raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-21 11:26:56 +02:00
Code Issues Packages Projects Releases Wiki Activity

Implement vertex swizzle

Browse Source
This commit is contained in:
nmzik
2026-07-18 03:36:09 +02:00
parent 18479daabd
commit f4b52cea2b
10 changed files with 282 additions and 195 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/host_gpu/renderer/shaders.cpp
+20 -73
View File
@@ -22,7 +22,7 @@
namespace Libs::Graphics { namespace Libs::Graphics {
//IDK: maybe we can remove it? // IDK: maybe we can remove it?
constexpr uint32_t kTemporaryPs5BufferFormat121 = 121u; constexpr uint32_t kTemporaryPs5BufferFormat121 = 121u;
static bool NarrowInputFormat(VkFormat* format, uint32_t* size, uint32_t used_components) { static bool NarrowInputFormat(VkFormat* format, uint32_t* size, uint32_t used_components) {
@@ -561,13 +561,15 @@ void CreatePipelineInternal(PipelineCache::GraphicsPipeline* pipeline, VkRenderP
input_attr[index].location = index; input_attr[index].location = index;
input_attr[index].offset = b.attr_offsets[ai]; input_attr[index].offset = b.attr_offsets[ai];
uint32_t attr_size = 4; uint32_t attr_size = 4;
auto registers_num = vs_input_info->resources_dst[index].registers_num; auto registers_num = vs_input_info->resources_dst[index].registers_num;
auto used_components = (vs_input_info->resource_fetch_components[index] > 0 auto used_components = (vs_input_info->resource_fetch_components[index] > 0
? vs_input_info->resource_fetch_components[index] ? vs_input_info->resource_fetch_components[index]
: registers_num); : registers_num);
const auto raw_components = VertexRawComponentCount(
vs_input_info->resources[index].DstSelXYZW(), static_cast<uint32_t>(registers_num));
GetInputFormat(vs_input_info->resources[index], &input_attr[index].format, &attr_size, GetInputFormat(vs_input_info->resources[index], &input_attr[index].format, &attr_size,
static_cast<uint32_t>(used_components)); raw_components);
if (graphics_debug_dump_enabled()) { if (graphics_debug_dump_enabled()) {
static std::atomic_uint log_count = 0; static std::atomic_uint log_count = 0;
@@ -602,73 +604,18 @@ void CreatePipelineInternal(PipelineCache::GraphicsPipeline* pipeline, VkRenderP
EXIT_NOT_IMPLEMENTED(vs_input_info->resources[index].AddTid()); EXIT_NOT_IMPLEMENTED(vs_input_info->resources[index].AddTid());
EXIT_NOT_IMPLEMENTED(vs_input_info->resources[index].SwizzleEnabled()); EXIT_NOT_IMPLEMENTED(vs_input_info->resources[index].SwizzleEnabled());
auto log_unsupported_vertex_swizzle = [index, attr_size, registers_num]( if (registers_num < 1 || registers_num > 4) {
uint32_t swizzle, uint32_t expected) { EXIT("invalid vertex destination register count: %u\n", registers_num);
static bool logged = false; }
if (!logged) { for (uint32_t component = 0; component < static_cast<uint32_t>(registers_num);
LOGF( component++) {
"VertexInput: temporary: accepting unsupported dst swizzle at attr %" PRIu32 const auto selector =
" (attr_size=%" PRIu32 ", regs=%" PRIu32 ", swizzle=0x%03" PRIx32 GetDstSel(vs_input_info->resources[index].DstSelXYZW(), component);
", expected=0x%03" PRIx32 ")\n", if (DecodeVertexDstSelector(selector).kind == VertexDstSelectKind::Reserved) {
static_cast<uint32_t>(index), attr_size, registers_num, swizzle, expected); EXIT("unsupported reserved vertex dst selector: attr=%u component=%u "
logged = true; "selector=%u\n",
static_cast<uint32_t>(index), component, selector);
} }
};
switch (registers_num) {
case 1: {
auto swizzle = vs_input_info->resources[index].DstSelX();
if (swizzle != DstSel(4)) {
log_unsupported_vertex_swizzle(swizzle, DstSel(4));
}
break;
}
case 2: {
auto swizzle = vs_input_info->resources[index].DstSelXY();
auto expected = (attr_size == 1 ? DstSel(4, 0) : DstSel(4, 5));
if (swizzle != expected) {
log_unsupported_vertex_swizzle(swizzle, expected);
}
break;
}
case 3: {
auto swizzle = vs_input_info->resources[index].DstSelXYZ();
auto expected =
(attr_size == 1 ? DstSel(4, 0, 0)
: (attr_size == 2 ? DstSel(4, 5, 0) : DstSel(4, 5, 6)));
if (swizzle != expected) {
log_unsupported_vertex_swizzle(swizzle, expected);
}
break;
}
case 4: {
auto swizzle = vs_input_info->resources[index].DstSelXYZW();
auto expected = DstSel(4, 5, 6, 7);
bool supported = false;
switch (attr_size) {
case 1:
expected = DstSel(4, 0, 0, 1);
supported = (swizzle == expected);
break;
case 2:
expected = DstSel(4, 5, 0, 1);
supported = (swizzle == expected);
break;
case 3:
expected = DstSel(4, 5, 6, 1);
supported = (swizzle == expected || swizzle == DstSel(4, 5, 6, 0));
break;
default:
supported = (swizzle == expected || swizzle == DstSel(4, 5, 6, 1) ||
swizzle == DstSel(4, 5, 6, 0));
break;
}
if (!supported) {
log_unsupported_vertex_swizzle(swizzle, expected);
}
break;
}
default: EXIT("invalid registers_num");
} }
} }
} }
src/graphics/shader/recompiler/ShaderDecoder.h
+2
View File
@@ -639,6 +639,8 @@ struct Instruction {
bool slc = false; bool slc = false;
bool idxen = false; bool idxen = false;
bool offen = false; bool offen = false;
// V_CMPX writes the comparison result to its normal scalar destination and EXEC.
bool compare_exec = false;
int32_t branch_offset = 0; int32_t branch_offset = 0;
uint32_t branch_target = 0; uint32_t branch_target = 0;
struct { struct {
src/graphics/shader/recompiler/VectorAluOps.cpp
+8 -8
View File
@@ -1103,9 +1103,8 @@ bool DecodeVopcSdwa(uint32_t pc, std::span<const uint32_t> code, uint32_t word_i
!DecodeScalarSource(vsrc1 + (fields.s1 == 0u ? 256u : 0u), pc, &inst->src1, error)) { !DecodeScalarSource(vsrc1 + (fields.s1 == 0u ? 256u : 0u), pc, &inst->src1, error)) {
return false; return false;
} }
if (IsVopcCompareExec(inst->opcode)) { inst->compare_exec = IsVopcCompareExec(inst->opcode);
inst->dst.kind = OperandKind::ExecLo; if (fields.sd == 0u) {
} else if (fields.sd == 0u) {
inst->dst.kind = OperandKind::VccLo; inst->dst.kind = OperandKind::VccLo;
} else if (!DecodeScalarDestination(fields.sdst, pc, &inst->dst, error)) { } else if (!DecodeScalarDestination(fields.sdst, pc, &inst->dst, error)) {
return false; return false;
@@ -1142,7 +1141,8 @@ bool DecodeVopcDpp(uint32_t pc, std::span<const uint32_t> code, uint32_t word_in
!DecodeVectorGpr(vsrc1, &inst->src1, error)) { !DecodeVectorGpr(vsrc1, &inst->src1, error)) {
return false; return false;
} }
inst->dst.kind = IsVopcCompareExec(inst->opcode) ? OperandKind::ExecLo : OperandKind::VccLo; inst->dst.kind = OperandKind::VccLo;
inst->compare_exec = IsVopcCompareExec(inst->opcode);
inst->src0.negate = ((modifier >> 20u) & 0x1u) != 0u; inst->src0.negate = ((modifier >> 20u) & 0x1u) != 0u;
inst->src0.absolute = ((modifier >> 21u) & 0x1u) != 0u; inst->src0.absolute = ((modifier >> 21u) & 0x1u) != 0u;
inst->src0.dpp = true; inst->src0.dpp = true;
@@ -1532,7 +1532,8 @@ bool DecodeVopc(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
inst->family = Family::VOPC; inst->family = Family::VOPC;
inst->opcode_id = opcode; inst->opcode_id = opcode;
inst->opcode = Lookup(VOPC_OPS, static_cast<uint32_t>(std::size(VOPC_OPS)), opcode); inst->opcode = Lookup(VOPC_OPS, static_cast<uint32_t>(std::size(VOPC_OPS)), opcode);
inst->dst.kind = IsVopcCompareExec(inst->opcode) ? OperandKind::ExecLo : OperandKind::VccLo; inst->dst.kind = OperandKind::VccLo;
inst->compare_exec = IsVopcCompareExec(inst->opcode);
SetRawWords(inst, code, word_index, 1); SetRawWords(inst, code, word_index, 1);
switch (src0) { switch (src0) {
@@ -1618,9 +1619,8 @@ bool DecodeVop3(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
return true; return true;
} }
bool dst_ok = true; bool dst_ok = true;
if (compare_exec) { inst->compare_exec = compare_exec;
inst->dst.kind = OperandKind::ExecLo; if (scalar_dst) {
} else if (scalar_dst) {
// VOP3A uses VDST for VOPC and the scalar-destination lane-read opcodes. // VOP3A uses VDST for VOPC and the scalar-destination lane-read opcodes.
dst_ok = DecodeScalarDestination(vdst, pc, &inst->dst, error); dst_ok = DecodeScalarDestination(vdst, pc, &inst->dst, error);
} else { } else {
src/graphics/shader/recompiler/shaderIR/ShaderIR.cpp
+16
View File
@@ -1165,6 +1165,22 @@ bool LowerImplemented(const Decoder::Instruction& decoded, BasicBlock* block, st
} }
block->instructions.push_back(inst); block->instructions.push_back(inst);
if (decoded.compare_exec) {
// V_CMPX has two architectural destinations. Keep the comparison's normal VCC/SDST
// result, then mirror the complete lane mask to EXEC. A mask is always a 64-bit scalar pair, including wave32 shaders.
for (uint32_t i = 0; i < 2u; i++) {
Instruction move;
move.pc = decoded.pc;
move.op = Opcode::MoveU32;
move.src_count = 1;
move.src[0] = inst.dst;
move.src[0].reg.index += i;
move.dst.kind = OperandKind::Register;
move.dst.reg.file = RegisterFile::Exec;
move.dst.reg.index = i;
block->instructions.push_back(move);
}
}
return AppendScalarResultSccNonZero(decoded, block, error); return AppendScalarResultSccNonZero(decoded, block, error);
} }
src/graphics/shader/recompiler/spirvEmitter/spirvEmitterModule.cpp
+5 -1
View File
@@ -94,8 +94,12 @@ uint32_t VertexParameterComponentCount(const EmitterState& state, const InputBin
input.location < ShaderVertexInputInfo::RES_MAX && input.location < ShaderVertexInputInfo::RES_MAX &&
input.location < static_cast<uint32_t>(state.vertex_input_info->resources_num) && input.location < static_cast<uint32_t>(state.vertex_input_info->resources_num) &&
state.vertex_input_info->resources_dst[input.location].registers_num > 0) { state.vertex_input_info->resources_dst[input.location].registers_num > 0) {
count = static_cast<uint32_t>( const auto& descriptor = state.vertex_input_info->resources[input.location];
const auto dst_count = static_cast<uint32_t>(
state.vertex_input_info->resources_dst[input.location].registers_num); state.vertex_input_info->resources_dst[input.location].registers_num);
// Keep the SPIR-V interface and native VkFormat width identical. A destination
// selector may request a raw component beyond the written VGPR count (W -> X).
count = VertexRawComponentCount(descriptor.DstSelXYZW(), dst_count);
} }
return std::clamp(count, 1u, 4u); return std::clamp(count, 1u, 4u);
} }
src/graphics/shader/recompiler/spirvEmitter/spirvEmitterValues.cpp
+20 -2
View File
@@ -450,7 +450,25 @@ uint32_t EmitVertexParameterComponentU32(EmitterState* state, const InputBinding
uint32_t component) { uint32_t component) {
const auto count = VertexParameterComponentCount(*state, input); const auto count = VertexParameterComponentCount(*state, input);
const auto kind = VertexParameterScalarKind(*state, input.location); const auto kind = VertexParameterScalarKind(*state, input.location);
if (component >= count) { uint32_t source_component = component;
if (input.location < ShaderVertexInputInfo::RES_MAX &&
input.location < static_cast<uint32_t>(state->vertex_input_info->resources_num)) {
const auto& descriptor = state->vertex_input_info->resources[input.location];
const auto selection =
DecodeVertexDstSelector(GetDstSel(descriptor.DstSelXYZW(), component & 3u));
if (selection.kind == VertexDstSelectKind::Zero) {
return ConstantU32(state, 0u);
}
if (selection.kind == VertexDstSelectKind::One) {
return ConstantU32(state,
kind == VertexInputScalarKind::Float ? 0x3f800000u : 1u);
}
if (selection.kind == VertexDstSelectKind::Reserved) {
EXIT("reserved vertex destination selector\n");
}
source_component = selection.source_component;
}
if (source_component >= count) {
return VertexInputDefaultComponentU32(state, kind, component); return VertexInputDefaultComponentU32(state, kind, component);
} }
@@ -462,7 +480,7 @@ uint32_t EmitVertexParameterComponentU32(EmitterState* state, const InputBinding
const auto pointer_type = VertexParameterScalarPointerType(*state, kind); const auto pointer_type = VertexParameterScalarPointerType(*state, kind);
const auto pointer = state->builder.AllocateId(); const auto pointer = state->builder.AllocateId();
state->builder.AddFunction({OpAccessChain, pointer_type, pointer, input.variable_id, state->builder.AddFunction({OpAccessChain, pointer_type, pointer, input.variable_id,
ConstantU32(state, component)}); ConstantU32(state, source_component)});
state->builder.AddFunction({OpLoad, scalar_type, raw, pointer}); state->builder.AddFunction({OpLoad, scalar_type, raw, pointer});
} }
src/graphics/shader/shader.h
+55 -20
View File
@@ -64,6 +64,41 @@ inline uint8_t GetDstSel(uint32_t swizzle, uint32_t channel) {
return (swizzle >> (channel * 3u)) & 0x7u; return (swizzle >> (channel * 3u)) & 0x7u;
} }
enum class VertexDstSelectKind : uint8_t { Zero, One, Component, Reserved };
struct VertexDstSelection {
VertexDstSelectKind kind = VertexDstSelectKind::Reserved;
uint8_t source_component = 0;
};
constexpr VertexDstSelection DecodeVertexDstSelector(uint8_t selector) {
if (selector == 0u) {
return {VertexDstSelectKind::Zero, 0};
}
if (selector == 1u) {
return {VertexDstSelectKind::One, 0};
}
if (selector >= 4u && selector <= 7u) {
return {VertexDstSelectKind::Component, static_cast<uint8_t>(selector - 4u)};
}
return {VertexDstSelectKind::Reserved, 0};
}
constexpr uint32_t VertexRawComponentCount(uint32_t swizzle, uint32_t destination_components) {
const auto dst_count = destination_components < 1u
? 1u
: (destination_components > 4u ? 4u : destination_components);
auto count = dst_count;
for (uint32_t component = 0; component < dst_count; component++) {
const auto selection = DecodeVertexDstSelector(GetDstSel(swizzle, component));
if (selection.kind == VertexDstSelectKind::Component) {
const auto source_count = static_cast<uint32_t>(selection.source_component) + 1u;
count = count > source_count ? count : source_count;
}
}
return count;
}
struct ShaderVertexInputInfo { struct ShaderVertexInputInfo {
static constexpr int RES_MAX = 32; static constexpr int RES_MAX = 32;
@@ -96,26 +131,26 @@ struct ShaderComputeInputInfo {
}; };
struct ShaderPixelInputInfo { struct ShaderPixelInputInfo {
uint32_t interpolator_settings[32] = {0}; uint32_t interpolator_settings[32] = {0};
uint32_t input_num = 0; uint32_t input_num = 0;
uint32_t ps_system_input_base = 0; uint32_t ps_system_input_base = 0;
uint8_t target_output_mode[8] = {}; uint8_t target_output_mode[8] = {};
std::array<Prospero::ColorComponentMapping, 8> target_export_mapping = {}; std::array<Prospero::ColorComponentMapping, 8> target_export_mapping = {};
uint32_t mrt_output_mask = 0; uint32_t mrt_output_mask = 0;
uint32_t descriptor_set = 0; uint32_t descriptor_set = 0;
bool ps_pos_x = false; bool ps_pos_x = false;
bool ps_pos_y = false; bool ps_pos_y = false;
bool ps_pos_xy = false; bool ps_pos_xy = false;
bool ps_pos_z = false; bool ps_pos_z = false;
bool ps_pos_w = false; bool ps_pos_w = false;
bool ps_front_face = false; bool ps_front_face = false;
bool ps_no_perspective = false; bool ps_no_perspective = false;
bool ps_pixel_kill_enable = false; bool ps_pixel_kill_enable = false;
bool ps_depth_export_enable = false; bool ps_depth_export_enable = false;
bool ps_sample_mask_export_enable = false; bool ps_sample_mask_export_enable = false;
bool ps_early_z = false; bool ps_early_z = false;
bool ps_execute_on_noop = false; bool ps_execute_on_noop = false;
ShaderStageRuntime stage; ShaderStageRuntime stage;
bool HasPositionInput() const { return ps_pos_x || ps_pos_y || ps_pos_z || ps_pos_w; } bool HasPositionInput() const { return ps_pos_x || ps_pos_y || ps_pos_z || ps_pos_w; }
}; };
tests/ShaderRecompilerComputeTests.cpp
+38 -15
View File
@@ -5700,24 +5700,24 @@ TestCase VectorVop3FloatCompareNegSourceModifier() {
{O::VMovB32, O::VCmpLtF32, O::BufferStoreDword, O::SEndpgm}}; {O::VMovB32, O::VCmpLtF32, O::BufferStoreDword, O::SEndpgm}};
} }
TestCase VectorVop3CmpxWritesExecMask() { TestCase VectorVop3CmpxWritesScalarDstAndExecMask() {
using O = ShaderOpcode; using O = ShaderOpcode;
std::vector<u32> code; std::vector<u32> code;
AppendVMovU32(&code, 0, 2); AppendVMovU32(&code, 0, 1);
AppendVMovU32(&code, 1, 1); AppendVMovU32(&code, 1, 1);
AppendVMovU32(&code, 2, 0); AppendVMovU32(&code, 2, 7);
AppendVMovU32(&code, 30, 0); AppendVMovU32(&code, 30, 0);
AppendVop3(&code, 0xd1, 5, Vgpr(0), Vgpr(1)); // v_cmpx_lt_u32, false AppendVop3(&code, 0xd2, 5, Vgpr(0), Vgpr(1)); // v_cmpx_eq_u32 s[5:6], true
AppendVMovU32(&code, 2, 7); code.push_back(EncodeSop1(0x0a, 126, 5)); // s_wqm_b64 exec, s[5:6]
AppendBufferStoreDword(&code, 2, 30); AppendBufferStoreDword(&code, 2, 30);
AppendEnd(&code); AppendEnd(&code);
return {"VectorVop3CmpxWritesExecMask", return {"VectorVop3CmpxWritesScalarDstAndExecMask",
code, code,
{0}, {0},
{0}, {7},
{O::VMovB32, O::VCmpxLtU32, O::BufferStoreDword, O::SEndpgm}}; {O::VMovB32, O::VCmpxEqU32, O::SWqmB64, O::BufferStoreDword, O::SEndpgm}};
} }
TestCase VectorVopcSdwaCmpxWritesExecMask() { TestCase VectorVopcSdwaCmpxWritesExecMask() {
@@ -5741,6 +5741,28 @@ TestCase VectorVopcSdwaCmpxWritesExecMask() {
{O::VMovB32, O::VCmpxLtU32, O::BufferStoreDword, O::SEndpgm}}; {O::VMovB32, O::VCmpxLtU32, O::BufferStoreDword, O::SEndpgm}};
} }
TestCase VectorVopcCmpxPreservesVccForWqm() {
using O = ShaderOpcode;
std::vector<u32> code;
AppendVMovU32(&code, 0, 1);
AppendVMovU32(&code, 1, 1);
AppendVMovU32(&code, 2, 7);
AppendVMovU32(&code, 30, 0);
code.push_back(EncodeSop1(0x04, 106, InlineU32(0))); // s_mov_b64 vcc, 0
code.push_back(EncodeVopc(0xd2, Vgpr(0), 1)); // v_cmpx_eq_u32, true
code.push_back(EncodeSop1(0x0a, 126, 106)); // s_wqm_b64 exec, vcc
AppendBufferStoreDword(&code, 2, 30);
AppendEnd(&code);
return {"VectorVopcCmpxPreservesVccForWqm",
code,
{0},
{7},
{O::VMovB32, O::SMovB64, O::VCmpxEqU32, O::SWqmB64,
O::BufferStoreDword, O::SEndpgm}};
}
TestCase VectorCompareInvertedMaskSelect() { TestCase VectorCompareInvertedMaskSelect() {
using O = ShaderOpcode; using O = ShaderOpcode;
@@ -8638,8 +8660,9 @@ std::vector<TestCase> MakeCases() {
AddCase(Vop3CndmaskAllowsDataSourceModifier); AddCase(Vop3CndmaskAllowsDataSourceModifier);
AddCase(VectorCompareExecOps); AddCase(VectorCompareExecOps);
AddCase(VectorVop3FloatCompareNegSourceModifier); AddCase(VectorVop3FloatCompareNegSourceModifier);
AddCase(VectorVop3CmpxWritesExecMask); AddCase(VectorVop3CmpxWritesScalarDstAndExecMask);
AddCase(VectorVopcSdwaCmpxWritesExecMask); AddCase(VectorVopcSdwaCmpxWritesExecMask);
AddCase(VectorVopcCmpxPreservesVccForWqm);
AddCase(VectorCompareInvertedMaskSelect); AddCase(VectorCompareInvertedMaskSelect);
AddCase(BranchSelect); AddCase(BranchSelect);
AddCase(SimpleLoop); AddCase(SimpleLoop);
tests/ShaderVertexMetadataTests.cpp
+98 -61
View File
@@ -9,85 +9,122 @@ namespace {
using namespace Libs::Graphics; using namespace Libs::Graphics;
void Check(bool value, const char* text) { void Check(bool value, const char *text) {
if (!value) { if (!value) {
std::fprintf(stderr, "ShaderVertexMetadataTests: failed: %s\n", text); std::fprintf(stderr, "ShaderVertexMetadataTests: failed: %s\n", text);
std::abort(); std::abort();
} }
} }
struct Fixture { struct Fixture {
std::array<uint16_t, static_cast<size_t>(AgcDirectResourceType::Last) + 1> offsets {}; std::array<uint16_t, static_cast<size_t>(AgcDirectResourceType::Last) + 1>
ShaderUserData user_data {}; offsets{};
ShaderSemantic semantic {}; ShaderUserData user_data{};
ShaderMappedData mapped {}; ShaderSemantic semantic{};
ShaderMappedData mapped{};
Fixture() { Fixture() {
offsets.fill(AGC_ILLEGAL_DIRECT_OFFSET); offsets.fill(AGC_ILLEGAL_DIRECT_OFFSET);
offsets[static_cast<size_t>(AgcDirectResourceType::PtrVertexBufferTable)] = 2; offsets[static_cast<size_t>(AgcDirectResourceType::PtrVertexBufferTable)] =
offsets[static_cast<size_t>(AgcDirectResourceType::PtrVertexAttribDescTable)] = 4; 2;
user_data.direct_resource_offset = offsets.data(); offsets[static_cast<size_t>(
user_data.direct_resource_count = static_cast<uint16_t>(offsets.size()); AgcDirectResourceType::PtrVertexAttribDescTable)] = 4;
mapped.user_data = &user_data; user_data.direct_resource_offset = offsets.data();
mapped.input_semantics = &semantic; user_data.direct_resource_count = static_cast<uint16_t>(offsets.size());
mapped.num_input_semantics = 1; mapped.user_data = &user_data;
} mapped.input_semantics = &semantic;
mapped.num_input_semantics = 1;
}
}; };
void CheckRejected(const ShaderMappedData& data, const char* text) { void CheckRejected(const ShaderMappedData &data, const char *text) {
ShaderVertexMetadata output; ShaderVertexMetadata output;
output.vertex_buffer_reg = 37; output.vertex_buffer_reg = 37;
output.vertex_attrib_reg = 41; output.vertex_attrib_reg = 41;
output.input_semantics_count = 7; output.input_semantics_count = 7;
std::string error; std::string error;
Check(!ShaderReadVertexMetadata(data, 64, &output, &error), text); Check(!ShaderReadVertexMetadata(data, 64, &output, &error), text);
Check(!error.empty(), "metadata rejection omitted its diagnostic"); Check(!error.empty(), "metadata rejection omitted its diagnostic");
Check(output.vertex_buffer_reg == 37 && output.vertex_attrib_reg == 41 && Check(output.vertex_buffer_reg == 37 && output.vertex_attrib_reg == 41 &&
output.input_semantics_count == 7, output.input_semantics_count == 7,
"metadata rejection changed the prior output"); "metadata rejection changed the prior output");
} }
void TestValidAndInvalidMetadata() { void TestValidAndInvalidMetadata() {
Fixture fixture; Fixture fixture;
ShaderVertexMetadata output; ShaderVertexMetadata output;
std::string error; std::string error;
Check(ShaderReadVertexMetadata(fixture.mapped, 64, &output, &error), Check(ShaderReadVertexMetadata(fixture.mapped, 64, &output, &error),
"valid AGC vertex metadata was rejected"); "valid AGC vertex metadata was rejected");
Check(output.vertex_buffer_reg == 2 && output.vertex_attrib_reg == 4 && Check(output.vertex_buffer_reg == 2 && output.vertex_attrib_reg == 4 &&
output.input_semantics_count == 1, output.input_semantics_count == 1,
"valid AGC vertex metadata was decoded incorrectly"); "valid AGC vertex metadata was decoded incorrectly");
auto missing_header = fixture.mapped; auto missing_header = fixture.mapped;
missing_header.user_data = nullptr; missing_header.user_data = nullptr;
CheckRejected(missing_header, "missing AGC user-data header was accepted"); CheckRejected(missing_header, "missing AGC user-data header was accepted");
Fixture missing_offsets; Fixture missing_offsets;
missing_offsets.user_data.direct_resource_offset = nullptr; missing_offsets.user_data.direct_resource_offset = nullptr;
CheckRejected(missing_offsets.mapped, "missing direct-resource offsets were accepted"); CheckRejected(missing_offsets.mapped,
"missing direct-resource offsets were accepted");
Fixture excessive_resources; Fixture excessive_resources;
excessive_resources.user_data.direct_resource_count = excessive_resources.user_data.direct_resource_count =
static_cast<uint16_t>(excessive_resources.offsets.size() + 1); static_cast<uint16_t>(excessive_resources.offsets.size() + 1);
CheckRejected(excessive_resources.mapped, "excessive direct-resource count was accepted"); CheckRejected(excessive_resources.mapped,
"excessive direct-resource count was accepted");
Fixture excessive_semantics; Fixture excessive_semantics;
excessive_semantics.mapped.num_input_semantics = ShaderVertexInputInfo::RES_MAX + 1; excessive_semantics.mapped.num_input_semantics =
CheckRejected(excessive_semantics.mapped, "excessive vertex semantic count was accepted"); ShaderVertexInputInfo::RES_MAX + 1;
CheckRejected(excessive_semantics.mapped,
"excessive vertex semantic count was accepted");
Fixture excessive_register; Fixture excessive_register;
excessive_register.offsets[ excessive_register.offsets[static_cast<size_t>(
static_cast<size_t>(AgcDirectResourceType::PtrVertexBufferTable)] = 63; AgcDirectResourceType::PtrVertexBufferTable)] = 63;
CheckRejected(excessive_register.mapped, "out-of-domain vertex table SGPR was accepted"); CheckRejected(excessive_register.mapped,
"out-of-domain vertex table SGPR was accepted");
Fixture missing_semantics; Fixture missing_semantics;
missing_semantics.mapped.input_semantics = nullptr; missing_semantics.mapped.input_semantics = nullptr;
CheckRejected(missing_semantics.mapped, "missing vertex semantic array was accepted"); CheckRejected(missing_semantics.mapped,
"missing vertex semantic array was accepted");
}
void TestVertexDstSelectors() {
const auto zero = DecodeVertexDstSelector(0);
const auto one = DecodeVertexDstSelector(1);
Check(zero.kind == VertexDstSelectKind::Zero &&
one.kind == VertexDstSelectKind::One,
"vertex constant selectors were decoded incorrectly");
for (uint8_t selector = 4; selector <= 7; selector++) {
const auto component = DecodeVertexDstSelector(selector);
Check(component.kind == VertexDstSelectKind::Component &&
component.source_component == selector - 4,
"vertex component selector was decoded incorrectly");
}
Check(DecodeVertexDstSelector(2).kind == VertexDstSelectKind::Reserved &&
DecodeVertexDstSelector(3).kind == VertexDstSelectKind::Reserved,
"reserved vertex selectors were accepted");
Check(GetDstSel(DstSel(6, 5, 4, 7), 0) == 6 &&
GetDstSel(DstSel(6, 5, 4, 7), 1) == 5 &&
GetDstSel(DstSel(6, 5, 4, 7), 2) == 4 &&
GetDstSel(DstSel(6, 5, 4, 7), 3) == 7,
"BGRA vertex selector packing was decoded incorrectly");
Check(VertexRawComponentCount(DstSel(7, 0, 0, 0), 1) == 4,
"W-to-X did not widen the raw vertex interface to four components");
Check(VertexRawComponentCount(DstSel(4, 4, 4, 4), 4) == 4 &&
VertexRawComponentCount(DstSel(4, 0, 0, 1), 4) == 4,
"replication or constant selectors changed raw vertex width");
} }
} // namespace } // namespace
int main() { int main() {
TestValidAndInvalidMetadata(); TestValidAndInvalidMetadata();
std::puts("ShaderVertexMetadataTests: all cases passed"); TestVertexDstSelectors();
return 0; std::puts("ShaderVertexMetadataTests: all cases passed");
return 0;
} }
tests/shaderCfgTests.cpp
+20 -15
View File
@@ -1473,9 +1473,11 @@ void TestNewShaderRecompilerMoreAluFamilies() {
"new decoder did not decode old-backed V_CMP_NEQ_F32"); "new decoder did not decode old-backed V_CMP_NEQ_F32");
Check(Common::ContainsStr(result.decoded_dump, "v_cmp_neq_f32 s0, 0.500000, v1"), Check(Common::ContainsStr(result.decoded_dump, "v_cmp_neq_f32 s0, 0.500000, v1"),
"new decoder did not decode old-backed V_CMP_NEQ_F32 SDWA scalar destination"); "new decoder did not decode old-backed V_CMP_NEQ_F32 SDWA scalar destination");
Check(Common::ContainsStr(result.decoded_dump, "v_cmpx_lt_u32 exec_lo, v5.sdwa(sel=4") && Check(Common::ContainsStr(result.decoded_dump, "v_cmpx_lt_u32 vcc_lo, v5.sdwa(sel=4") &&
Common::ContainsStr(result.ir_dump, "CompareMaskLtU32 exec_lo"), Common::ContainsStr(result.ir_dump, "CompareMaskLtU32 vcc_lo") &&
"new decoder did not route V_CMPX SDWA destination to exec"); Common::ContainsStr(result.ir_dump, "MoveU32 exec_lo, vcc_lo") &&
Common::ContainsStr(result.ir_dump, "MoveU32 exec_hi, vcc_hi"),
"new decoder did not preserve V_CMPX SDWA destination and mirror its mask to exec");
Check(Common::ContainsStr(result.decoded_dump, "v_cmp_nlt_f32"), Check(Common::ContainsStr(result.decoded_dump, "v_cmp_nlt_f32"),
"new decoder did not decode old-backed V_CMP_NLT_F32"); "new decoder did not decode old-backed V_CMP_NLT_F32");
Check(Common::ContainsStr(result.decoded_dump, "v_cmpx_nge_f32"), Check(Common::ContainsStr(result.decoded_dump, "v_cmpx_nge_f32"),
@@ -1774,10 +1776,12 @@ void TestNewShaderRecompilerMoreAluFamilies() {
"VOP3-encoded VOP1 find-first-bit-low did not lower through shared IR"); "VOP3-encoded VOP1 find-first-bit-low did not lower through shared IR");
Check(Common::ContainsStr(result.ir_dump, "CompareGtF32"), Check(Common::ContainsStr(result.ir_dump, "CompareGtF32"),
"VOPC float compare did not lower to IR"); "VOPC float compare did not lower to IR");
Check(Common::ContainsStr(result.ir_dump, "CompareMaskGtF32 exec_lo"), Check(Common::ContainsStr(result.ir_dump, "CompareMaskGtF32 vcc_lo") &&
"VOPC float compare-and-mask did not lower to exec mask IR"); Common::ContainsStr(result.ir_dump, "MoveU32 exec_lo, vcc_lo"),
Check(Common::ContainsStr(result.ir_dump, "CompareMaskGtU32 exec_lo"), "VOPC float compare-and-mask did not lower to VCC plus EXEC IR");
"VOPC uint compare-and-mask did not lower to exec mask IR"); Check(Common::ContainsStr(result.ir_dump, "CompareMaskGtU32 vcc_lo") &&
Common::ContainsStr(result.ir_dump, "MoveU32 exec_hi, vcc_hi"),
"VOPC uint compare-and-mask did not lower to VCC plus EXEC IR");
Check(Common::ContainsStr(result.ir_dump, "CompareFalse vcc_lo, v6, v6"), Check(Common::ContainsStr(result.ir_dump, "CompareFalse vcc_lo, v6, v6"),
"VOPC false compare did not lower to shared IR"); "VOPC false compare did not lower to shared IR");
Check(Common::ContainsStr(result.ir_dump, "CompareTrue vcc_lo, v6, v6"), Check(Common::ContainsStr(result.ir_dump, "CompareTrue vcc_lo, v6, v6"),
@@ -1800,17 +1804,17 @@ void TestNewShaderRecompilerMoreAluFamilies() {
"VOPC SDWA unordered-not-equal compare did not lower to scalar-destination IR"); "VOPC SDWA unordered-not-equal compare did not lower to scalar-destination IR");
Check(Common::ContainsStr(result.ir_dump, "CompareUnordGeF32 vcc_lo, v6, v6"), Check(Common::ContainsStr(result.ir_dump, "CompareUnordGeF32 vcc_lo, v6, v6"),
"VOPC unordered-greater-equal compare did not lower to shared IR"); "VOPC unordered-greater-equal compare did not lower to shared IR");
Check(Common::ContainsStr(result.ir_dump, "CompareMaskUnordLtF32 exec_lo, v6, v6"), Check(Common::ContainsStr(result.ir_dump, "CompareMaskUnordLtF32 vcc_lo, v6, v6"),
"VOPC unordered-less compare-and-mask did not lower to shared IR"); "VOPC unordered-less compare-and-mask did not lower to shared IR");
Check(Common::ContainsStr(result.ir_dump, "CompareMaskUnordEqF32 exec_lo, v6, v6"), Check(Common::ContainsStr(result.ir_dump, "CompareMaskUnordEqF32 vcc_lo, v6, v6"),
"VOPC unordered-equal compare-and-mask did not lower to shared IR"); "VOPC unordered-equal compare-and-mask did not lower to shared IR");
Check(Common::ContainsStr(result.ir_dump, "CompareMaskUnordLeF32 exec_lo, v6, v6"), Check(Common::ContainsStr(result.ir_dump, "CompareMaskUnordLeF32 vcc_lo, v6, v6"),
"VOPC unordered-less-equal compare-and-mask did not lower to shared IR"); "VOPC unordered-less-equal compare-and-mask did not lower to shared IR");
Check(Common::ContainsStr(result.ir_dump, "CompareMaskUnordGtF32 exec_lo, v6, v6"), Check(Common::ContainsStr(result.ir_dump, "CompareMaskUnordGtF32 vcc_lo, v6, v6"),
"VOPC unordered-greater compare-and-mask did not lower to shared IR"); "VOPC unordered-greater compare-and-mask did not lower to shared IR");
Check(Common::ContainsStr(result.ir_dump, "CompareMaskUnordNeF32 exec_lo, v6, v6"), Check(Common::ContainsStr(result.ir_dump, "CompareMaskUnordNeF32 vcc_lo, v6, v6"),
"VOPC unordered-not-equal compare-and-mask did not lower to shared IR"); "VOPC unordered-not-equal compare-and-mask did not lower to shared IR");
Check(Common::ContainsStr(result.ir_dump, "CompareMaskUnordGeF32 exec_lo, v6, v6"), Check(Common::ContainsStr(result.ir_dump, "CompareMaskUnordGeF32 vcc_lo, v6, v6"),
"VOPC unordered-greater-equal compare-and-mask did not lower to shared IR"); "VOPC unordered-greater-equal compare-and-mask did not lower to shared IR");
Check(Common::ContainsStr(result.ir_dump, "CompareFalse vcc_lo, v5, v5"), Check(Common::ContainsStr(result.ir_dump, "CompareFalse vcc_lo, v5, v5"),
"VOPC integer false compare did not lower to shared IR"); "VOPC integer false compare did not lower to shared IR");
@@ -2687,8 +2691,9 @@ void TestNewShaderRecompilerSignedCompareAlu() {
"signed halfword greater-or-equal compare did not lower to IR"); "signed halfword greater-or-equal compare did not lower to IR");
Check(Common::ContainsStr(result.ir_dump, "CompareLtU16"), Check(Common::ContainsStr(result.ir_dump, "CompareLtU16"),
"unsigned halfword less-than compare did not lower to IR"); "unsigned halfword less-than compare did not lower to IR");
Check(Common::ContainsStr(result.ir_dump, "CompareMaskGtI32 exec_lo"), Check(Common::ContainsStr(result.ir_dump, "CompareMaskGtI32 vcc_lo") &&
"signed compare-and-mask did not lower to exec mask IR"); Common::ContainsStr(result.ir_dump, "MoveU32 exec_lo, vcc_lo"),
"signed compare-and-mask did not lower to VCC plus EXEC IR");
Check(SpirvContainsOpcode(result.spirv, 173), "SPIR-V binary does not contain OpSGreaterThan"); Check(SpirvContainsOpcode(result.spirv, 173), "SPIR-V binary does not contain OpSGreaterThan");
Check(SpirvContainsOpcode(result.spirv, 177), "SPIR-V binary does not contain OpSLessThan"); Check(SpirvContainsOpcode(result.spirv, 177), "SPIR-V binary does not contain OpSLessThan");
Check(SpirvContainsOpcode(result.spirv, 202), Check(SpirvContainsOpcode(result.spirv, 202),