Mirrors/yuzu
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge pull request #3409 from ReinUsesLisp/host-queries

Browse source 
query_cache: Implement a query cache and query 21 (samples passed)
This commit is contained in:
Fernando Sahmkow 2020-02-18 11:31:06 -04:00 • committed by GitHub
commit 93acfbd3a5
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
18 changed files with 1043 additions and 51 deletions
Download patch file Download diff file Expand all files Collapse all files

77
src/video_core/engines/maxwell_3d.cpp
View file

@ -4,6 +4,7 @@
#include <cinttypes>
#include <cstring>
#include <optional>
#include "common/assert.h"
#include "core/core.h"
#include "core/core_timing.h"
@ -16,6 +17,8 @@
namespace Tegra::Engines {
using VideoCore::QueryType;
/// First register id that is actually a Macro call.
constexpr u32 MacroRegistersStart = 0xE00;
@ -400,6 +403,10 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
ProcessQueryCondition();
break;
}
case MAXWELL3D_REG_INDEX(counter_reset): {
ProcessCounterReset();
break;
}
case MAXWELL3D_REG_INDEX(sync_info): {
ProcessSyncPoint();
break;
@ -544,40 +551,28 @@ void Maxwell3D::ProcessQueryGet() {
"Units other than CROP are unimplemented");
switch (regs.query.query_get.operation) {
case Regs::QueryOperation::Release: {
const u64 result = regs.query.query_sequence;
StampQueryResult(result, regs.query.query_get.short_query == 0);
case Regs::QueryOperation::Release:
StampQueryResult(regs.query.query_sequence, regs.query.query_get.short_query == 0);
break;
}
case Regs::QueryOperation::Acquire: {
// Todo(Blinkhawk): Under this operation, the GPU waits for the CPU
// to write a value that matches the current payload.
case Regs::QueryOperation::Acquire:
// TODO(Blinkhawk): Under this operation, the GPU waits for the CPU to write a value that
// matches the current payload.
UNIMPLEMENTED_MSG("Unimplemented query operation ACQUIRE");
break;
}
case Regs::QueryOperation::Counter: {
u64 result{};
switch (regs.query.query_get.select) {
case Regs::QuerySelect::Zero:
result = 0;
break;
default:
result = 1;
UNIMPLEMENTED_MSG("Unimplemented query select type {}",
static_cast<u32>(regs.query.query_get.select.Value()));
case Regs::QueryOperation::Counter:
if (const std::optional<u64> result = GetQueryResult()) {
// If the query returns an empty optional it means it's cached and deferred.
// In this case we have a non-empty result, so we stamp it immediately.
StampQueryResult(*result, regs.query.query_get.short_query == 0);
}
StampQueryResult(result, regs.query.query_get.short_query == 0);
break;
}
case Regs::QueryOperation::Trap: {
case Regs::QueryOperation::Trap:
UNIMPLEMENTED_MSG("Unimplemented query operation TRAP");
break;
}
default: {
default:
UNIMPLEMENTED_MSG("Unknown query operation");
break;
}
}
}
void Maxwell3D::ProcessQueryCondition() {
@ -593,20 +588,20 @@ void Maxwell3D::ProcessQueryCondition() {
}
case Regs::ConditionMode::ResNonZero: {
Regs::QueryCompare cmp;
memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
memory_manager.ReadBlock(condition_address, &cmp, sizeof(cmp));
execute_on = cmp.initial_sequence != 0U && cmp.initial_mode != 0U;
break;
}
case Regs::ConditionMode::Equal: {
Regs::QueryCompare cmp;
memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
memory_manager.ReadBlock(condition_address, &cmp, sizeof(cmp));
execute_on =
cmp.initial_sequence == cmp.current_sequence && cmp.initial_mode == cmp.current_mode;
break;
}
case Regs::ConditionMode::NotEqual: {
Regs::QueryCompare cmp;
memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
memory_manager.ReadBlock(condition_address, &cmp, sizeof(cmp));
execute_on =
cmp.initial_sequence != cmp.current_sequence || cmp.initial_mode != cmp.current_mode;
break;
@ -619,6 +614,18 @@ void Maxwell3D::ProcessQueryCondition() {
}
}
void Maxwell3D::ProcessCounterReset() {
switch (regs.counter_reset) {
case Regs::CounterReset::SampleCnt:
rasterizer.ResetCounter(QueryType::SamplesPassed);
break;
default:
LOG_WARNING(Render_OpenGL, "Unimplemented counter reset={}",
static_cast<int>(regs.counter_reset));
break;
}
}
void Maxwell3D::ProcessSyncPoint() {
const u32 sync_point = regs.sync_info.sync_point.Value();
const u32 increment = regs.sync_info.increment.Value();
@ -661,6 +668,22 @@ void Maxwell3D::DrawArrays() {
}
}
std::optional<u64> Maxwell3D::GetQueryResult() {
switch (regs.query.query_get.select) {
case Regs::QuerySelect::Zero:
return 0;
case Regs::QuerySelect::SamplesPassed:
// Deferred.
rasterizer.Query(regs.query.QueryAddress(), VideoCore::QueryType::SamplesPassed,
system.GPU().GetTicks());
return {};
default:
UNIMPLEMENTED_MSG("Unimplemented query select type {}",
static_cast<u32>(regs.query.query_get.select.Value()));
return 1;
}
}
void Maxwell3D::ProcessCBBind(std::size_t stage_index) {
// Bind the buffer currently in CB_ADDRESS to the specified index in the desired shader stage.
auto& shader = state.shader_stages[stage_index];

42
src/video_core/engines/maxwell_3d.h
View file

@ -6,6 +6,7 @@
#include <array>
#include <bitset>
#include <optional>
#include <type_traits>
#include <unordered_map>
#include <vector>
@ -409,6 +410,27 @@ public:
Linear = 1,
};
enum class CounterReset : u32 {
SampleCnt = 0x01,
Unk02 = 0x02,
Unk03 = 0x03,
Unk04 = 0x04,
EmittedPrimitives = 0x10, // Not tested
Unk11 = 0x11,
Unk12 = 0x12,
Unk13 = 0x13,
Unk15 = 0x15,
Unk16 = 0x16,
Unk17 = 0x17,
Unk18 = 0x18,
Unk1A = 0x1A,
Unk1B = 0x1B,
Unk1C = 0x1C,
Unk1D = 0x1D,
Unk1E = 0x1E,
GeneratedPrimitives = 0x1F,
};
struct Cull {
enum class FrontFace : u32 {
ClockWise = 0x0900,
@ -857,7 +879,7 @@ public:
BitField<7, 1, u32> c7;
} clip_distance_enabled;
INSERT_UNION_PADDING_WORDS(0x1);
u32 samplecnt_enable;
float point_size;
@ -865,7 +887,11 @@ public:
u32 point_sprite_enable;
INSERT_UNION_PADDING_WORDS(0x5);
INSERT_UNION_PADDING_WORDS(0x3);
CounterReset counter_reset;
INSERT_UNION_PADDING_WORDS(0x1);
u32 zeta_enable;
@ -1412,12 +1438,15 @@ private:
/// Handles a write to the QUERY_GET register.
void ProcessQueryGet();
// Writes the query result accordingly
/// Writes the query result accordingly.
void StampQueryResult(u64 payload, bool long_query);
// Handles Conditional Rendering
/// Handles conditional rendering.
void ProcessQueryCondition();
/// Handles counter resets.
void ProcessCounterReset();
/// Handles writes to syncing register.
void ProcessSyncPoint();
@ -1434,6 +1463,9 @@ private:
// Handles a instance drawcall from MME
void StepInstance(MMEDrawMode expected_mode, u32 count);
/// Returns a query's value or an empty object if the value will be deferred through a cache.
std::optional<u64> GetQueryResult();
};
#define ASSERT_REG_POSITION(field_name, position) \
@ -1499,8 +1531,10 @@ ASSERT_REG_POSITION(screen_y_control, 0x4EB);
ASSERT_REG_POSITION(vb_element_base, 0x50D);
ASSERT_REG_POSITION(vb_base_instance, 0x50E);
ASSERT_REG_POSITION(clip_distance_enabled, 0x544);
ASSERT_REG_POSITION(samplecnt_enable, 0x545);
ASSERT_REG_POSITION(point_size, 0x546);
ASSERT_REG_POSITION(point_sprite_enable, 0x548);
ASSERT_REG_POSITION(counter_reset, 0x54C);
ASSERT_REG_POSITION(zeta_enable, 0x54E);
ASSERT_REG_POSITION(multisample_control, 0x54F);
ASSERT_REG_POSITION(condition, 0x554);