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

[DXBC] dxbc.h with non-translator-specific parts

This commit is contained in:
Triang3l 2021-01-04 16:15:34 +03:00
parent 2d9326e02d
commit 9a74df491f
7 changed files with 4784 additions and 4772 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1611
src/xenia/gpu/dxbc.h Normal file
View File

File diff suppressed because it is too large Load Diff

1388
src/xenia/gpu/dxbc_shader_translator.cc
View File

File diff suppressed because it is too large Load Diff

1603
src/xenia/gpu/dxbc_shader_translator.h
View File

File diff suppressed because it is too large Load Diff

934
src/xenia/gpu/dxbc_shader_translator_alu.cc
View File

File diff suppressed because it is too large Load Diff

1127
src/xenia/gpu/dxbc_shader_translator_fetch.cc
View File

File diff suppressed because it is too large Load Diff

381
src/xenia/gpu/dxbc_shader_translator_memexport.cc
View File

@ -15,7 +15,7 @@ using namespace ucode;
void DxbcShaderTranslator::ExportToMemory_PackFixed32( void DxbcShaderTranslator::ExportToMemory_PackFixed32(
const uint32_t* eM_temps, uint32_t eM_count, const uint32_t bits[4], const uint32_t* eM_temps, uint32_t eM_count, const uint32_t bits[4],
const DxbcSrc& is_integer, const DxbcSrc& is_signed) { const dxbc::Src& is_integer, const dxbc::Src& is_signed) {
// Will insert with BFI - sign extension of red will be overwritten, not // Will insert with BFI - sign extension of red will be overwritten, not
// truncated. // truncated.
assert_not_zero(bits[0]); assert_not_zero(bits[0]);
@ -26,64 +26,64 @@ void DxbcShaderTranslator::ExportToMemory_PackFixed32(
mask |= 1 << i; mask |= 1 << i;
} }
} }
DxbcOpIf(true, is_signed); a_.OpIf(true, is_signed);
{ {
float range[4]; float range[4];
for (uint32_t i = 0; i < 4; ++i) { for (uint32_t i = 0; i < 4; ++i) {
range[i] = bits[i] ? float((uint32_t(1) << (bits[i] - 1)) - 1) : 0.0f; range[i] = bits[i] ? float((uint32_t(1) << (bits[i] - 1)) - 1) : 0.0f;
} }
DxbcSrc range_src(DxbcSrc::LP(range)); dxbc::Src range_src(dxbc::Src::LP(range));
DxbcOpIf(false, is_integer); a_.OpIf(false, is_integer);
for (uint32_t i = 0; i < eM_count; ++i) { for (uint32_t i = 0; i < eM_count; ++i) {
uint32_t eM_temp = eM_temps[i]; uint32_t eM_temp = eM_temps[i];
DxbcOpMul(DxbcDest::R(eM_temp, mask), DxbcSrc::R(eM_temp), range_src); a_.OpMul(dxbc::Dest::R(eM_temp, mask), dxbc::Src::R(eM_temp), range_src);
} }
DxbcOpEndIf(); a_.OpEndIf();
for (uint32_t i = 0; i < eM_count; ++i) { for (uint32_t i = 0; i < eM_count; ++i) {
DxbcDest eM_dest(DxbcDest::R(eM_temps[i], mask)); dxbc::Dest eM_dest(dxbc::Dest::R(eM_temps[i], mask));
DxbcSrc eM_src(DxbcSrc::R(eM_temps[i])); dxbc::Src eM_src(dxbc::Src::R(eM_temps[i]));
// TODO(Triang3l): NaN should become zero, not -range. // TODO(Triang3l): NaN should become zero, not -range.
DxbcOpMax(eM_dest, eM_src, -range_src); a_.OpMax(eM_dest, eM_src, -range_src);
DxbcOpMin(eM_dest, eM_src, range_src); a_.OpMin(eM_dest, eM_src, range_src);
} }
} }
DxbcOpElse(); a_.OpElse();
{ {
float range[4]; float range[4];
for (uint32_t i = 0; i < 4; ++i) { for (uint32_t i = 0; i < 4; ++i) {
range[i] = float((uint32_t(1) << bits[i]) - 1); range[i] = float((uint32_t(1) << bits[i]) - 1);
} }
DxbcSrc range_src(DxbcSrc::LP(range)); dxbc::Src range_src(dxbc::Src::LP(range));
DxbcOpIf(false, is_integer); a_.OpIf(false, is_integer);
for (uint32_t i = 0; i < eM_count; ++i) { for (uint32_t i = 0; i < eM_count; ++i) {
uint32_t eM_temp = eM_temps[i]; uint32_t eM_temp = eM_temps[i];
DxbcOpMul(DxbcDest::R(eM_temp, mask), DxbcSrc::R(eM_temp), range_src); a_.OpMul(dxbc::Dest::R(eM_temp, mask), dxbc::Src::R(eM_temp), range_src);
} }
DxbcOpEndIf(); a_.OpEndIf();
for (uint32_t i = 0; i < eM_count; ++i) { for (uint32_t i = 0; i < eM_count; ++i) {
DxbcDest eM_dest(DxbcDest::R(eM_temps[i], mask)); dxbc::Dest eM_dest(dxbc::Dest::R(eM_temps[i], mask));
DxbcSrc eM_src(DxbcSrc::R(eM_temps[i])); dxbc::Src eM_src(dxbc::Src::R(eM_temps[i]));
DxbcOpMax(eM_dest, eM_src, DxbcSrc::LF(0.0f)); a_.OpMax(eM_dest, eM_src, dxbc::Src::LF(0.0f));
DxbcOpMin(eM_dest, eM_src, range_src); a_.OpMin(eM_dest, eM_src, range_src);
} }
} }
DxbcOpEndIf(); a_.OpEndIf();
for (uint32_t i = 0; i < eM_count; ++i) { for (uint32_t i = 0; i < eM_count; ++i) {
uint32_t eM_temp = eM_temps[i]; uint32_t eM_temp = eM_temps[i];
// Round to the nearest integer, according to the rules of handling integer // Round to the nearest integer, according to the rules of handling integer
// formats in Direct3D. // formats in Direct3D.
// TODO(Triang3l): Round by adding +-0.5, not with round_ne. // TODO(Triang3l): Round by adding +-0.5, not with round_ne.
DxbcOpRoundNE(DxbcDest::R(eM_temp, mask), DxbcSrc::R(eM_temp)); a_.OpRoundNE(dxbc::Dest::R(eM_temp, mask), dxbc::Src::R(eM_temp));
DxbcOpFToI(DxbcDest::R(eM_temp, mask), DxbcSrc::R(eM_temp)); a_.OpFToI(dxbc::Dest::R(eM_temp, mask), dxbc::Src::R(eM_temp));
DxbcDest eM_packed_dest(DxbcDest::R(eM_temp, 0b0001)); dxbc::Dest eM_packed_dest(dxbc::Dest::R(eM_temp, 0b0001));
DxbcSrc eM_packed_src(DxbcSrc::R(eM_temp, DxbcSrc::kXXXX)); dxbc::Src eM_packed_src(dxbc::Src::R(eM_temp, dxbc::Src::kXXXX));
uint32_t offset = bits[0]; uint32_t offset = bits[0];
for (uint32_t j = 1; j < 4; ++j) { for (uint32_t j = 1; j < 4; ++j) {
if (!bits[j]) { if (!bits[j]) {
continue; continue;
} }
DxbcOpBFI(eM_packed_dest, DxbcSrc::LU(bits[j]), DxbcSrc::LU(offset), a_.OpBFI(eM_packed_dest, dxbc::Src::LU(bits[j]), dxbc::Src::LU(offset),
DxbcSrc::R(eM_temp).Select(j), eM_packed_src); dxbc::Src::R(eM_temp).Select(j), eM_packed_src);
offset += bits[j]; offset += bits[j];
} }
} }
@ -100,40 +100,40 @@ void DxbcShaderTranslator::ExportToMemory() {
// Safety check if the shared memory is bound as UAV. // Safety check if the shared memory is bound as UAV.
system_constants_used_ |= 1ull << kSysConst_Flags_Index; system_constants_used_ |= 1ull << kSysConst_Flags_Index;
DxbcOpAnd(DxbcDest::R(control_temp, 0b0001), a_.OpAnd(dxbc::Dest::R(control_temp, 0b0001),
DxbcSrc::CB(cbuffer_index_system_constants_, dxbc::Src::CB(cbuffer_index_system_constants_,
uint32_t(CbufferRegister::kSystemConstants), uint32_t(CbufferRegister::kSystemConstants),
kSysConst_Flags_Vec) kSysConst_Flags_Vec)
.Select(kSysConst_Flags_Comp), .Select(kSysConst_Flags_Comp),
DxbcSrc::LU(kSysFlag_SharedMemoryIsUAV)); dxbc::Src::LU(kSysFlag_SharedMemoryIsUAV));
if (is_pixel_shader()) { if (is_pixel_shader()) {
// Disable memexport in pixel shaders with supersampling since VPOS is // Disable memexport in pixel shaders with supersampling since VPOS is
// ambiguous. // ambiguous.
if (edram_rov_used_) { if (edram_rov_used_) {
system_constants_used_ |= 1ull system_constants_used_ |= 1ull
<< kSysConst_EdramResolutionSquareScale_Index; << kSysConst_EdramResolutionSquareScale_Index;
DxbcOpULT(DxbcDest::R(control_temp, 0b0010), a_.OpULT(dxbc::Dest::R(control_temp, 0b0010),
DxbcSrc::CB(cbuffer_index_system_constants_, dxbc::Src::CB(cbuffer_index_system_constants_,
uint32_t(CbufferRegister::kSystemConstants), uint32_t(CbufferRegister::kSystemConstants),
kSysConst_EdramResolutionSquareScale_Vec) kSysConst_EdramResolutionSquareScale_Vec)
.Select(kSysConst_EdramResolutionSquareScale_Comp), .Select(kSysConst_EdramResolutionSquareScale_Comp),
DxbcSrc::LU(2)); dxbc::Src::LU(2));
DxbcOpAnd(DxbcDest::R(control_temp, 0b0001), a_.OpAnd(dxbc::Dest::R(control_temp, 0b0001),
DxbcSrc::R(control_temp, DxbcSrc::kXXXX), dxbc::Src::R(control_temp, dxbc::Src::kXXXX),
DxbcSrc::R(control_temp, DxbcSrc::kYYYY)); dxbc::Src::R(control_temp, dxbc::Src::kYYYY));
} else { } else {
// Enough to check just Y because it's scaled for both 2x and 4x. // Enough to check just Y because it's scaled for both 2x and 4x.
system_constants_used_ |= 1ull << kSysConst_SampleCountLog2_Index; system_constants_used_ |= 1ull << kSysConst_SampleCountLog2_Index;
DxbcOpMovC(DxbcDest::R(control_temp, 0b0001), a_.OpMovC(dxbc::Dest::R(control_temp, 0b0001),
DxbcSrc::CB(cbuffer_index_system_constants_, dxbc::Src::CB(cbuffer_index_system_constants_,
uint32_t(CbufferRegister::kSystemConstants), uint32_t(CbufferRegister::kSystemConstants),
kSysConst_SampleCountLog2_Vec) kSysConst_SampleCountLog2_Vec)
.Select(kSysConst_SampleCountLog2_Comp + 1), .Select(kSysConst_SampleCountLog2_Comp + 1),
DxbcSrc::LU(0), DxbcSrc::R(control_temp, DxbcSrc::kXXXX)); dxbc::Src::LU(0), dxbc::Src::R(control_temp, dxbc::Src::kXXXX));
} }
} }
// Check if memexport can be done. // Check if memexport can be done.
DxbcOpIf(true, DxbcSrc::R(control_temp, DxbcSrc::kXXXX)); a_.OpIf(true, dxbc::Src::R(control_temp, dxbc::Src::kXXXX));
// control_temp.x is now free. // control_temp.x is now free.
for (uint32_t i = 0; i < Shader::kMaxMemExports; ++i) { for (uint32_t i = 0; i < Shader::kMaxMemExports; ++i) {
@ -160,21 +160,21 @@ void DxbcShaderTranslator::ExportToMemory() {
} }
// Swap red and blue if needed. // Swap red and blue if needed.
DxbcOpAnd(DxbcDest::R(control_temp, 0b0001), a_.OpAnd(dxbc::Dest::R(control_temp, 0b0001),
DxbcSrc::R(eA_temp, DxbcSrc::kZZZZ), dxbc::Src::R(eA_temp, dxbc::Src::kZZZZ),
DxbcSrc::LU(uint32_t(1) << 19)); dxbc::Src::LU(uint32_t(1) << 19));
for (uint32_t j = 0; j < eM_count; ++j) { for (uint32_t j = 0; j < eM_count; ++j) {
uint32_t eM_temp = eM_temps[j]; uint32_t eM_temp = eM_temps[j];
DxbcOpMovC(DxbcDest::R(eM_temp, 0b0101), a_.OpMovC(dxbc::Dest::R(eM_temp, 0b0101),
DxbcSrc::R(control_temp, DxbcSrc::kXXXX), dxbc::Src::R(control_temp, dxbc::Src::kXXXX),
DxbcSrc::R(eM_temp, 0b000010), DxbcSrc::R(eM_temp)); dxbc::Src::R(eM_temp, 0b000010), dxbc::Src::R(eM_temp));
} }
// Initialize element size in control_temp.x to 4 bytes as this is the most // Initialize element size in control_temp.x to 4 bytes as this is the most
// common size. // common size.
DxbcDest element_size_dest(DxbcDest::R(control_temp, 0b0001)); dxbc::Dest element_size_dest(dxbc::Dest::R(control_temp, 0b0001));
DxbcSrc element_size_src(DxbcSrc::R(control_temp, DxbcSrc::kXXXX)); dxbc::Src element_size_src(dxbc::Src::R(control_temp, dxbc::Src::kXXXX));
DxbcOpMov(element_size_dest, DxbcSrc::LU(4)); a_.OpMov(element_size_dest, dxbc::Src::LU(4));
// Each eM should get a packed value in the destination format now. // Each eM should get a packed value in the destination format now.
@ -182,285 +182,288 @@ void DxbcShaderTranslator::ExportToMemory() {
// Y - signedness if fixed-point. // Y - signedness if fixed-point.
// Z - fractional/integer if fixed-point. // Z - fractional/integer if fixed-point.
// W - color format. // W - color format.
DxbcOpUBFE(DxbcDest::R(control_temp, 0b1110), DxbcSrc::LU(0, 1, 1, 6), a_.OpUBFE(dxbc::Dest::R(control_temp, 0b1110), dxbc::Src::LU(0, 1, 1, 6),
DxbcSrc::LU(0, 16, 17, 8), DxbcSrc::R(eA_temp, DxbcSrc::kZZZZ)); dxbc::Src::LU(0, 16, 17, 8),
DxbcSrc is_signed(DxbcSrc::R(control_temp, DxbcSrc::kYYYY)); dxbc::Src::R(eA_temp, dxbc::Src::kZZZZ));
DxbcSrc is_integer(DxbcSrc::R(control_temp, DxbcSrc::kZZZZ)); dxbc::Src is_signed(dxbc::Src::R(control_temp, dxbc::Src::kYYYY));
dxbc::Src is_integer(dxbc::Src::R(control_temp, dxbc::Src::kZZZZ));
// Convert and pack the format. // Convert and pack the format.
DxbcOpSwitch(DxbcSrc::R(control_temp, DxbcSrc::kWWWW)); a_.OpSwitch(dxbc::Src::R(control_temp, dxbc::Src::kWWWW));
// control_temp.w is now free. // control_temp.w is now free.
{ {
// k_8_8_8_8 // k_8_8_8_8
// k_8_8_8_8_AS_16_16_16_16 // k_8_8_8_8_AS_16_16_16_16
DxbcOpCase(DxbcSrc::LU(uint32_t(xenos::ColorFormat::k_8_8_8_8))); a_.OpCase(dxbc::Src::LU(uint32_t(xenos::ColorFormat::k_8_8_8_8)));
DxbcOpCase( a_.OpCase(dxbc::Src::LU(
DxbcSrc::LU(uint32_t(xenos::ColorFormat::k_8_8_8_8_AS_16_16_16_16))); uint32_t(xenos::ColorFormat::k_8_8_8_8_AS_16_16_16_16)));
{ {
uint32_t bits[4] = {8, 8, 8, 8}; uint32_t bits[4] = {8, 8, 8, 8};
ExportToMemory_PackFixed32(eM_temps, eM_count, bits, is_integer, ExportToMemory_PackFixed32(eM_temps, eM_count, bits, is_integer,
is_signed); is_signed);
} }
DxbcOpBreak(); a_.OpBreak();
// k_2_10_10_10 // k_2_10_10_10
// k_2_10_10_10_AS_16_16_16_16 // k_2_10_10_10_AS_16_16_16_16
DxbcOpCase(DxbcSrc::LU(uint32_t(xenos::ColorFormat::k_2_10_10_10))); a_.OpCase(dxbc::Src::LU(uint32_t(xenos::ColorFormat::k_2_10_10_10)));
DxbcOpCase(DxbcSrc::LU( a_.OpCase(dxbc::Src::LU(
uint32_t(xenos::ColorFormat::k_2_10_10_10_AS_16_16_16_16))); uint32_t(xenos::ColorFormat::k_2_10_10_10_AS_16_16_16_16)));
{ {
uint32_t bits[4] = {10, 10, 10, 2}; uint32_t bits[4] = {10, 10, 10, 2};
ExportToMemory_PackFixed32(eM_temps, eM_count, bits, is_integer, ExportToMemory_PackFixed32(eM_temps, eM_count, bits, is_integer,
is_signed); is_signed);
} }
DxbcOpBreak(); a_.OpBreak();
// k_10_11_11 // k_10_11_11
// k_10_11_11_AS_16_16_16_16 // k_10_11_11_AS_16_16_16_16
DxbcOpCase(DxbcSrc::LU(uint32_t(xenos::ColorFormat::k_10_11_11))); a_.OpCase(dxbc::Src::LU(uint32_t(xenos::ColorFormat::k_10_11_11)));
DxbcOpCase( a_.OpCase(dxbc::Src::LU(
DxbcSrc::LU(uint32_t(xenos::ColorFormat::k_10_11_11_AS_16_16_16_16))); uint32_t(xenos::ColorFormat::k_10_11_11_AS_16_16_16_16)));
{ {
uint32_t bits[4] = {11, 11, 10}; uint32_t bits[4] = {11, 11, 10};
ExportToMemory_PackFixed32(eM_temps, eM_count, bits, is_integer, ExportToMemory_PackFixed32(eM_temps, eM_count, bits, is_integer,
is_signed); is_signed);
} }
DxbcOpBreak(); a_.OpBreak();
// k_11_11_10 // k_11_11_10
// k_11_11_10_AS_16_16_16_16 // k_11_11_10_AS_16_16_16_16
DxbcOpCase(DxbcSrc::LU(uint32_t(xenos::ColorFormat::k_11_11_10))); a_.OpCase(dxbc::Src::LU(uint32_t(xenos::ColorFormat::k_11_11_10)));
DxbcOpCase( a_.OpCase(dxbc::Src::LU(
DxbcSrc::LU(uint32_t(xenos::ColorFormat::k_11_11_10_AS_16_16_16_16))); uint32_t(xenos::ColorFormat::k_11_11_10_AS_16_16_16_16)));
{ {
uint32_t bits[4] = {10, 11, 11}; uint32_t bits[4] = {10, 11, 11};
ExportToMemory_PackFixed32(eM_temps, eM_count, bits, is_integer, ExportToMemory_PackFixed32(eM_temps, eM_count, bits, is_integer,
is_signed); is_signed);
} }
DxbcOpBreak(); a_.OpBreak();
// k_16_16 // k_16_16
DxbcOpCase(DxbcSrc::LU(uint32_t(xenos::ColorFormat::k_16_16))); a_.OpCase(dxbc::Src::LU(uint32_t(xenos::ColorFormat::k_16_16)));
{ {
uint32_t bits[4] = {16, 16}; uint32_t bits[4] = {16, 16};
ExportToMemory_PackFixed32(eM_temps, eM_count, bits, is_integer, ExportToMemory_PackFixed32(eM_temps, eM_count, bits, is_integer,
is_signed); is_signed);
} }
DxbcOpBreak(); a_.OpBreak();
// k_16_16_16_16 // k_16_16_16_16
DxbcOpCase(DxbcSrc::LU(uint32_t(xenos::ColorFormat::k_16_16_16_16))); a_.OpCase(dxbc::Src::LU(uint32_t(xenos::ColorFormat::k_16_16_16_16)));
DxbcOpMov(element_size_dest, DxbcSrc::LU(8)); a_.OpMov(element_size_dest, dxbc::Src::LU(8));
DxbcOpIf(true, is_signed); a_.OpIf(true, is_signed);
{ {
DxbcOpIf(false, is_integer); a_.OpIf(false, is_integer);
for (uint32_t j = 0; j < eM_count; ++j) { for (uint32_t j = 0; j < eM_count; ++j) {
uint32_t eM_temp = eM_temps[j]; uint32_t eM_temp = eM_temps[j];
DxbcOpMul(DxbcDest::R(eM_temp), DxbcSrc::R(eM_temp), a_.OpMul(dxbc::Dest::R(eM_temp), dxbc::Src::R(eM_temp),
DxbcSrc::LF(32767.0f)); dxbc::Src::LF(32767.0f));
} }
DxbcOpEndIf(); a_.OpEndIf();
for (uint32_t j = 0; j < eM_count; ++j) { for (uint32_t j = 0; j < eM_count; ++j) {
DxbcDest eM_dest(DxbcDest::R(eM_temps[j])); dxbc::Dest eM_dest(dxbc::Dest::R(eM_temps[j]));
DxbcSrc eM_src(DxbcSrc::R(eM_temps[j])); dxbc::Src eM_src(dxbc::Src::R(eM_temps[j]));
// TODO(Triang3l): NaN should become zero, not -range. // TODO(Triang3l): NaN should become zero, not -range.
DxbcOpMax(eM_dest, eM_src, DxbcSrc::LF(-32767.0f)); a_.OpMax(eM_dest, eM_src, dxbc::Src::LF(-32767.0f));
DxbcOpMin(eM_dest, eM_src, DxbcSrc::LF(32767.0f)); a_.OpMin(eM_dest, eM_src, dxbc::Src::LF(32767.0f));
} }
} }
DxbcOpElse(); a_.OpElse();
{ {
DxbcOpIf(false, is_integer); a_.OpIf(false, is_integer);
for (uint32_t j = 0; j < eM_count; ++j) { for (uint32_t j = 0; j < eM_count; ++j) {
uint32_t eM_temp = eM_temps[j]; uint32_t eM_temp = eM_temps[j];
DxbcOpMul(DxbcDest::R(eM_temp), DxbcSrc::R(eM_temp), a_.OpMul(dxbc::Dest::R(eM_temp), dxbc::Src::R(eM_temp),
DxbcSrc::LF(65535.0f)); dxbc::Src::LF(65535.0f));
} }
DxbcOpEndIf(); a_.OpEndIf();
for (uint32_t j = 0; j < eM_count; ++j) { for (uint32_t j = 0; j < eM_count; ++j) {
DxbcDest eM_dest(DxbcDest::R(eM_temps[j])); dxbc::Dest eM_dest(dxbc::Dest::R(eM_temps[j]));
DxbcSrc eM_src(DxbcSrc::R(eM_temps[j])); dxbc::Src eM_src(dxbc::Src::R(eM_temps[j]));
DxbcOpMax(eM_dest, eM_src, DxbcSrc::LF(0.0f)); a_.OpMax(eM_dest, eM_src, dxbc::Src::LF(0.0f));
DxbcOpMin(eM_dest, eM_src, DxbcSrc::LF(65535.0f)); a_.OpMin(eM_dest, eM_src, dxbc::Src::LF(65535.0f));
} }
} }
DxbcOpEndIf(); a_.OpEndIf();
for (uint32_t j = 0; j < eM_count; ++j) { for (uint32_t j = 0; j < eM_count; ++j) {
uint32_t eM_temp = eM_temps[j]; uint32_t eM_temp = eM_temps[j];
// Round to the nearest integer, according to the rules of handling // Round to the nearest integer, according to the rules of handling
// integer formats in Direct3D. // integer formats in Direct3D.
// TODO(Triang3l): Round by adding +-0.5, not with round_ne. // TODO(Triang3l): Round by adding +-0.5, not with round_ne.
DxbcOpRoundNE(DxbcDest::R(eM_temp), DxbcSrc::R(eM_temp)); a_.OpRoundNE(dxbc::Dest::R(eM_temp), dxbc::Src::R(eM_temp));
DxbcOpFToI(DxbcDest::R(eM_temp), DxbcSrc::R(eM_temp)); a_.OpFToI(dxbc::Dest::R(eM_temp), dxbc::Src::R(eM_temp));
DxbcOpBFI(DxbcDest::R(eM_temp, 0b0011), DxbcSrc::LU(16), a_.OpBFI(dxbc::Dest::R(eM_temp, 0b0011), dxbc::Src::LU(16),
DxbcSrc::LU(16), DxbcSrc::R(eM_temp, 0b1101), dxbc::Src::LU(16), dxbc::Src::R(eM_temp, 0b1101),
DxbcSrc::R(eM_temp, 0b1000)); dxbc::Src::R(eM_temp, 0b1000));
} }
DxbcOpBreak(); a_.OpBreak();
// k_16_16_FLOAT // k_16_16_FLOAT
DxbcOpCase(DxbcSrc::LU(uint32_t(xenos::ColorFormat::k_16_16_FLOAT))); a_.OpCase(dxbc::Src::LU(uint32_t(xenos::ColorFormat::k_16_16_FLOAT)));
for (uint32_t j = 0; j < eM_count; ++j) { for (uint32_t j = 0; j < eM_count; ++j) {
uint32_t eM_temp = eM_temps[j]; uint32_t eM_temp = eM_temps[j];
DxbcOpF32ToF16(DxbcDest::R(eM_temp, 0b0011), DxbcSrc::R(eM_temp)); a_.OpF32ToF16(dxbc::Dest::R(eM_temp, 0b0011), dxbc::Src::R(eM_temp));
DxbcOpBFI(DxbcDest::R(eM_temp, 0b0001), DxbcSrc::LU(16), a_.OpBFI(dxbc::Dest::R(eM_temp, 0b0001), dxbc::Src::LU(16),
DxbcSrc::LU(16), DxbcSrc::R(eM_temp, DxbcSrc::kYYYY), dxbc::Src::LU(16), dxbc::Src::R(eM_temp, dxbc::Src::kYYYY),
DxbcSrc::R(eM_temp, DxbcSrc::kXXXX)); dxbc::Src::R(eM_temp, dxbc::Src::kXXXX));
} }
DxbcOpBreak(); a_.OpBreak();
// k_16_16_16_16_FLOAT // k_16_16_16_16_FLOAT
DxbcOpCase( a_.OpCase(
DxbcSrc::LU(uint32_t(xenos::ColorFormat::k_16_16_16_16_FLOAT))); dxbc::Src::LU(uint32_t(xenos::ColorFormat::k_16_16_16_16_FLOAT)));
DxbcOpMov(element_size_dest, DxbcSrc::LU(8)); a_.OpMov(element_size_dest, dxbc::Src::LU(8));
for (uint32_t j = 0; j < eM_count; ++j) { for (uint32_t j = 0; j < eM_count; ++j) {
uint32_t eM_temp = eM_temps[j]; uint32_t eM_temp = eM_temps[j];
DxbcOpF32ToF16(DxbcDest::R(eM_temp), DxbcSrc::R(eM_temp)); a_.OpF32ToF16(dxbc::Dest::R(eM_temp), dxbc::Src::R(eM_temp));
DxbcOpBFI(DxbcDest::R(eM_temp, 0b0011), DxbcSrc::LU(16), a_.OpBFI(dxbc::Dest::R(eM_temp, 0b0011), dxbc::Src::LU(16),
DxbcSrc::LU(16), DxbcSrc::R(eM_temp, 0b1101), dxbc::Src::LU(16), dxbc::Src::R(eM_temp, 0b1101),
DxbcSrc::R(eM_temp, 0b1000)); dxbc::Src::R(eM_temp, 0b1000));
} }
DxbcOpBreak(); a_.OpBreak();
// k_32_FLOAT // k_32_FLOAT
// Already in the destination format, 4 bytes per element already // Already in the destination format, 4 bytes per element already
// selected. // selected.
// k_32_32_FLOAT // k_32_32_FLOAT
DxbcOpCase(DxbcSrc::LU(uint32_t(xenos::ColorFormat::k_32_32_FLOAT))); a_.OpCase(dxbc::Src::LU(uint32_t(xenos::ColorFormat::k_32_32_FLOAT)));
DxbcOpMov(element_size_dest, DxbcSrc::LU(8)); a_.OpMov(element_size_dest, dxbc::Src::LU(8));
// Already in the destination format. // Already in the destination format.
DxbcOpBreak(); a_.OpBreak();
// k_32_32_32_32_FLOAT // k_32_32_32_32_FLOAT
DxbcOpCase( a_.OpCase(
DxbcSrc::LU(uint32_t(xenos::ColorFormat::k_32_32_32_32_FLOAT))); dxbc::Src::LU(uint32_t(xenos::ColorFormat::k_32_32_32_32_FLOAT)));
DxbcOpMov(element_size_dest, DxbcSrc::LU(16)); a_.OpMov(element_size_dest, dxbc::Src::LU(16));
// Already in the destination format. // Already in the destination format.
DxbcOpBreak(); a_.OpBreak();
} }
DxbcOpEndSwitch(); a_.OpEndSwitch();
// control_temp.yz are now free. // control_temp.yz are now free.
// Do endian swap. // Do endian swap.
{ {
DxbcDest endian_dest(DxbcDest::R(control_temp, 0b0010)); dxbc::Dest endian_dest(dxbc::Dest::R(control_temp, 0b0010));
DxbcSrc endian_src(DxbcSrc::R(control_temp, DxbcSrc::kYYYY)); dxbc::Src endian_src(dxbc::Src::R(control_temp, dxbc::Src::kYYYY));
// Extract endianness into control_temp.y. // Extract endianness into control_temp.y.
DxbcOpAnd(endian_dest, DxbcSrc::R(eA_temp, DxbcSrc::kZZZZ), a_.OpAnd(endian_dest, dxbc::Src::R(eA_temp, dxbc::Src::kZZZZ),
DxbcSrc::LU(0b111)); dxbc::Src::LU(0b111));
// Change 8-in-64 and 8-in-128 to 8-in-32. // Change 8-in-64 and 8-in-128 to 8-in-32.
for (uint32_t j = 0; j < 2; ++j) { for (uint32_t j = 0; j < 2; ++j) {
DxbcOpIEq(DxbcDest::R(control_temp, 0b0100), endian_src, a_.OpIEq(dxbc::Dest::R(control_temp, 0b0100), endian_src,
DxbcSrc::LU(uint32_t(j ? xenos::Endian128::k8in128 dxbc::Src::LU(uint32_t(j ? xenos::Endian128::k8in128
: xenos::Endian128::k8in64))); : xenos::Endian128::k8in64)));
for (uint32_t k = 0; k < eM_count; ++k) { for (uint32_t k = 0; k < eM_count; ++k) {
uint32_t eM_temp = eM_temps[k]; uint32_t eM_temp = eM_temps[k];
DxbcOpMovC(DxbcDest::R(eM_temp), a_.OpMovC(dxbc::Dest::R(eM_temp),
DxbcSrc::R(control_temp, DxbcSrc::kZZZZ), dxbc::Src::R(control_temp, dxbc::Src::kZZZZ),
DxbcSrc::R(eM_temp, j ? 0b00011011 : 0b10110001), dxbc::Src::R(eM_temp, j ? 0b00011011 : 0b10110001),
DxbcSrc::R(eM_temp)); dxbc::Src::R(eM_temp));
} }
DxbcOpMovC(endian_dest, DxbcSrc::R(control_temp, DxbcSrc::kZZZZ), a_.OpMovC(endian_dest, dxbc::Src::R(control_temp, dxbc::Src::kZZZZ),
DxbcSrc::LU(uint32_t(xenos::Endian128::k8in32)), endian_src); dxbc::Src::LU(uint32_t(xenos::Endian128::k8in32)),
endian_src);
} }
uint32_t swap_temp = PushSystemTemp(); uint32_t swap_temp = PushSystemTemp();
DxbcDest swap_temp_dest(DxbcDest::R(swap_temp)); dxbc::Dest swap_temp_dest(dxbc::Dest::R(swap_temp));
DxbcSrc swap_temp_src(DxbcSrc::R(swap_temp)); dxbc::Src swap_temp_src(dxbc::Src::R(swap_temp));
// 8-in-16 or one half of 8-in-32. // 8-in-16 or one half of 8-in-32.
DxbcOpSwitch(endian_src); a_.OpSwitch(endian_src);
DxbcOpCase(DxbcSrc::LU(uint32_t(xenos::Endian128::k8in16))); a_.OpCase(dxbc::Src::LU(uint32_t(xenos::Endian128::k8in16)));
DxbcOpCase(DxbcSrc::LU(uint32_t(xenos::Endian128::k8in32))); a_.OpCase(dxbc::Src::LU(uint32_t(xenos::Endian128::k8in32)));
for (uint32_t j = 0; j < eM_count; ++j) { for (uint32_t j = 0; j < eM_count; ++j) {
DxbcDest eM_dest(DxbcDest::R(eM_temps[j])); dxbc::Dest eM_dest(dxbc::Dest::R(eM_temps[j]));
DxbcSrc eM_src(DxbcSrc::R(eM_temps[j])); dxbc::Src eM_src(dxbc::Src::R(eM_temps[j]));
// Temp = X0Z0. // Temp = X0Z0.
DxbcOpAnd(swap_temp_dest, eM_src, DxbcSrc::LU(0x00FF00FF)); a_.OpAnd(swap_temp_dest, eM_src, dxbc::Src::LU(0x00FF00FF));
// eM = YZW0. // eM = YZW0.
DxbcOpUShR(eM_dest, eM_src, DxbcSrc::LU(8)); a_.OpUShR(eM_dest, eM_src, dxbc::Src::LU(8));
// eM = Y0W0. // eM = Y0W0.
DxbcOpAnd(eM_dest, eM_src, DxbcSrc::LU(0x00FF00FF)); a_.OpAnd(eM_dest, eM_src, dxbc::Src::LU(0x00FF00FF));
// eM = YXWZ. // eM = YXWZ.
DxbcOpUMAd(eM_dest, swap_temp_src, DxbcSrc::LU(256), eM_src); a_.OpUMAd(eM_dest, swap_temp_src, dxbc::Src::LU(256), eM_src);
} }
DxbcOpBreak(); a_.OpBreak();
DxbcOpEndSwitch(); a_.OpEndSwitch();
// 16-in-32 or another half of 8-in-32. // 16-in-32 or another half of 8-in-32.
DxbcOpSwitch(endian_src); a_.OpSwitch(endian_src);
DxbcOpCase(DxbcSrc::LU(uint32_t(xenos::Endian128::k8in32))); a_.OpCase(dxbc::Src::LU(uint32_t(xenos::Endian128::k8in32)));
DxbcOpCase(DxbcSrc::LU(uint32_t(xenos::Endian128::k16in32))); a_.OpCase(dxbc::Src::LU(uint32_t(xenos::Endian128::k16in32)));
for (uint32_t j = 0; j < eM_count; ++j) { for (uint32_t j = 0; j < eM_count; ++j) {
DxbcDest eM_dest(DxbcDest::R(eM_temps[j])); dxbc::Dest eM_dest(dxbc::Dest::R(eM_temps[j]));
DxbcSrc eM_src(DxbcSrc::R(eM_temps[j])); dxbc::Src eM_src(dxbc::Src::R(eM_temps[j]));
// Temp = ZW00. // Temp = ZW00.
DxbcOpUShR(swap_temp_dest, eM_src, DxbcSrc::LU(16)); a_.OpUShR(swap_temp_dest, eM_src, dxbc::Src::LU(16));
// eM = ZWXY. // eM = ZWXY.
DxbcOpBFI(eM_dest, DxbcSrc::LU(16), DxbcSrc::LU(16), eM_src, a_.OpBFI(eM_dest, dxbc::Src::LU(16), dxbc::Src::LU(16), eM_src,
swap_temp_src); swap_temp_src);
} }
DxbcOpBreak(); a_.OpBreak();
DxbcOpEndSwitch(); a_.OpEndSwitch();
// Release swap_temp. // Release swap_temp.
PopSystemTemp(); PopSystemTemp();
} }
// control_temp.yz are now free. // control_temp.yz are now free.
DxbcDest address_dest(DxbcDest::R(eA_temp, 0b0001)); dxbc::Dest address_dest(dxbc::Dest::R(eA_temp, 0b0001));
DxbcSrc address_src(DxbcSrc::R(eA_temp, DxbcSrc::kXXXX)); dxbc::Src address_src(dxbc::Src::R(eA_temp, dxbc::Src::kXXXX));
// Multiply the base address by dword size, also dropping the 0x40000000 // Multiply the base address by dword size, also dropping the 0x40000000
// bit. // bit.
DxbcOpIShL(address_dest, address_src, DxbcSrc::LU(2)); a_.OpIShL(address_dest, address_src, dxbc::Src::LU(2));
// Drop the exponent in the element index. // Drop the exponent in the element index.
DxbcOpAnd(DxbcDest::R(eA_temp, 0b0010), DxbcSrc::R(eA_temp, DxbcSrc::kYYYY), a_.OpAnd(dxbc::Dest::R(eA_temp, 0b0010),
DxbcSrc::LU((1 << 23) - 1)); dxbc::Src::R(eA_temp, dxbc::Src::kYYYY),
dxbc::Src::LU((1 << 23) - 1));
// Add the offset of the first written element to the base address. // Add the offset of the first written element to the base address.
DxbcOpUMAd(address_dest, DxbcSrc::R(eA_temp, DxbcSrc::kYYYY), a_.OpUMAd(address_dest, dxbc::Src::R(eA_temp, dxbc::Src::kYYYY),
element_size_src, address_src); element_size_src, address_src);
// Do the writes. // Do the writes.
DxbcSrc eM_written_src( dxbc::Src eM_written_src(
DxbcSrc::R(system_temp_memexport_written_).Select(i >> 2)); dxbc::Src::R(system_temp_memexport_written_).Select(i >> 2));
uint32_t eM_written_base = 1u << ((i & 3) << 3); uint32_t eM_written_base = 1u << ((i & 3) << 3);
for (uint32_t j = 0; j < eM_count; ++j) { for (uint32_t j = 0; j < eM_count; ++j) {
// Go to the next eM#. // Go to the next eM#.
uint32_t eM_relative_offset = eM_offsets[j] - (j ? eM_offsets[j - 1] : 0); uint32_t eM_relative_offset = eM_offsets[j] - (j ? eM_offsets[j - 1] : 0);
if (eM_relative_offset) { if (eM_relative_offset) {
if (eM_relative_offset == 1) { if (eM_relative_offset == 1) {
DxbcOpIAdd(address_dest, element_size_src, address_src); a_.OpIAdd(address_dest, element_size_src, address_src);
} else { } else {
DxbcOpUMAd(address_dest, DxbcSrc::LU(eM_relative_offset), a_.OpUMAd(address_dest, dxbc::Src::LU(eM_relative_offset),
element_size_src, address_src); element_size_src, address_src);
} }
} }
// Check if the eM# was actually written to on the execution path. // Check if the eM# was actually written to on the execution path.
DxbcOpAnd(DxbcDest::R(control_temp, 0b0010), eM_written_src, a_.OpAnd(dxbc::Dest::R(control_temp, 0b0010), eM_written_src,
DxbcSrc::LU(eM_written_base << eM_offsets[j])); dxbc::Src::LU(eM_written_base << eM_offsets[j]));
DxbcOpIf(true, DxbcSrc::R(control_temp, DxbcSrc::kYYYY)); a_.OpIf(true, dxbc::Src::R(control_temp, dxbc::Src::kYYYY));
// Write the element of the needed size. // Write the element of the needed size.
DxbcSrc eM_src(DxbcSrc::R(eM_temps[j])); dxbc::Src eM_src(dxbc::Src::R(eM_temps[j]));
DxbcOpSwitch(element_size_src); a_.OpSwitch(element_size_src);
for (uint32_t k = 1; k <= 4; k <<= 1) { for (uint32_t k = 1; k <= 4; k <<= 1) {
DxbcOpCase(DxbcSrc::LU(k * 4)); a_.OpCase(dxbc::Src::LU(k * 4));
if (uav_index_shared_memory_ == kBindingIndexUnallocated) { if (uav_index_shared_memory_ == kBindingIndexUnallocated) {
uav_index_shared_memory_ = uav_count_++; uav_index_shared_memory_ = uav_count_++;
} }
DxbcOpStoreRaw( a_.OpStoreRaw(
DxbcDest::U(uav_index_shared_memory_, dxbc::Dest::U(uav_index_shared_memory_,
uint32_t(UAVRegister::kSharedMemory), (1 << k) - 1), uint32_t(UAVRegister::kSharedMemory), (1 << k) - 1),
address_src, eM_src); address_src, eM_src);
DxbcOpBreak(); a_.OpBreak();
} }
DxbcOpEndSwitch(); a_.OpEndSwitch();
DxbcOpEndIf(); a_.OpEndIf();
} }
// control_temp.y is now free. // control_temp.y is now free.
} }
// Close the memexport possibility check. // Close the memexport possibility check.
DxbcOpEndIf(); a_.OpEndIf();
// Release control_temp. // Release control_temp.
PopSystemTemp(); PopSystemTemp();

2512
src/xenia/gpu/dxbc_shader_translator_om.cc
View File

File diff suppressed because it is too large Load Diff