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

[GPU] Ownership-transfer-based RT cache, 3x3 resolution scaling

Browse Source 
The ROV path is also disabled by default because of lower performance
This commit is contained in:
Triang3l
2021-04-26 22:12:09 +03:00
parent 30ea6e3ea3
commit 913e1e949c
362 changed files with 185259 additions and 38291 deletions
Download Patch File Download Diff File Expand all files Collapse all files

125
src/xenia/gpu/dxbc_shader_translator_memexport.cc
View File

@@ -7,6 +7,8 @@
******************************************************************************
*/
#include "xenia/base/assert.h"
#include "xenia/gpu/draw_util.h"
#include "xenia/gpu/dxbc_shader_translator.h"
namespace xe {
@@ -100,40 +102,94 @@ void DxbcShaderTranslator::ExportToMemory() {
// Safety check if the shared memory is bound as UAV.
system_constants_used_ |= 1ull << kSysConst_Flags_Index;
a_.OpAnd(dxbc::Dest::R(control_temp, 0b0001),
dxbc::Src::CB(cbuffer_index_system_constants_,
uint32_t(CbufferRegister::kSystemConstants),
kSysConst_Flags_Vec)
.Select(kSysConst_Flags_Comp),
dxbc::Src::LU(kSysFlag_SharedMemoryIsUAV));
a_.OpUBFE(dxbc::Dest::R(control_temp, 0b0001), dxbc::Src::LU(1),
dxbc::Src::LU(kSysFlag_SharedMemoryIsUAV_Shift),
dxbc::Src::CB(cbuffer_index_system_constants_,
uint32_t(CbufferRegister::kSystemConstants),
kSysConst_Flags_Vec)
.Select(kSysConst_Flags_Comp));
// Open the `if` with the uniform condition for the shared memory buffer being
// bound as a UAV (more fine-grained checks are vector and likely divergent).
a_.OpIf(true, dxbc::Src::R(control_temp, dxbc::Src::kXXXX));
// Check more fine-grained limitations.
// The flag in control_temp.x can be 0 or 1 for simplicity, not necessarily
// 0 or 0xFFFFFFFF.
bool inner_condition_provided = false;
if (is_pixel_shader()) {
// Disable memexport in pixel shaders with supersampling since VPOS is
// ambiguous.
if (edram_rov_used_) {
system_constants_used_ |= 1ull
<< kSysConst_EdramResolutionSquareScale_Index;
a_.OpULT(dxbc::Dest::R(control_temp, 0b0010),
dxbc::Src::CB(cbuffer_index_system_constants_,
uint32_t(CbufferRegister::kSystemConstants),
kSysConst_EdramResolutionSquareScale_Vec)
.Select(kSysConst_EdramResolutionSquareScale_Comp),
dxbc::Src::LU(2));
a_.OpAnd(dxbc::Dest::R(control_temp, 0b0001),
dxbc::Src::R(control_temp, dxbc::Src::kXXXX),
dxbc::Src::R(control_temp, dxbc::Src::kYYYY));
} else {
// Enough to check just Y because it's scaled for both 2x and 4x.
system_constants_used_ |= 1ull << kSysConst_SampleCountLog2_Index;
a_.OpMovC(dxbc::Dest::R(control_temp, 0b0001),
dxbc::Src::CB(cbuffer_index_system_constants_,
uint32_t(CbufferRegister::kSystemConstants),
kSysConst_SampleCountLog2_Vec)
.Select(kSysConst_SampleCountLog2_Comp + 1),
dxbc::Src::LU(0), dxbc::Src::R(control_temp, dxbc::Src::kXXXX));
if (draw_resolution_scale_ > 1) {
// Only do memexport for one host pixel in a guest pixel.
// For 2x - (1, 1) because it's covered with half-pixel offset that
// becomes full-pixel.
// For 3x - also (1, 1) because it's still covered with half-pixel offset,
// but close to the center.
in_position_used_ |= 0b0011;
a_.OpFToU(
dxbc::Dest::R(control_temp, 0b0110),
dxbc::Src::V(uint32_t(InOutRegister::kPSInPosition), 0b0100 << 2));
switch (draw_resolution_scale_) {
case 2:
a_.OpAnd(dxbc::Dest::R(control_temp, 0b0110),
dxbc::Src::R(control_temp), dxbc::Src::LU(1));
// No need to do IEq - already 1 for right / bottom, 0 for left / top.
break;
case 3:
// xy % 3 == 1.
for (uint32_t i = 1; i <= 2; ++i) {
a_.OpUMul(dxbc::Dest::R(control_temp, 0b1000), dxbc::Dest::Null(),
dxbc::Src::R(control_temp).Select(i),
dxbc::Src::LU(draw_util::kDivideScale3));
a_.OpUShR(dxbc::Dest::R(control_temp, 0b1000),
dxbc::Src::R(control_temp, dxbc::Src::kWWWW),
dxbc::Src::LU(draw_util::kDivideUpperShift3));
a_.OpIMAd(dxbc::Dest::R(control_temp, 1 << i),
dxbc::Src::R(control_temp, dxbc::Src::kWWWW),
dxbc::Src::LI(-3), dxbc::Src::R(control_temp).Select(i));
}
a_.OpIEq(dxbc::Dest::R(control_temp, 0b0110),
dxbc::Src::R(control_temp), dxbc::Src::LU(1));
break;
default:
assert_unhandled_case(draw_resolution_scale_);
}
a_.OpAnd(dxbc::Dest::R(control_temp,
inner_condition_provided ? 0b0010 : 0b0001),
dxbc::Src::R(control_temp, dxbc::Src::kYYYY),
dxbc::Src::R(control_temp, dxbc::Src::kZZZZ));
if (inner_condition_provided) {
// Merge with the previous condition in control_temp.x.
a_.OpAnd(dxbc::Dest::R(control_temp, 0b0001),
dxbc::Src::R(control_temp, dxbc::Src::kXXXX),
dxbc::Src::R(control_temp, dxbc::Src::kYYYY));
}
inner_condition_provided = true;
}
// With sample-rate shading (with float24 conversion), only do memexport
// from one sample (as the shader is invoked multiple times for a pixel),
// if SV_SampleIndex == firstbit_lo(SV_Coverage). For zero coverage,
// firstbit_lo returns 0xFFFFFFFF.
if (IsSampleRate()) {
a_.OpFirstBitLo(dxbc::Dest::R(control_temp, 0b0010),
dxbc::Src::VCoverage());
a_.OpIEq(
dxbc::Dest::R(control_temp,
inner_condition_provided ? 0b0010 : 0b0001),
dxbc::Src::V(uint32_t(InOutRegister::kPSInFrontFaceAndSampleIndex),
dxbc::Src::kYYYY),
dxbc::Src::R(control_temp, dxbc::Src::kYYYY));
if (inner_condition_provided) {
// Merge with the previous condition in control_temp.x.
a_.OpAnd(dxbc::Dest::R(control_temp, 0b0001),
dxbc::Src::R(control_temp, dxbc::Src::kXXXX),
dxbc::Src::R(control_temp, dxbc::Src::kYYYY));
}
inner_condition_provided = true;
}
}
// Check if memexport can be done.
a_.OpIf(true, dxbc::Src::R(control_temp, dxbc::Src::kXXXX));
// Open the inner (vector) conditional if needed.
if (inner_condition_provided) {
a_.OpIf(true, dxbc::Src::R(control_temp, dxbc::Src::kXXXX));
}
// control_temp.x is now free.
for (uint32_t i = 0; i < Shader::kMaxMemExports; ++i) {
@@ -462,7 +518,12 @@ void DxbcShaderTranslator::ExportToMemory() {
// control_temp.y is now free.
}
// Close the memexport possibility check.
// Close the inner memexport possibility conditional.
if (inner_condition_provided) {
a_.OpEndIf();
}
// Close the outer memexport possibility conditional.
a_.OpEndIf();
// Release control_temp.