mirrors/shadPS4
1
0
Fork 0
mirror of https://github.com/shadps4-emu/shadPS4.git synced 2025-12-12 14:48:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

Initial support of Geometry shaders (#1244)

Browse Source 
* video_core: initial GS support

* fix for components mapping; missing prim type
This commit is contained in:
psucien
2024-10-06 00:26:50 +02:00
committed by GitHub
parent 5bb45dc7ba
commit 927bb0c175
40 changed files with 944 additions and 268 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/shader_recompiler/frontend/translate/data_share.cpp
View File

@@ -55,12 +55,6 @@ void Translator::EmitDataShare(const GcnInst& inst) {
}
}
// SOPP
void Translator::S_BARRIER() {
ir.Barrier();
}
// VOP2
void Translator::V_READFIRSTLANE_B32(const GcnInst& inst) {

8
src/shader_recompiler/frontend/translate/scalar_alu.cpp
View File

@@ -540,14 +540,6 @@ void Translator::S_BREV_B32(const GcnInst& inst) {
SetDst(inst.dst[0], ir.BitReverse(GetSrc(inst.src[0])));
}
void Translator::S_GETPC_B64(u32 pc, const GcnInst& inst) {
// This only really exists to let resource tracking pass know
// there is an inline cbuf.
const IR::ScalarReg dst{inst.dst[0].code};
ir.SetScalarReg(dst, ir.Imm32(pc));
ir.SetScalarReg(dst + 1, ir.Imm32(0));
}
void Translator::S_AND_SAVEEXEC_B64(const GcnInst& inst) {
// This instruction normally operates on 64-bit data (EXEC, VCC, SGPRs)
// However here we flatten it to 1-bit EXEC and 1-bit VCC. For the destination

75
src/shader_recompiler/frontend/translate/scalar_flow.cpp Normal file
View File

@@ -0,0 +1,75 @@
// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/frontend/opcodes.h"
#include "shader_recompiler/frontend/translate/translate.h"
namespace Shader::Gcn {
void Translator::EmitFlowControl(u32 pc, const GcnInst& inst) {
switch (inst.opcode) {
case Opcode::S_BARRIER:
return S_BARRIER();
case Opcode::S_TTRACEDATA:
LOG_WARNING(Render_Vulkan, "S_TTRACEDATA instruction!");
return;
case Opcode::S_GETPC_B64:
return S_GETPC_B64(pc, inst);
case Opcode::S_WAITCNT:
case Opcode::S_NOP:
case Opcode::S_ENDPGM:
case Opcode::S_CBRANCH_EXECZ:
case Opcode::S_CBRANCH_SCC0:
case Opcode::S_CBRANCH_SCC1:
case Opcode::S_CBRANCH_VCCNZ:
case Opcode::S_CBRANCH_VCCZ:
case Opcode::S_CBRANCH_EXECNZ:
case Opcode::S_BRANCH:
return;
case Opcode::S_SENDMSG:
S_SENDMSG(inst);
return;
default:
UNREACHABLE();
}
}
void Translator::S_BARRIER() {
ir.Barrier();
}
void Translator::S_GETPC_B64(u32 pc, const GcnInst& inst) {
// This only really exists to let resource tracking pass know
// there is an inline cbuf.
const IR::ScalarReg dst{inst.dst[0].code};
ir.SetScalarReg(dst, ir.Imm32(pc));
ir.SetScalarReg(dst + 1, ir.Imm32(0));
}
void Translator::S_SENDMSG(const GcnInst& inst) {
const auto& simm = reinterpret_cast<const SendMsgSimm&>(inst.control.sopp.simm);
switch (simm.msg) {
case SendMsgSimm::Message::Gs: {
switch (simm.op) {
case SendMsgSimm::GsOp::Nop:
break;
case SendMsgSimm::GsOp::Cut:
ir.EmitPrimitive();
break;
case SendMsgSimm::GsOp::Emit:
ir.EmitVertex();
break;
default:
UNREACHABLE();
}
break;
}
case SendMsgSimm::Message::GsDone: {
break;
}
default:
UNREACHABLE();
}
}
} // namespace Shader::Gcn

45
src/shader_recompiler/frontend/translate/translate.cpp
View File

@@ -10,6 +10,7 @@
#include "shader_recompiler/info.h"
#include "shader_recompiler/runtime_info.h"
#include "video_core/amdgpu/resource.h"
#include "video_core/amdgpu/types.h"
#define MAGIC_ENUM_RANGE_MIN 0
#define MAGIC_ENUM_RANGE_MAX 1515
@@ -35,6 +36,7 @@ void Translator::EmitPrologue() {
IR::VectorReg dst_vreg = IR::VectorReg::V0;
switch (info.stage) {
case Stage::Vertex:
case Stage::Export:
// v0: vertex ID, always present
ir.SetVectorReg(dst_vreg++, ir.GetAttributeU32(IR::Attribute::VertexId));
// v1: instance ID, step rate 0
@@ -76,6 +78,20 @@ void Translator::EmitPrologue() {
ir.SetScalarReg(dst_sreg++, ir.GetAttributeU32(IR::Attribute::WorkgroupId, 2));
}
break;
case Stage::Geometry:
switch (runtime_info.gs_info.out_primitive[0]) {
case AmdGpu::GsOutputPrimitiveType::TriangleStrip:
ir.SetVectorReg(IR::VectorReg::V3, ir.Imm32(2u)); // vertex 2
[[fallthrough]];
case AmdGpu::GsOutputPrimitiveType::LineStrip:
ir.SetVectorReg(IR::VectorReg::V1, ir.Imm32(1u)); // vertex 1
[[fallthrough]];
default:
ir.SetVectorReg(IR::VectorReg::V0, ir.Imm32(0u)); // vertex 0
break;
}
ir.SetVectorReg(IR::VectorReg::V2, ir.GetAttributeU32(IR::Attribute::PrimitiveId));
break;
default:
throw NotImplementedException("Unknown shader stage");
}
@@ -359,7 +375,7 @@ void Translator::EmitFetch(const GcnInst& inst) {
if (!std::filesystem::exists(dump_dir)) {
std::filesystem::create_directories(dump_dir);
}
const auto filename = fmt::format("vs_{:#018x}_fetch.bin", info.pgm_hash);
const auto filename = fmt::format("vs_{:#018x}.fetch.bin", info.pgm_hash);
const auto file = IOFile{dump_dir / filename, FileAccessMode::Write};
file.WriteRaw<u8>(code, fetch_size);
}
@@ -424,31 +440,6 @@ void Translator::EmitFetch(const GcnInst& inst) {
}
}
void Translator::EmitFlowControl(u32 pc, const GcnInst& inst) {
switch (inst.opcode) {
case Opcode::S_BARRIER:
return S_BARRIER();
case Opcode::S_TTRACEDATA:
LOG_WARNING(Render_Vulkan, "S_TTRACEDATA instruction!");
return;
case Opcode::S_GETPC_B64:
return S_GETPC_B64(pc, inst);
case Opcode::S_WAITCNT:
case Opcode::S_NOP:
case Opcode::S_ENDPGM:
case Opcode::S_CBRANCH_EXECZ:
case Opcode::S_CBRANCH_SCC0:
case Opcode::S_CBRANCH_SCC1:
case Opcode::S_CBRANCH_VCCNZ:
case Opcode::S_CBRANCH_VCCZ:
case Opcode::S_CBRANCH_EXECNZ:
case Opcode::S_BRANCH:
return;
default:
UNREACHABLE();
}
}
void Translator::LogMissingOpcode(const GcnInst& inst) {
LOG_ERROR(Render_Recompiler, "Unknown opcode {} ({}, category = {})",
magic_enum::enum_name(inst.opcode), u32(inst.opcode),
@@ -467,7 +458,7 @@ void Translate(IR::Block* block, u32 pc, std::span<const GcnInst> inst_list, Inf
// Special case for emitting fetch shader.
if (inst.opcode == Opcode::S_SWAPPC_B64) {
ASSERT(info.stage == Stage::Vertex);
ASSERT(info.stage == Stage::Vertex || info.stage == Stage::Export);
translator.EmitFetch(inst);
continue;
}

1
src/shader_recompiler/frontend/translate/translate.h
View File

@@ -116,6 +116,7 @@ public:
// SOPP
void S_BARRIER();
void S_SENDMSG(const GcnInst& inst);
// Scalar Memory
// SMRD

49
src/shader_recompiler/frontend/translate/vector_memory.cpp
View File

@@ -160,9 +160,19 @@ void Translator::EmitVectorMemory(const GcnInst& inst) {
void Translator::BUFFER_LOAD(u32 num_dwords, bool is_typed, const GcnInst& inst) {
const auto& mtbuf = inst.control.mtbuf;
const bool is_ring = mtbuf.glc && mtbuf.slc;
const IR::VectorReg vaddr{inst.src[0].code};
const IR::ScalarReg sharp{inst.src[2].code * 4};
const IR::Value soffset{GetSrc(inst.src[3])};
if (info.stage != Stage::Geometry) {
ASSERT_MSG(soffset.IsImmediate() && soffset.U32() == 0,
"Non immediate offset not supported");
}
const IR::Value address = [&] -> IR::Value {
if (is_ring) {
return ir.CompositeConstruct(ir.GetVectorReg(vaddr), soffset);
}
if (mtbuf.idxen && mtbuf.offen) {
return ir.CompositeConstruct(ir.GetVectorReg(vaddr), ir.GetVectorReg(vaddr + 1));
}
@@ -171,13 +181,12 @@ void Translator::BUFFER_LOAD(u32 num_dwords, bool is_typed, const GcnInst& inst)
}
return {};
}();
const IR::Value soffset{GetSrc(inst.src[3])};
ASSERT_MSG(soffset.IsImmediate() && soffset.U32() == 0, "Non immediate offset not supported");
IR::BufferInstInfo info{};
info.index_enable.Assign(mtbuf.idxen);
info.offset_enable.Assign(mtbuf.offen);
info.inst_offset.Assign(mtbuf.offset);
IR::BufferInstInfo buffer_info{};
buffer_info.index_enable.Assign(mtbuf.idxen);
buffer_info.offset_enable.Assign(mtbuf.offen);
buffer_info.inst_offset.Assign(mtbuf.offset);
buffer_info.ring_access.Assign(is_ring);
if (is_typed) {
const auto dmft = static_cast<AmdGpu::DataFormat>(mtbuf.dfmt);
const auto nfmt = static_cast<AmdGpu::NumberFormat>(mtbuf.nfmt);
@@ -190,7 +199,7 @@ void Translator::BUFFER_LOAD(u32 num_dwords, bool is_typed, const GcnInst& inst)
const IR::Value handle =
ir.CompositeConstruct(ir.GetScalarReg(sharp), ir.GetScalarReg(sharp + 1),
ir.GetScalarReg(sharp + 2), ir.GetScalarReg(sharp + 3));
const IR::Value value = ir.LoadBuffer(num_dwords, handle, address, info);
const IR::Value value = ir.LoadBuffer(num_dwords, handle, address, buffer_info);
const IR::VectorReg dst_reg{inst.src[1].code};
if (num_dwords == 1) {
ir.SetVectorReg(dst_reg, IR::U32{value});
@@ -230,9 +239,20 @@ void Translator::BUFFER_LOAD_FORMAT(u32 num_dwords, const GcnInst& inst) {
void Translator::BUFFER_STORE(u32 num_dwords, bool is_typed, const GcnInst& inst) {
const auto& mtbuf = inst.control.mtbuf;
const bool is_ring = mtbuf.glc && mtbuf.slc;
const IR::VectorReg vaddr{inst.src[0].code};
const IR::ScalarReg sharp{inst.src[2].code * 4};
const IR::Value address = [&] -> IR::Value {
const IR::Value soffset{GetSrc(inst.src[3])};
if (info.stage != Stage::Export && info.stage != Stage::Geometry) {
ASSERT_MSG(soffset.IsImmediate() && soffset.U32() == 0,
"Non immediate offset not supported");
}
IR::Value address = [&] -> IR::Value {
if (is_ring) {
return ir.CompositeConstruct(ir.GetVectorReg(vaddr), soffset);
}
if (mtbuf.idxen && mtbuf.offen) {
return ir.CompositeConstruct(ir.GetVectorReg(vaddr), ir.GetVectorReg(vaddr + 1));
}
@@ -241,13 +261,12 @@ void Translator::BUFFER_STORE(u32 num_dwords, bool is_typed, const GcnInst& inst
}
return {};
}();
const IR::Value soffset{GetSrc(inst.src[3])};
ASSERT_MSG(soffset.IsImmediate() && soffset.U32() == 0, "Non immediate offset not supported");
IR::BufferInstInfo info{};
info.index_enable.Assign(mtbuf.idxen);
info.offset_enable.Assign(mtbuf.offen);
info.inst_offset.Assign(mtbuf.offset);
IR::BufferInstInfo buffer_info{};
buffer_info.index_enable.Assign(mtbuf.idxen);
buffer_info.offset_enable.Assign(mtbuf.offen);
buffer_info.inst_offset.Assign(mtbuf.offset);
buffer_info.ring_access.Assign(is_ring);
if (is_typed) {
const auto dmft = static_cast<AmdGpu::DataFormat>(mtbuf.dfmt);
const auto nfmt = static_cast<AmdGpu::NumberFormat>(mtbuf.nfmt);
@@ -278,7 +297,7 @@ void Translator::BUFFER_STORE(u32 num_dwords, bool is_typed, const GcnInst& inst
const IR::Value handle =
ir.CompositeConstruct(ir.GetScalarReg(sharp), ir.GetScalarReg(sharp + 1),
ir.GetScalarReg(sharp + 2), ir.GetScalarReg(sharp + 3));
ir.StoreBuffer(num_dwords, handle, address, value, info);
ir.StoreBuffer(num_dwords, handle, address, value, buffer_info);
}
void Translator::BUFFER_STORE_FORMAT(u32 num_dwords, const GcnInst& inst) {