video_core: implement PM4 disassembler

This commit is contained in:
Random06457 2024-08-25 23:25:11 +09:00
parent a6a9fff666
commit db619463dd
3 changed files with 328 additions and 15 deletions

View File

@ -4,6 +4,9 @@
#include "gnm_error.h"
#include "gnmdriver.h"
#define MAGIC_ENUM_RANGE_MIN 0
#define MAGIC_ENUM_RANGE_MAX 0x88
#include <magic_enum.hpp>
#include "common/assert.h"
#include "common/config.h"
#include "common/debug.h"
@ -320,6 +323,296 @@ static void WaitGpuIdle() {
cv_lock.wait(lock, [] { return submission_lock == 0; });
}
static std::string DisassemblePM4(std::span<const u32> cmds) {
std::string ret = "";
// TODO: replace with something better?, appending strings like this is not the most optimized
// thing one can do
auto append = [&ret](std::string_view sv) { ret += sv; };
auto hexdump = [&append](const u32* data, size_t size) {
for (size_t i = 0; i < size / 4;) {
append(fmt::format(" {:08X}: ", i * 4));
for (size_t j = 0; j < 4 && i < size / 4; j++, i++)
append(fmt::format("{:08X} ", data[i]));
append("\n");
}
};
while (!cmds.empty()) {
const auto* header = reinterpret_cast<const PM4Header*>(cmds.data());
const u32 type = header->type;
switch (type) {
case 0:
append(fmt::format("PM4Type0: base={}, count={}\n", header->type0.base.Value(),
header->type0.count.Value()));
for (size_t i = 0; i < header->type0.count.Value(); i++)
append(fmt::format("- 0x{:08X}\n", cmds.data()[1 + i]));
break;
cmds = cmds.subspan(header->type0.NumWords() + 1);
case 1:
UNREACHABLE();
case 2:
append(fmt::format("PM4Type2:\n"));
cmds = cmds.subspan(1);
break;
case 3:
append(fmt::format(
"PM4Type3: predicate={}, shader_type={}, opcode={}(0x{:X}), count={}\n",
magic_enum::enum_name(header->type3.predicate.Value()),
magic_enum::enum_name(header->type3.shader_type.Value()),
magic_enum::enum_name(header->type3.opcode.Value()),
static_cast<u32>(header->type3.opcode.Value()), header->type3.count.Value()));
switch (header->type3.opcode.Value()) {
case PM4ItOpcode::SetContextReg:
case PM4ItOpcode::SetConfigReg:
case PM4ItOpcode::SetShReg:
case PM4ItOpcode::SetUconfigReg: {
const auto* set_data = reinterpret_cast<const PM4CmdSetData*>(header);
append(fmt::format("- reg_offset=0x{:08X}\n", set_data->reg_offset.Value()));
append(fmt::format("- index=0x{:08X}\n", set_data->index.Value()));
append("- data:\n");
hexdump(set_data->data, set_data->Size());
break;
}
case PM4ItOpcode::WriteConstRam: {
const auto* write_const = reinterpret_cast<const PM4WriteConstRam*>(header);
append(fmt::format("- offset=0x{:X}\n", write_const->Offset()));
append(fmt::format("- size=0x{:X}\n", write_const->Size()));
append("- data:\n");
hexdump(write_const->data, write_const->Size());
break;
}
case PM4ItOpcode::DmaData: {
const auto* dma_data = reinterpret_cast<const PM4DmaData*>(header);
append(fmt::format("- engine=0x{:X}\n", dma_data->engine.Value()));
append(fmt::format("- command=0x{:X}\n", dma_data->command));
append(fmt::format("- cp_sync=0x{:X}\n", dma_data->cp_sync.Value()));
append(fmt::format("- src: atc=0x{:X}, cache_policy=0x{:X}, volatile=0x{:X}, "
"sel=0x{:X}, addr=0x{:016X}\n",
dma_data->src_atc.Value(), dma_data->src_cache_policy.Value(),
dma_data->src_volatile.Value(), dma_data->src_sel.Value(),
((u64)dma_data->src_addr_hi << 32) | dma_data->src_addr_lo));
append(fmt::format("- dst: atc=0x{:X}, cache_policy=0x{:X}, volatile=0x{:X}, "
"sel=0x{:X}, addr=0x{:016X}\n",
dma_data->dst_atc.Value(), dma_data->dst_cache_policy.Value(),
dma_data->dst_volatile.Value(), dma_data->dst_sel.Value(),
((u64)dma_data->dst_addr_hi << 32) | dma_data->dst_addr_lo));
break;
}
case PM4ItOpcode::ContextControl: {
const auto* ctx_ctl = reinterpret_cast<const PM4CmdContextControl*>(header);
append(fmt::format("- load_control:\n"));
append(fmt::format(" - enable_single_cntx_config_reg={}\n",
ctx_ctl->load_control.enable_single_cntx_config_reg.Value()));
append(fmt::format(" - enable_multi_cntx_render_reg={}\n",
ctx_ctl->load_control.enable_multi_cntx_render_reg.Value()));
append(fmt::format(" - enable_user_config_reg__CI={}\n",
ctx_ctl->load_control.enable_user_config_reg__CI.Value()));
append(fmt::format(" - enable_gfx_sh_reg={}\n",
ctx_ctl->load_control.enable_gfx_sh_reg.Value()));
append(fmt::format(" - enable_cs_sh_reg={}\n",
ctx_ctl->load_control.enable_cs_sh_reg.Value()));
append(fmt::format(" - enable_dw={}\n", ctx_ctl->load_control.enable_dw.Value()));
append(fmt::format("- shadow_enable:\n"));
append(fmt::format(" - enable_single_cntx_config_reg={}\n",
ctx_ctl->shadow_enable.enable_single_cntx_config_reg.Value()));
append(fmt::format(" - enable_multi_cntx_render_reg={}\n",
ctx_ctl->shadow_enable.enable_multi_cntx_render_reg.Value()));
append(fmt::format(" - enable_user_config_reg__CI={}\n",
ctx_ctl->shadow_enable.enable_user_config_reg__CI.Value()));
append(fmt::format(" - enable_gfx_sh_reg={}\n",
ctx_ctl->shadow_enable.enable_gfx_sh_reg.Value()));
append(fmt::format(" - enable_cs_sh_reg={}\n",
ctx_ctl->shadow_enable.enable_cs_sh_reg.Value()));
append(fmt::format(" - enable_dw={}\n", ctx_ctl->shadow_enable.enable_dw.Value()));
break;
}
case PM4ItOpcode::NumInstances: {
const auto* num_instances = reinterpret_cast<const PM4CmdDrawNumInstances*>(header);
append(fmt::format("- num_instances={}\n", num_instances->num_instances));
break;
}
case PM4ItOpcode::DrawIndexAuto: {
const auto* draw_index = reinterpret_cast<const PM4CmdDrawIndexAuto*>(header);
append(fmt::format("- index_count={}\n", draw_index->index_count));
append(fmt::format("- draw_initiator={}\n", draw_index->draw_initiator));
break;
}
case PM4ItOpcode::EventWriteEop: {
const auto* event_eop = reinterpret_cast<const PM4CmdEventWriteEop*>(header);
append(fmt::format("- event_type=0x{:X}\n", event_eop->event_type.Value()));
append(fmt::format("- event_index={}\n", event_eop->event_index.Value()));
append(fmt::format("- address=0x{:016X}\n",
reinterpret_cast<uintptr_t>(event_eop->Address<void>())));
append(fmt::format("- data=0x{:016X}\n", event_eop->DataQWord()));
append(fmt::format("- int_sel={}\n",
magic_enum::enum_name(event_eop->int_sel.Value())));
append(fmt::format("- data_sel={}\n",
magic_enum::enum_name(event_eop->data_sel.Value())));
break;
}
case PM4ItOpcode::AcquireMem: {
const auto* acquire_mem = reinterpret_cast<const PM4CmdAcquireMem*>(header);
append(fmt::format("- cp_coher_cntl=0x{:X}\n", acquire_mem->cp_coher_cntl));
append(fmt::format("- cp_coher_size=0x{:X}\n",
((u64)acquire_mem->cp_coher_size_hi << 32) |
acquire_mem->cp_coher_size_lo));
append(fmt::format("- cp_coher_base=0x{:X}\n",
((u64)acquire_mem->cp_coher_base_hi << 32) |
acquire_mem->cp_coher_base_lo));
append(fmt::format("- poll_interval=0x{:X}\n", acquire_mem->poll_interval));
break;
}
case PM4ItOpcode::WriteData: {
const auto* write_data = reinterpret_cast<const PM4CmdWriteData*>(header);
append(fmt::format("- dst_sel=0x{:X}\n", write_data->dst_sel.Value()));
append(fmt::format("- wr_one_addr={}\n", write_data->wr_one_addr.Value()));
append(fmt::format("- wr_confirm={}\n", write_data->wr_confirm.Value()));
append(fmt::format("- engine_sel={}\n", write_data->engine_sel.Value()));
append(fmt::format("- dst_addr=0x{:016X}\n", write_data->addr64));
append("- data:\n");
hexdump(write_data->data, write_data->Size());
break;
}
case PM4ItOpcode::WaitRegMem: {
const auto* write_reg_mem = reinterpret_cast<const PM4CmdWaitRegMem*>(header);
append(fmt::format("- function={}\n",
magic_enum::enum_name(write_reg_mem->function.Value())));
append(fmt::format("- mem_space={}\n",
magic_enum::enum_name(write_reg_mem->mem_space.Value())));
append(fmt::format("- engine={}\n",
magic_enum::enum_name(write_reg_mem->engine.Value())));
append(fmt::format("- poll_addr=0x{:X}\n", write_reg_mem->Address<uintptr_t>()));
append(fmt::format("- ref=0x{:X}\n", write_reg_mem->ref));
append(fmt::format("- mask=0x{:X}\n", write_reg_mem->mask));
append(fmt::format("- poll_interval=0x{:X}\n", write_reg_mem->poll_interval));
break;
}
case PM4ItOpcode::ReleaseMem: {
const auto* release_mem = reinterpret_cast<const PM4CmdReleaseMem*>(header);
append(fmt::format("- event_type=0x{:X}", release_mem->event_type.Value()));
append(fmt::format("- event_index=0x{:X}", release_mem->event_index.Value()));
append(fmt::format("- tcl1_vol_action_ena=0x{:X}",
release_mem->tcl1_vol_action_ena.Value()));
append(fmt::format("- tc_vol_action_ena=0x{:X}",
release_mem->tc_vol_action_ena.Value()));
append(fmt::format("- tc_wb_action_ena=0x{:X}",
release_mem->tc_wb_action_ena.Value()));
append(fmt::format("- tcl1__action_ena=0x{:X}",
release_mem->tcl1__action_ena.Value()));
append(fmt::format("- tc_action_ena=0x{:X}", release_mem->tc_action_ena.Value()));
append(fmt::format("- cache_policy=0x{:X}", release_mem->cache_policy.Value()));
append(fmt::format("- dst_sel=0x{:X}", release_mem->dst_sel.Value()));
append(fmt::format("- int_sel={}\n",
magic_enum::enum_name(release_mem->int_sel.Value())));
append(fmt::format("- data_sel={}\n",
magic_enum::enum_name(release_mem->data_sel.Value())));
append(fmt::format("- addr=0x{:X}\n",
reinterpret_cast<uintptr_t>(release_mem->Address<void>())));
append(fmt::format("- data=0x{:X}", release_mem->DataQWord()));
break;
}
case PM4ItOpcode::IndirectBuffer: {
const auto* indirect_buffer = reinterpret_cast<const PM4CmdIndirectBuffer*>(header);
append(fmt::format("- address=0x{:X}\n",
reinterpret_cast<uintptr_t>(indirect_buffer->Address<void>())));
append(fmt::format("- size=0x{:X}\n", indirect_buffer->ib_size.Value()));
append(fmt::format("- chain=0x{:X}\n", indirect_buffer->chain.Value()));
append(fmt::format("- vmid=0x{:X}\n", indirect_buffer->vmid.Value()));
break;
}
case PM4ItOpcode::IndexBufferSize: {
const auto* index_size = reinterpret_cast<const PM4CmdDrawIndexBufferSize*>(header);
append(fmt::format("- num_indices=0x{:X}\n", index_size->num_indices));
break;
}
case PM4ItOpcode::IndexBase: {
const auto* index_base = reinterpret_cast<const PM4CmdDrawIndexBase*>(header);
append(fmt::format("- address=0x{:X}\n",
((u64)index_base->addr_hi << 32) | index_base->addr_lo));
break;
}
case PM4ItOpcode::DispatchDirect: {
const auto* dispatch_direct = reinterpret_cast<const PM4CmdDispatchDirect*>(header);
append(fmt::format("- dim_x=0x{:X}\n", dispatch_direct->dim_x));
append(fmt::format("- dim_y=0x{:X}\n", dispatch_direct->dim_y));
append(fmt::format("- dim_z=0x{:X}\n", dispatch_direct->dim_z));
append(fmt::format("- dispatch_initiator=0x{:X}\n",
dispatch_direct->dispatch_initiator));
break;
}
case PM4ItOpcode::DumpConstRam: {
const auto* dump_const = reinterpret_cast<const PM4DumpConstRam*>(header);
append(fmt::format("- address=0x{:X}\n", dump_const->Address<uintptr_t>()));
append(fmt::format("- offset=0x{:X}\n", dump_const->Offset()));
append(fmt::format("- size=0x{:X}\n", dump_const->Size()));
break;
}
case PM4ItOpcode::EventWriteEos: {
const auto* event_eos = reinterpret_cast<const PM4CmdEventWriteEos*>(header);
append(fmt::format("- event_type=0x{:X}\n", event_eos->event_type.Value()));
append(fmt::format("- event_index=0x{:X}\n", event_eos->event_index.Value()));
append(fmt::format("- address=0x{:X}\n", event_eos->Address<uintptr_t>()));
append(fmt::format("- command={}\n",
magic_enum::enum_name(event_eos->command.Value())));
append(fmt::format("- gds_index=0x{:X}\n", event_eos->gds_index.Value()));
append(fmt::format("- size=0x{:X}\n", event_eos->size.Value()));
break;
}
case PM4ItOpcode::DrawIndirect: {
const auto* draw_indirect = reinterpret_cast<const PM4CmdDrawIndirect*>(header);
append(fmt::format("- data_offset=0x{:X}\n", draw_indirect->data_offset));
append(fmt::format("- base_vtx_loc=0x{:X}\n", draw_indirect->base_vtx_loc.Value()));
append(fmt::format("- start_inst_loc=0x{:X}\n",
draw_indirect->start_inst_loc.Value()));
append(fmt::format("- draw_initiator=0x{:X}\n", draw_indirect->draw_initiator));
break;
}
case PM4ItOpcode::IndexType: {
const auto* draw_index_type = reinterpret_cast<const PM4CmdDrawIndexType*>(header);
append(fmt::format("- index_type=0x{:X}\n", draw_index_type->index_type.Value()));
append(fmt::format("- swap_mode=0x{:X}\n", draw_index_type->swap_mode.Value()));
break;
}
case PM4ItOpcode::DrawIndex2: {
const auto* draw_index = reinterpret_cast<const PM4CmdDrawIndex2*>(header);
append(fmt::format("- max_size=0x{:X}\n", draw_index->max_size));
append(fmt::format("- index_base=0x{:X}\n", ((u64)draw_index->index_base_hi << 32) |
draw_index->index_base_lo));
append(fmt::format("- index_count=0x{:X}\n", draw_index->index_count));
append(fmt::format("- draw_initiator=0x{:X}\n", draw_index->draw_initiator));
break;
}
case PM4ItOpcode::DrawIndexOffset2: {
const auto* draw_index_off =
reinterpret_cast<const PM4CmdDrawIndexOffset2*>(header);
append(fmt::format("- max_size=0x{:X}\n", draw_index_off->max_size));
append(fmt::format("- index_offset=0x{:X}\n", draw_index_off->index_offset));
append(fmt::format("- index_count=0x{:X}\n", draw_index_off->index_count));
append(fmt::format("- draw_initiator=0x{:X}\n", draw_index_off->draw_initiator));
break;
}
default:
for (size_t i = 0; i < header->type0.count + 1; i++)
append(fmt::format("- 0x{:08X}\n", cmds.data()[1 + i]));
break;
}
cmds = cmds.subspan(header->type3.NumWords() + 1);
break;
default:
UNREACHABLE();
}
append("\n");
}
return ret;
}
static void DumpCommandList(std::span<const u32> cmd_list, const std::string& postfix) {
using namespace Common::FS;
const auto dump_dir = GetUserPath(PathType::PM4Dir);
@ -332,6 +625,9 @@ static void DumpCommandList(std::span<const u32> cmd_list, const std::string& po
const auto filename = fmt::format("{:08}_{}", frames_submitted, postfix);
const auto file = IOFile{dump_dir / filename, FileAccessMode::Write};
file.WriteSpan(cmd_list);
IOFile txt{dump_dir / (filename + ".txt"), FileAccessMode::Write, FileType::TextFile};
txt.WriteString(DisassemblePM4(cmd_list));
}
// Write a special ending NOP packet with N DWs data block

View File

@ -229,16 +229,14 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
case PM4ItOpcode::SetConfigReg: {
const auto* set_data = reinterpret_cast<const PM4CmdSetData*>(header);
const auto reg_addr = ConfigRegWordOffset + set_data->reg_offset;
const auto* payload = reinterpret_cast<const u32*>(header + 2);
std::memcpy(&regs.reg_array[reg_addr], payload, (count - 1) * sizeof(u32));
std::memcpy(&regs.reg_array[reg_addr], set_data->data, set_data->Size());
break;
}
case PM4ItOpcode::SetContextReg: {
const auto* set_data = reinterpret_cast<const PM4CmdSetData*>(header);
const auto reg_addr = ContextRegWordOffset + set_data->reg_offset;
const auto* payload = reinterpret_cast<const u32*>(header + 2);
std::memcpy(&regs.reg_array[reg_addr], payload, (count - 1) * sizeof(u32));
const auto* payload = set_data->data;
std::memcpy(&regs.reg_array[reg_addr], payload, set_data->Size());
// In the case of HW, render target memory has alignment as color block operates on
// tiles. There is no information of actual resource extents stored in CB context
@ -308,14 +306,14 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
}
case PM4ItOpcode::SetShReg: {
const auto* set_data = reinterpret_cast<const PM4CmdSetData*>(header);
std::memcpy(&regs.reg_array[ShRegWordOffset + set_data->reg_offset], header + 2,
(count - 1) * sizeof(u32));
const auto reg_addr = ShRegWordOffset + set_data->reg_offset;
std::memcpy(&regs.reg_array[reg_addr], set_data->data, set_data->Size());
break;
}
case PM4ItOpcode::SetUconfigReg: {
const auto* set_data = reinterpret_cast<const PM4CmdSetData*>(header);
std::memcpy(&regs.reg_array[UconfigRegWordOffset + set_data->reg_offset],
header + 2, (count - 1) * sizeof(u32));
const auto reg_addr = UconfigRegWordOffset + set_data->reg_offset;
std::memcpy(&regs.reg_array[reg_addr], set_data->data, set_data->Size());
break;
}
case PM4ItOpcode::IndexType: {
@ -522,8 +520,8 @@ Liverpool::Task Liverpool::ProcessCompute(std::span<const u32> acb, int vqid) {
}
case PM4ItOpcode::SetShReg: {
const auto* set_data = reinterpret_cast<const PM4CmdSetData*>(header);
std::memcpy(&regs.reg_array[ShRegWordOffset + set_data->reg_offset], header + 2,
(count - 1) * sizeof(u32));
const auto reg_addr = ShRegWordOffset + set_data->reg_offset;
std::memcpy(&regs.reg_array[reg_addr], set_data->data, set_data->Size());
break;
}
case PM4ItOpcode::DispatchDirect: {

View File

@ -187,6 +187,11 @@ struct PM4CmdSetData {
BitField<28, 4, u32> index; ///< Index for UCONFIG/CONTEXT on CI+
///< Program to zero for other opcodes and on SI
};
u32 data[0];
[[nodiscard]] u32 Size() const {
return header.count << 2u;
}
template <PM4ShaderType type = PM4ShaderType::ShaderGraphics, typename... Args>
static constexpr u32* SetContextReg(u32* cmdbuf, Args... data) {
@ -364,6 +369,16 @@ struct PM4CmdEventWriteEop {
}
};
struct PM4CmdAcquireMem {
PM4Type3Header header;
u32 cp_coher_cntl;
u32 cp_coher_size_lo;
u32 cp_coher_size_hi;
u32 cp_coher_base_lo;
u32 cp_coher_base_hi;
u32 poll_interval;
};
struct PM4DmaData {
PM4Type3Header header;
union {
@ -475,6 +490,10 @@ struct PM4CmdWriteData {
addr64 = static_cast<u64>(addr);
}
u32 Size() const {
return (header.count.Value() - 2) * 4;
}
template <typename T>
T Address() const {
return reinterpret_cast<T>(addr64);