hdkv/yuzu
1
Fork 0
Code Issues Pull Requests Releases 1 Activity

early-access version 3012

Browse Source
This commit is contained in:
pineappleEA
2022-10-10 02:05:13 +02:00
parent 6834d23fab
commit b16ace6ba4
14 changed files with 140 additions and 169 deletions
Download Patch File Download Diff File Expand all files Collapse all files

172
src/video_core/engines/maxwell_3d.cpp
View File

@@ -117,10 +117,12 @@ void Maxwell3D::InitializeRegisterDefaults() {
shadow_state = regs;
mme_inline[MAXWELL3D_REG_INDEX(draw.end)] = true;
mme_inline[MAXWELL3D_REG_INDEX(draw.begin)] = true;
mme_inline[MAXWELL3D_REG_INDEX(vertex_buffer.count)] = true;
mme_inline[MAXWELL3D_REG_INDEX(index_buffer.count)] = true;
inline_draw[MAXWELL3D_REG_INDEX(draw.end)] = true;
inline_draw[MAXWELL3D_REG_INDEX(draw.begin)] = true;
inline_draw[MAXWELL3D_REG_INDEX(vertex_buffer.first)] = true;
inline_draw[MAXWELL3D_REG_INDEX(vertex_buffer.count)] = true;
inline_draw[MAXWELL3D_REG_INDEX(index_buffer.first)] = true;
inline_draw[MAXWELL3D_REG_INDEX(index_buffer.count)] = true;
}
void Maxwell3D::ProcessMacro(u32 method, const u32* base_start, u32 amount, bool is_last_call) {
@@ -208,25 +210,21 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
return ProcessCBBind(3);
case MAXWELL3D_REG_INDEX(bind_groups[4].raw_config):
return ProcessCBBind(4);
case MAXWELL3D_REG_INDEX(draw.end):
return DrawArrays();
case MAXWELL3D_REG_INDEX(index_buffer32_first):
regs.index_buffer.count = regs.index_buffer32_first.count;
regs.index_buffer.first = regs.index_buffer32_first.first;
dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
return DrawArrays();
return FlushInlineDraw();
case MAXWELL3D_REG_INDEX(index_buffer16_first):
regs.index_buffer.count = regs.index_buffer16_first.count;
regs.index_buffer.first = regs.index_buffer16_first.first;
dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
return DrawArrays();
return FlushInlineDraw();
case MAXWELL3D_REG_INDEX(index_buffer8_first):
regs.index_buffer.count = regs.index_buffer8_first.count;
regs.index_buffer.first = regs.index_buffer8_first.first;
dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
// a macro calls this one over and over, should it increase instancing?
// Used by Hades and likely other Vulkan games.
return DrawArrays();
return FlushInlineDraw();
case MAXWELL3D_REG_INDEX(topology_override):
use_topology_override = true;
return;
@@ -261,14 +259,14 @@ void Maxwell3D::CallMacroMethod(u32 method, const std::vector<u32>& parameters)
// Execute the current macro.
macro_engine->Execute(macro_positions[entry], parameters);
if (mme_draw.current_mode != MMEDrawMode::Undefined) {
FlushMMEInlineDraw();
if (draw_state.current_mode != DrawMode::Undefined) {
FlushInlineDraw();
}
}
void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
// It is an error to write to a register other than the current macro's ARG register before it
// has finished execution.
// It is an error to write to a register other than the current macro's ARG register before
// it has finished execution.
if (executing_macro != 0) {
ASSERT(method == executing_macro + 1);
}
@@ -283,9 +281,39 @@ void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
ASSERT_MSG(method < Regs::NUM_REGS,
"Invalid Maxwell3D register, increase the size of the Regs structure");
const u32 argument = ProcessShadowRam(method, method_argument);
ProcessDirtyRegisters(method, argument);
ProcessMethodCall(method, argument, method_argument, is_last_call);
if (inline_draw[method]) {
regs.reg_array[method] = method_argument;
switch (method) {
case MAXWELL3D_REG_INDEX(vertex_buffer.count):
case MAXWELL3D_REG_INDEX(index_buffer.count): {
const DrawMode expected_mode = method == MAXWELL3D_REG_INDEX(vertex_buffer.count)
? DrawMode::Array
: DrawMode::Indexed;
StepInstance(expected_mode, method_argument);
break;
}
case MAXWELL3D_REG_INDEX(draw.begin):
draw_state.instance_mode =
(regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Subsequent) ||
(regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Unchanged);
draw_state.gl_begin_consume = true;
break;
case MAXWELL3D_REG_INDEX(draw.end):
draw_state.gl_end_count++;
break;
case MAXWELL3D_REG_INDEX(vertex_buffer.first):
case MAXWELL3D_REG_INDEX(index_buffer.first):
break;
}
} else {
if (draw_state.current_mode != DrawMode::Undefined) {
FlushInlineDraw();
}
const u32 argument = ProcessShadowRam(method, method_argument);
ProcessDirtyRegisters(method, argument);
ProcessMethodCall(method, argument, method_argument, is_last_call);
}
}
void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
@@ -326,55 +354,30 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
}
}
void Maxwell3D::StepInstance(const MMEDrawMode expected_mode, const u32 count) {
if (mme_draw.current_mode == MMEDrawMode::Undefined) {
if (mme_draw.gl_begin_consume) {
mme_draw.current_mode = expected_mode;
mme_draw.current_count = count;
mme_draw.instance_count = 1;
mme_draw.gl_begin_consume = false;
mme_draw.gl_end_count = 0;
void Maxwell3D::StepInstance(const DrawMode expected_mode, const u32 count) {
if (draw_state.current_mode == DrawMode::Undefined) {
if (draw_state.gl_begin_consume) {
draw_state.current_mode = expected_mode;
draw_state.current_count = count;
draw_state.instance_count = 1;
draw_state.gl_begin_consume = false;
draw_state.gl_end_count = 0;
}
return;
} else {
if (mme_draw.current_mode == expected_mode && count == mme_draw.current_count &&
mme_draw.instance_mode && mme_draw.gl_begin_consume) {
mme_draw.instance_count++;
mme_draw.gl_begin_consume = false;
if (draw_state.current_mode == expected_mode && count == draw_state.current_count &&
draw_state.instance_mode && draw_state.gl_begin_consume) {
draw_state.instance_count++;
draw_state.gl_begin_consume = false;
return;
} else {
FlushMMEInlineDraw();
FlushInlineDraw();
}
}
// Tail call in case it needs to retry.
StepInstance(expected_mode, count);
}
void Maxwell3D::CallMethodFromMME(u32 method, u32 method_argument) {
if (mme_inline[method]) {
regs.reg_array[method] = method_argument;
if (method == MAXWELL3D_REG_INDEX(vertex_buffer.count) ||
method == MAXWELL3D_REG_INDEX(index_buffer.count)) {
const MMEDrawMode expected_mode = method == MAXWELL3D_REG_INDEX(vertex_buffer.count)
? MMEDrawMode::Array
: MMEDrawMode::Indexed;
StepInstance(expected_mode, method_argument);
} else if (method == MAXWELL3D_REG_INDEX(draw.begin)) {
mme_draw.instance_mode =
(regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Subsequent) ||
(regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Unchanged);
mme_draw.gl_begin_consume = true;
} else {
mme_draw.gl_end_count++;
}
} else {
if (mme_draw.current_mode != MMEDrawMode::Undefined) {
FlushMMEInlineDraw();
}
CallMethod(method, method_argument, true);
}
}
void Maxwell3D::ProcessTopologyOverride() {
using PrimitiveTopology = Maxwell3D::Regs::PrimitiveTopology;
using PrimitiveTopologyOverride = Maxwell3D::Regs::PrimitiveTopologyOverride;
@@ -404,11 +407,12 @@ void Maxwell3D::ProcessTopologyOverride() {
}
}
void Maxwell3D::FlushMMEInlineDraw() {
void Maxwell3D::FlushInlineDraw() {
LOG_TRACE(HW_GPU, "called, topology={}, count={}", regs.draw.topology.Value(),
regs.vertex_buffer.count);
ASSERT_MSG(!(regs.index_buffer.count && regs.vertex_buffer.count), "Both indexed and direct?");
ASSERT(mme_draw.instance_count == mme_draw.gl_end_count);
ASSERT(draw_state.instance_count == draw_state.gl_end_count);
// Both instance configuration registers can not be set at the same time.
ASSERT_MSG(regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::First ||
@@ -417,9 +421,9 @@ void Maxwell3D::FlushMMEInlineDraw() {
ProcessTopologyOverride();
const bool is_indexed = mme_draw.current_mode == MMEDrawMode::Indexed;
const bool is_indexed = draw_state.current_mode == DrawMode::Indexed;
if (ShouldExecute()) {
rasterizer->Draw(is_indexed, true);
rasterizer->Draw(is_indexed);
}
// TODO(bunnei): Below, we reset vertex count so that we can use these registers to determine if
@@ -431,12 +435,12 @@ void Maxwell3D::FlushMMEInlineDraw() {
} else {
regs.vertex_buffer.count = 0;
}
mme_draw.current_mode = MMEDrawMode::Undefined;
mme_draw.current_count = 0;
mme_draw.instance_count = 0;
mme_draw.instance_mode = false;
mme_draw.gl_begin_consume = false;
mme_draw.gl_end_count = 0;
draw_state.current_mode = DrawMode::Undefined;
draw_state.current_count = 0;
draw_state.instance_count = 0;
draw_state.instance_mode = false;
draw_state.gl_begin_consume = false;
draw_state.gl_end_count = 0;
}
void Maxwell3D::ProcessMacroUpload(u32 data) {
@@ -576,42 +580,6 @@ void Maxwell3D::ProcessSyncPoint() {
}
}
void Maxwell3D::DrawArrays() {
LOG_TRACE(HW_GPU, "called, topology={}, count={}", regs.draw.topology.Value(),
regs.vertex_buffer.count);
ASSERT_MSG(!(regs.index_buffer.count && regs.vertex_buffer.count), "Both indexed and direct?");
// Both instance configuration registers can not be set at the same time.
ASSERT_MSG(regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::First ||
regs.draw.instance_id != Maxwell3D::Regs::Draw::InstanceId::Unchanged,
"Illegal combination of instancing parameters");
ProcessTopologyOverride();
if (regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Subsequent) {
// Increment the current instance *before* drawing.
state.current_instance++;
} else if (regs.draw.instance_id != Maxwell3D::Regs::Draw::InstanceId::Unchanged) {
// Reset the current instance to 0.
state.current_instance = 0;
}
const bool is_indexed{regs.index_buffer.count && !regs.vertex_buffer.count};
if (ShouldExecute()) {
rasterizer->Draw(is_indexed, false);
}
// TODO(bunnei): Below, we reset vertex count so that we can use these registers to determine if
// the game is trying to draw indexed or direct mode. This needs to be verified on HW still -
// it's possible that it is incorrect and that there is some other register used to specify the
// drawing mode.
if (is_indexed) {
regs.index_buffer.count = 0;
} else {
regs.vertex_buffer.count = 0;
}
}
std::optional<u64> Maxwell3D::GetQueryResult() {
switch (regs.report_semaphore.query.report) {
case Regs::ReportSemaphore::Report::Payload:

22
src/video_core/engines/maxwell_3d.h
View File

@@ -3050,8 +3050,6 @@ public:
};
std::array<ShaderStageInfo, Regs::MaxShaderStage> shader_stages;
u32 current_instance = 0; ///< Current instance to be used to simulate instanced rendering.
};
State state{};
@@ -3066,10 +3064,7 @@ public:
void CallMultiMethod(u32 method, const u32* base_start, u32 amount,
u32 methods_pending) override;
/// Write the value to the register identified by method.
void CallMethodFromMME(u32 method, u32 method_argument);
void FlushMMEInlineDraw();
void FlushInlineDraw();
bool ShouldExecute() const {
return execute_on;
@@ -3083,20 +3078,20 @@ public:
return *rasterizer;
}
enum class MMEDrawMode : u32 {
enum class DrawMode : u32 {
Undefined,
Array,
Indexed,
};
struct MMEDrawState {
MMEDrawMode current_mode{MMEDrawMode::Undefined};
struct DrawState {
DrawMode current_mode{DrawMode::Undefined};
u32 current_count{};
u32 instance_count{};
bool instance_mode{};
bool gl_begin_consume{};
u32 gl_end_count{};
} mme_draw;
} draw_state;
struct DirtyState {
using Flags = std::bitset<std::numeric_limits<u8>::max()>;
@@ -3166,14 +3161,11 @@ private:
/// Handles a write to the CB_BIND register.
void ProcessCBBind(size_t stage_index);
/// Handles a write to the VERTEX_END_GL register, triggering a draw.
void DrawArrays();
/// Handles use of topology overrides (e.g., to avoid using a topology assigned from a macro)
void ProcessTopologyOverride();
// Handles a instance drawcall from MME
void StepInstance(MMEDrawMode expected_mode, u32 count);
void StepInstance(DrawMode 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();
@@ -3186,7 +3178,7 @@ private:
/// Start offsets of each macro in macro_memory
std::array<u32, 0x80> macro_positions{};
std::array<bool, Regs::NUM_REGS> mme_inline{};
std::array<bool, Regs::NUM_REGS> inline_draw{};
/// Macro method that is currently being executed / being fed parameters.
u32 executing_macro = 0;