version, separate unsigned and signed load shaders
// completely
// (as one doesn't need decompression for this format, while another
// does).
{DXGI_FORMAT_R8G8_B8G8_UNORM, DXGI_FORMAT_R8G8_B8G8_UNORM,
kLoadShaderIndexBGRG8ToRGBG8, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, true, DXGI_FORMAT_R8G8B8A8_UNORM,
kLoadShaderIndexBGRG8ToRGB8, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_16_16_EDRAM
// Not usable as a texture, also has -32...32 range.
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_8_8_8_8_A
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_4_4_4_4
// Red and blue swapped in the load shader for simplicity.
{DXGI_FORMAT_B4G4R4A4_UNORM, DXGI_FORMAT_B4G4R4A4_UNORM,
kLoadShaderIndexRGBA4ToBGRA4, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_10_11_11
{DXGI_FORMAT_R16G16B16A16_TYPELESS, DXGI_FORMAT_R16G16B16A16_UNORM,
kLoadShaderIndexR11G11B10ToRGBA16, DXGI_FORMAT_R16G16B16A16_SNORM,
kLoadShaderIndexR11G11B10ToRGBA16SNorm, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_11_11_10
{DXGI_FORMAT_R16G16B16A16_TYPELESS, DXGI_FORMAT_R16G16B16A16_UNORM,
kLoadShaderIndexR10G11B11ToRGBA16, DXGI_FORMAT_R16G16B16A16_SNORM,
kLoadShaderIndexR10G11B11ToRGBA16SNorm, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_DXT1
{DXGI_FORMAT_BC1_UNORM, DXGI_FORMAT_BC1_UNORM, kLoadShaderIndex64bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true,
DXGI_FORMAT_R8G8B8A8_UNORM, kLoadShaderIndexDXT1ToRGBA8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_DXT2_3
{DXGI_FORMAT_BC2_UNORM, DXGI_FORMAT_BC2_UNORM, kLoadShaderIndex128bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true,
DXGI_FORMAT_R8G8B8A8_UNORM, kLoadShaderIndexDXT3ToRGBA8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_DXT4_5
{DXGI_FORMAT_BC3_UNORM, DXGI_FORMAT_BC3_UNORM, kLoadShaderIndex128bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true,
DXGI_FORMAT_R8G8B8A8_UNORM, kLoadShaderIndexDXT5ToRGBA8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_16_16_16_16_EDRAM
// Not usable as a texture, also has -32...32 range.
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// R32_FLOAT for depth because shaders would require an additional SRV
// to
// sample stencil, which we don't provide.
// k_24_8
{DXGI_FORMAT_R32_FLOAT, DXGI_FORMAT_R32_FLOAT, kLoadShaderIndexDepthUnorm,
DXGI_FORMAT_R32_FLOAT, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_24_8_FLOAT
{DXGI_FORMAT_R32_FLOAT, DXGI_FORMAT_R32_FLOAT, kLoadShaderIndexDepthFloat,
DXGI_FORMAT_R32_FLOAT, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_16
{DXGI_FORMAT_R16_TYPELESS, DXGI_FORMAT_R16_UNORM, kLoadShaderIndex16bpb,
DXGI_FORMAT_R16_SNORM, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_16_16
{DXGI_FORMAT_R16G16_TYPELESS, DXGI_FORMAT_R16G16_UNORM,
kLoadShaderIndex32bpb, DXGI_FORMAT_R16G16_SNORM, kLoadShaderIndexUnknown,
false, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_16_16_16_16
{DXGI_FORMAT_R16G16B16A16_TYPELESS, DXGI_FORMAT_R16G16B16A16_UNORM,
kLoadShaderIndex64bpb, DXGI_FORMAT_R16G16B16A16_SNORM,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_16_EXPAND
{DXGI_FORMAT_R16_FLOAT, DXGI_FORMAT_R16_FLOAT, kLoadShaderIndex16bpb,
DXGI_FORMAT_R16_FLOAT, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_16_16_EXPAND
{DXGI_FORMAT_R16G16_FLOAT, DXGI_FORMAT_R16G16_FLOAT,
kLoadShaderIndex32bpb, DXGI_FORMAT_R16G16_FLOAT, kLoadShaderIndexUnknown,
false, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_16_16_16_16_EXPAND
{DXGI_FORMAT_R16G16B16A16_FLOAT, DXGI_FORMAT_R16G16B16A16_FLOAT,
kLoadShaderIndex64bpb, DXGI_FORMAT_R16G16B16A16_FLOAT,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_16_FLOAT
{DXGI_FORMAT_R16_FLOAT, DXGI_FORMAT_R16_FLOAT, kLoadShaderIndex16bpb,
DXGI_FORMAT_R16_FLOAT, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_16_16_FLOAT
{DXGI_FORMAT_R16G16_FLOAT, DXGI_FORMAT_R16G16_FLOAT,
kLoadShaderIndex32bpb, DXGI_FORMAT_R16G16_FLOAT, kLoadShaderIndexUnknown,
false, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_16_16_16_16_FLOAT
{DXGI_FORMAT_R16G16B16A16_FLOAT, DXGI_FORMAT_R16G16B16A16_FLOAT,
kLoadShaderIndex64bpb, DXGI_FORMAT_R16G16B16A16_FLOAT,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_32
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_32_32
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_32_32_32_32
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_32_FLOAT
{DXGI_FORMAT_R32_FLOAT, DXGI_FORMAT_R32_FLOAT, kLoadShaderIndex32bpb,
DXGI_FORMAT_R32_FLOAT, kLoadShaderIndexUnknown, false,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_32_32_FLOAT
{DXGI_FORMAT_R32G32_FLOAT, DXGI_FORMAT_R32G32_FLOAT,
kLoadShaderIndex64bpb, DXGI_FORMAT_R32G32_FLOAT, kLoadShaderIndexUnknown,
false, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_32_32_32_32_FLOAT
{DXGI_FORMAT_R32G32B32A32_FLOAT, DXGI_FORMAT_R32G32B32A32_FLOAT,
kLoadShaderIndex128bpb, DXGI_FORMAT_R32G32B32A32_FLOAT,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_32_AS_8
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_32_AS_8_8
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_16_MPEG
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_16_16_MPEG
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_8_INTERLACED
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_32_AS_8_INTERLACED
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_32_AS_8_8_INTERLACED
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_16_INTERLACED
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_16_MPEG_INTERLACED
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_16_16_MPEG_INTERLACED
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_DXN
{DXGI_FORMAT_BC5_UNORM, DXGI_FORMAT_BC5_UNORM, kLoadShaderIndex128bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true,
DXGI_FORMAT_R8G8_UNORM, kLoadShaderIndexDXNToRG8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_8_8_8_8_AS_16_16_16_16
{DXGI_FORMAT_R8G8B8A8_TYPELESS, DXGI_FORMAT_R8G8B8A8_UNORM,
kLoadShaderIndex32bpb, DXGI_FORMAT_R8G8B8A8_SNORM,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_DXT1_AS_16_16_16_16
{DXGI_FORMAT_BC1_UNORM, DXGI_FORMAT_BC1_UNORM, kLoadShaderIndex64bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true,
DXGI_FORMAT_R8G8B8A8_UNORM, kLoadShaderIndexDXT1ToRGBA8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_DXT2_3_AS_16_16_16_16
{DXGI_FORMAT_BC2_UNORM, DXGI_FORMAT_BC2_UNORM, kLoadShaderIndex128bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true,
DXGI_FORMAT_R8G8B8A8_UNORM, kLoadShaderIndexDXT3ToRGBA8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_DXT4_5_AS_16_16_16_16
{DXGI_FORMAT_BC3_UNORM, DXGI_FORMAT_BC3_UNORM, kLoadShaderIndex128bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true,
DXGI_FORMAT_R8G8B8A8_UNORM, kLoadShaderIndexDXT5ToRGBA8,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_2_10_10_10_AS_16_16_16_16
{DXGI_FORMAT_R10G10B10A2_UNORM, DXGI_FORMAT_R10G10B10A2_UNORM,
kLoadShaderIndex32bpb, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
false, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_10_11_11_AS_16_16_16_16
{DXGI_FORMAT_R16G16B16A16_TYPELESS, DXGI_FORMAT_R16G16B16A16_UNORM,
kLoadShaderIndexR11G11B10ToRGBA16, DXGI_FORMAT_R16G16B16A16_SNORM,
kLoadShaderIndexR11G11B10ToRGBA16SNorm, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_11_11_10_AS_16_16_16_16
{DXGI_FORMAT_R16G16B16A16_TYPELESS, DXGI_FORMAT_R16G16B16A16_UNORM,
kLoadShaderIndexR10G11B11ToRGBA16, DXGI_FORMAT_R16G16B16A16_SNORM,
kLoadShaderIndexR10G11B11ToRGBA16SNorm, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_32_32_32_FLOAT
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBB},
// k_DXT3A
// R8_UNORM has the same size as BC2, but doesn't have the 4x4 size
// alignment requirement.
{DXGI_FORMAT_R8_UNORM, DXGI_FORMAT_R8_UNORM, kLoadShaderIndexDXT3A,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_DXT5A
{DXGI_FORMAT_BC4_UNORM, DXGI_FORMAT_BC4_UNORM, kLoadShaderIndex64bpb,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, true, DXGI_FORMAT_R8_UNORM,
kLoadShaderIndexDXT5AToR8, xenos::XE_GPU_TEXTURE_SWIZZLE_RRRR},
// k_CTX1
{DXGI_FORMAT_R8G8_UNORM, DXGI_FORMAT_R8G8_UNORM, kLoadShaderIndexCTX1,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGGG},
// k_DXT3A_AS_1_1_1_1
{DXGI_FORMAT_B4G4R4A4_UNORM, DXGI_FORMAT_B4G4R4A4_UNORM,
kLoadShaderIndexDXT3AAs1111ToBGRA4, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_8_8_8_8_GAMMA_EDRAM
// Not usable as a texture.
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
// k_2_10_10_10_FLOAT_EDRAM
// Not usable as a texture.
{DXGI_FORMAT_UNKNOWN, DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown,
DXGI_FORMAT_UNKNOWN, kLoadShaderIndexUnknown, false, DXGI_FORMAT_UNKNOWN,
kLoadShaderIndexUnknown, xenos::XE_GPU_TEXTURE_SWIZZLE_RGBA},
};
protected:
bool IsSignedVersionSeparateForFormat(TextureKey key) const override;
bool IsScaledResolveSupportedForFormat(TextureKey key) const override;
uint32_t GetHostFormatSwizzle(TextureKey key) const override;
uint32_t GetMaxHostTextureWidthHeight(
xenos::DataDimension dimension) const override;
uint32_t GetMaxHostTextureDepthOrArraySize(
xenos::DataDimension dimension) const override;
std::unique_ptr<Texture> CreateTexture(TextureKey key) override;
// This binds pipelines, allocates descriptors, and copies!
bool LoadTextureDataFromResidentMemoryImpl(Texture& texture, bool load_base,
bool load_mips) override;
void UpdateTextureBindingsImpl(uint32_t fetch_constant_mask) override;
private:
static constexpr uint32_t kLoadGuestXThreadsPerGroupLog2 = 2;
static constexpr uint32_t kLoadGuestYBlocksPerGroupLog2 = 5;
class D3D12Texture final : public Texture {
public:
@@ -467,8 +790,6 @@ class D3D12TextureCache final : public TextureCache {
xenos::ClampMode NormalizeClampMode(xenos::ClampMode clamp_mode) const;
static const HostFormat host_formats_[64];
D3D12CommandProcessor& command_processor_;
bool bindless_resources_used_;

2
src/xenia/gpu/graphics_system.cc
View File

@@ -198,7 +198,7 @@ uint32_t GraphicsSystem::ReadRegister(uint32_t addr) {
// maximum [width(0x0FFF), height(0x0FFF)]
return 0x050002D0;
default:
if (!register_file_.GetRegisterInfo(r)) {
if (!register_file_.IsValidRegister(r)) {
XELOGE("GPU: Read from unknown register ({:04X})", r);
}
}

60
src/xenia/gpu/register_file.cc
View File

@@ -8,7 +8,7 @@
*/
#include "xenia/gpu/register_file.h"
#include <array>
#include <cstring>
#include "xenia/base/math.h"
@@ -17,6 +17,52 @@ namespace xe {
namespace gpu {
RegisterFile::RegisterFile() { std::memset(values, 0, sizeof(values)); }
constexpr unsigned int GetHighestRegisterNumber() {
uint32_t highest = 0;
#define XE_GPU_REGISTER(index, type, name) \
highest = std::max<uint32_t>(highest, index);
#include "xenia/gpu/register_table.inc"
#undef XE_GPU_REGISTER
return highest;
}
constexpr unsigned int GetLowestRegisterNumber() {
uint32_t lowest = UINT_MAX;
#define XE_GPU_REGISTER(index, type, name) \
lowest = std::min<uint32_t>(lowest, index);
#include "xenia/gpu/register_table.inc"
#undef XE_GPU_REGISTER
return lowest;
}
static constexpr uint32_t lowest_register = GetLowestRegisterNumber();
static constexpr uint32_t highest_register = GetHighestRegisterNumber();
static constexpr uint32_t total_num_registers =
highest_register - lowest_register;
static constexpr uint32_t num_required_words_for_registers =
((total_num_registers + 63) & ~63) / 64;
// can't use bitset, its not constexpr in c++ 17
using ValidRegisterBitset = std::array<
uint64_t,
num_required_words_for_registers>; // std::bitset<highest_register
// - lowest_register>;
static constexpr ValidRegisterBitset BuildValidRegisterBitset() {
ValidRegisterBitset result{};
#define XE_GPU_REGISTER(index, type, name) \
result[(index - lowest_register) / 64] |= \
1ULL << ((index - lowest_register) % 64);
#include "xenia/gpu/register_table.inc"
#undef XE_GPU_REGISTER
return result;
}
static constexpr ValidRegisterBitset valid_register_bitset =
BuildValidRegisterBitset();
const RegisterInfo* RegisterFile::GetRegisterInfo(uint32_t index) {
switch (index) {
@@ -34,6 +80,18 @@ const RegisterInfo* RegisterFile::GetRegisterInfo(uint32_t index) {
return nullptr;
}
}
/*
todo: this still uses a lot of cpu! our bitset is too large
*/
bool RegisterFile::IsValidRegister(uint32_t index) {
if (XE_UNLIKELY(index < lowest_register) ||
XE_UNLIKELY(index > highest_register)) {
return false;
}
uint32_t register_linear_index = index - lowest_register;
return (valid_register_bitset[register_linear_index / 64] &
(1ULL << (register_linear_index % 64))) != 0;
}
} // namespace gpu
} // namespace xe

2
src/xenia/gpu/register_file.h
View File

@@ -32,7 +32,7 @@ class RegisterFile {
RegisterFile();
static const RegisterInfo* GetRegisterInfo(uint32_t index);
static bool IsValidRegister(uint32_t index);
static constexpr size_t kRegisterCount = 0x5003;
union RegisterValue {
uint32_t u32;

5
src/xenia/gpu/trace_viewer.cc
View File

@@ -41,9 +41,6 @@
#include "xenia/ui/windowed_app_context.h"
#include "xenia/xbox.h"
DEFINE_string(target_trace_file, "", "Specifies the trace file to load.",
"GPU");
namespace xe {
namespace gpu {
@@ -66,7 +63,7 @@ TraceViewer::TraceViewer(xe::ui::WindowedAppContext& app_context,
TraceViewer::~TraceViewer() = default;
bool TraceViewer::OnInitialize() {
std::string path = cvars::target_trace_file;
std::string path = cvars::target_trace_file.u8string();
// If no path passed, ask the user.
// On Android, however, there's no synchronous file picker, and the trace file

3
src/xenia/gpu/trace_viewer.h
View File

@@ -12,6 +12,7 @@
#include <string_view>
#include "xenia/base/cvar.h"
#include "xenia/emulator.h"
#include "xenia/gpu/shader.h"
#include "xenia/gpu/trace_player.h"
@@ -24,7 +25,7 @@
#include "xenia/ui/window.h"
#include "xenia/ui/window_listener.h"
#include "xenia/ui/windowed_app.h"
DECLARE_path(target_trace_file);
namespace xe {
namespace gpu {

4
src/xenia/gpu/trace_writer.cc
View File

@@ -25,7 +25,7 @@
namespace xe {
namespace gpu {
#if XE_ENABLE_TRACE_WRITER_INSTRUMENTATION == 1
TraceWriter::TraceWriter(uint8_t* membase)
: membase_(membase), file_(nullptr) {}
@@ -362,6 +362,6 @@ void TraceWriter::WriteGammaRamp(
fwrite(gamma_ramp_pwl_rgb, 1, kPWLUncompressedLength, file_);
}
}
#endif
} // namespace gpu
} // namespace xe

54
src/xenia/gpu/trace_writer.h
View File

@@ -17,11 +17,22 @@
#include "xenia/gpu/registers.h"
#include "xenia/gpu/trace_protocol.h"
// only enable trace writer in debug builds, measured hit from the trace
// function calls (even if they just immediately return) is 0.40-0.60% cpu time
// total. with inlining they just bloat the caller and negatively impact
// register allocation for the caller
#ifdef NDEBUG
#define XE_ENABLE_TRACE_WRITER_INSTRUMENTATION 0
#else
#define XE_ENABLE_TRACE_WRITER_INSTRUMENTATION 1
#endif
namespace xe {
namespace gpu {
class TraceWriter {
public:
#if XE_ENABLE_TRACE_WRITER_INSTRUMENTATION == 1
explicit TraceWriter(uint8_t* membase);
~TraceWriter();
@@ -61,6 +72,49 @@ class TraceWriter {
bool compress_output_ = true;
size_t compression_threshold_ = 1024; // Min. number of bytes to compress.
#else
// this could be annoying to maintain if new methods are added or the
// signatures change
constexpr explicit TraceWriter(uint8_t* membase) {}
static constexpr bool is_open() { return false; }
static constexpr bool Open(const std::filesystem::path& path,
uint32_t title_id) {
return false;
}
static constexpr void Flush() {}
static constexpr void Close() {}
static constexpr void WritePrimaryBufferStart(uint32_t base_ptr,
uint32_t count) {}
static constexpr void WritePrimaryBufferEnd() {}
static constexpr void WriteIndirectBufferStart(uint32_t base_ptr,
uint32_t count) {}
static constexpr void WriteIndirectBufferEnd() {}
static constexpr void WritePacketStart(uint32_t base_ptr, uint32_t count) {}
static constexpr void WritePacketEnd() {}
static constexpr void WriteMemoryRead(uint32_t base_ptr, size_t length,
const void* host_ptr = nullptr) {}
static constexpr void WriteMemoryReadCached(uint32_t base_ptr,
size_t length) {}
static constexpr void WriteMemoryReadCachedNop(uint32_t base_ptr,
size_t length) {}
static constexpr void WriteMemoryWrite(uint32_t base_ptr, size_t length,
const void* host_ptr = nullptr) {}
static constexpr void WriteEdramSnapshot(const void* snapshot) {}
static constexpr void WriteEvent(EventCommand::Type event_type) {}
static constexpr void WriteRegisters(uint32_t first_register,
const uint32_t* register_values,
uint32_t register_count,
bool execute_callbacks_on_play) {}
static constexpr void WriteGammaRamp(
const reg::DC_LUT_30_COLOR* gamma_ramp_256_entry_table,
const reg::DC_LUT_PWL_DATA* gamma_ramp_pwl_rgb,
uint32_t gamma_ramp_rw_component) {}
#endif
};
} // namespace gpu