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

[GPU] Mostly generic, not square-only resolution scaling

Browse Source
This commit is contained in:
Triang3l
2021-12-11 21:55:33 +03:00
parent e2da8597e1
commit 38b4741c8f
254 changed files with 86052 additions and 344263 deletions
Download Patch File Download Diff File Expand all files Collapse all files

101
src/xenia/gpu/dxbc_shader_translator_memexport.cc
View File

@@ -8,6 +8,7 @@
*/
#include "xenia/base/assert.h"
#include "xenia/base/math.h"
#include "xenia/gpu/draw_util.h"
#include "xenia/gpu/dxbc_shader_translator.h"
@@ -121,52 +122,74 @@ void DxbcShaderTranslator::ExportToMemory() {
// 0 or 0xFFFFFFFF.
bool inner_condition_provided = false;
if (is_pixel_shader()) {
if (draw_resolution_scale_ > 1) {
uint32_t resolution_scaled_axes =
uint32_t(draw_resolution_scale_x_ > 1) |
(uint32_t(draw_resolution_scale_y_ > 1) << 1);
if (resolution_scaled_axes) {
// Only do memexport for one host pixel in a guest pixel.
// For 2x - (1, 1) because it's covered with half-pixel offset that
// For 2x - pixel 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;
// For 3x - also pixel 1 because it's still covered with half-pixel
// offset, but close to the center.
// If X needs resolution scaling, writing 1 or 0 - whether the column is
// the one where memexport should be done - to control_temp.y.
// For Y, doing that to control_temp.z.
// Then, if both axes are resolution-scaled, merging the conditions for
// the two.
in_position_used_ |= resolution_scaled_axes;
a_.OpFToU(
dxbc::Dest::R(control_temp, 0b0110),
dxbc::Dest::R(control_temp, resolution_scaled_axes << 1),
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::Dest resolution_scaling_temp_dest(
dxbc::Dest::R(control_temp, 0b1000));
dxbc::Src resolution_scaling_temp_src(
dxbc::Src::R(control_temp, dxbc::Src::kWWWW));
for (uint32_t i = 0; i < 2; ++i) {
if (!(resolution_scaled_axes & (1 << i))) {
continue;
}
// If there's no inner condition in control_temp.x yet, the condition
// for the current axis can go directly to it. Otherwise, need to merge
// with the previous condition, using control_temp.w as an intermediate
// variable.
dxbc::Dest resolution_scaled_axis_result(
inner_condition_provided ? resolution_scaling_temp_dest
: dxbc::Dest::R(control_temp, 0b0001));
dxbc::Src resolution_scaled_axis_src(
dxbc::Src::R(control_temp).Select(1 + i));
uint32_t axis_resolution_scale =
i ? draw_resolution_scale_y_ : draw_resolution_scale_x_;
switch (axis_resolution_scale) {
case 2:
// xy & 1 == 1.
a_.OpAnd(resolution_scaled_axis_result, resolution_scaled_axis_src,
dxbc::Src::LU(1));
// No need to do IEq - already 1 for right / bottom, 0 for left /
// top.
break;
case 3:
// xy % 3 == 1.
a_.OpUMul(resolution_scaling_temp_dest, dxbc::Dest::Null(),
resolution_scaled_axis_src,
dxbc::Src::LU(draw_util::kDivideScale3));
a_.OpUShR(dxbc::Dest::R(control_temp, 0b1000),
dxbc::Src::R(control_temp, dxbc::Src::kWWWW),
a_.OpUShR(resolution_scaling_temp_dest, resolution_scaling_temp_src,
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_.OpIMAd(resolution_scaling_temp_dest, resolution_scaling_temp_src,
dxbc::Src::LI(-3), resolution_scaled_axis_src);
a_.OpIEq(resolution_scaled_axis_result, resolution_scaling_temp_src,
dxbc::Src::LU(1));
break;
default:
assert_unhandled_case(axis_resolution_scale);
}
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),
resolution_scaling_temp_src);
}
inner_condition_provided = true;
}
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),