Cool_Tools/citra-mk7
1
0
Fork 0
mirror of https://github.com/PabloMK7/citra.git synced 2024-11-17 04:17:15 +01:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #2103 from wwylele/gpu-reg-cleanup

Browse Source 
GPU: DisplayTransfer & MemoryFill cleanup and param check
This commit is contained in:
bunnei 2016-10-03 20:21:55 -04:00 committed by GitHub
commit 49b10339bf
5 changed files with 347 additions and 247 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

355
src/core/hw/gpu.cpp
View File

@ -80,31 +80,28 @@ static Math::Vec4<u8> DecodePixel(Regs::PixelFormat input_format, const u8* src_
MICROPROFILE_DEFINE(GPU_DisplayTransfer, "GPU", "DisplayTransfer", MP_RGB(100, 100, 255));
MICROPROFILE_DEFINE(GPU_CmdlistProcessing, "GPU", "Cmdlist Processing", MP_RGB(100, 255, 100));
template <typename T>
inline void Write(u32 addr, const T data) {
addr -= HW::VADDR_GPU;
u32 index = addr / 4;
static void MemoryFill(const Regs::MemoryFillConfig& config) {
const PAddr start_addr = config.GetStartAddress();
const PAddr end_addr = config.GetEndAddress();
// Writes other than u32 are untested, so I'd rather have them abort than silently fail
if (index >= Regs::NumIds() || !std::is_same<T, u32>::value) {
LOG_ERROR(HW_GPU, "unknown Write%lu 0x%08X @ 0x%08X", sizeof(data) * 8, (u32)data, addr);
// TODO: do hwtest with these cases
if (!Memory::IsValidPhysicalAddress(start_addr)) {
LOG_CRITICAL(HW_GPU, "invalid start address 0x%08X", start_addr);
return;
}
g_regs[index] = static_cast<u32>(data);
if (!Memory::IsValidPhysicalAddress(end_addr)) {
LOG_CRITICAL(HW_GPU, "invalid end address 0x%08X", end_addr);
return;
}
switch (index) {
if (end_addr <= start_addr) {
LOG_CRITICAL(HW_GPU, "invalid memory range from 0x%08X to 0x%08X", start_addr, end_addr);
return;
}
// Memory fills are triggered once the fill value is written.
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[0].trigger, 0x00004 + 0x3):
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[1].trigger, 0x00008 + 0x3): {
const bool is_second_filler = (index != GPU_REG_INDEX(memory_fill_config[0].trigger));
auto& config = g_regs.memory_fill_config[is_second_filler];
if (config.trigger) {
if (config.address_start) { // Some games pass invalid values here
u8* start = Memory::GetPhysicalPointer(config.GetStartAddress());
u8* end = Memory::GetPhysicalPointer(config.GetEndAddress());
u8* start = Memory::GetPhysicalPointer(start_addr);
u8* end = Memory::GetPhysicalPointer(end_addr);
// TODO: Consider always accelerating and returning vector of
// regions that the accelerated fill did not cover to
@ -115,10 +112,11 @@ inline void Write(u32 addr, const T data) {
// Then fill all completely covered surfaces, and return the
// regions that were between surfaces or within the touching
// ones for cpu to manually fill here.
if (!VideoCore::g_renderer->Rasterizer()->AccelerateFill(config)) {
if (VideoCore::g_renderer->Rasterizer()->AccelerateFill(config))
return;
Memory::RasterizerFlushAndInvalidateRegion(config.GetStartAddress(),
config.GetEndAddress() -
config.GetStartAddress());
config.GetEndAddress() - config.GetStartAddress());
if (config.fill_24bit) {
// fill with 24-bit values
@ -141,104 +139,60 @@ inline void Write(u32 addr, const T data) {
for (u8* ptr = start; ptr < end; ptr += sizeof(u16))
memcpy(ptr, &value_16bit, sizeof(u16));
}
}
static void DisplayTransfer(const Regs::DisplayTransferConfig& config) {
const PAddr src_addr = config.GetPhysicalInputAddress();
const PAddr dst_addr = config.GetPhysicalOutputAddress();
// TODO: do hwtest with these cases
if (!Memory::IsValidPhysicalAddress(src_addr)) {
LOG_CRITICAL(HW_GPU, "invalid input address 0x%08X", src_addr);
return;
}
LOG_TRACE(HW_GPU, "MemoryFill from 0x%08x to 0x%08x", config.GetStartAddress(),
config.GetEndAddress());
if (!is_second_filler) {
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC0);
} else {
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC1);
}
if (!Memory::IsValidPhysicalAddress(dst_addr)) {
LOG_CRITICAL(HW_GPU, "invalid output address 0x%08X", dst_addr);
return;
}
// Reset "trigger" flag and set the "finish" flag
// NOTE: This was confirmed to happen on hardware even if "address_start" is zero.
config.trigger.Assign(0);
config.finished.Assign(1);
}
break;
if (config.input_width == 0) {
LOG_CRITICAL(HW_GPU, "zero input width");
return;
}
case GPU_REG_INDEX(display_transfer_config.trigger): {
MICROPROFILE_SCOPE(GPU_DisplayTransfer);
const auto& config = g_regs.display_transfer_config;
if (config.trigger & 1) {
if (Pica::g_debug_context)
Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::IncomingDisplayTransfer,
nullptr);
if (!VideoCore::g_renderer->Rasterizer()->AccelerateDisplayTransfer(config)) {
u8* src_pointer = Memory::GetPhysicalPointer(config.GetPhysicalInputAddress());
u8* dst_pointer = Memory::GetPhysicalPointer(config.GetPhysicalOutputAddress());
if (config.is_texture_copy) {
u32 input_width = config.texture_copy.input_width * 16;
u32 input_gap = config.texture_copy.input_gap * 16;
u32 output_width = config.texture_copy.output_width * 16;
u32 output_gap = config.texture_copy.output_gap * 16;
size_t contiguous_input_size =
config.texture_copy.size / input_width * (input_width + input_gap);
Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(),
static_cast<u32>(contiguous_input_size));
size_t contiguous_output_size =
config.texture_copy.size / output_width * (output_width + output_gap);
Memory::RasterizerFlushAndInvalidateRegion(
config.GetPhysicalOutputAddress(),
static_cast<u32>(contiguous_output_size));
u32 remaining_size = config.texture_copy.size;
u32 remaining_input = input_width;
u32 remaining_output = output_width;
while (remaining_size > 0) {
u32 copy_size =
std::min({remaining_input, remaining_output, remaining_size});
std::memcpy(dst_pointer, src_pointer, copy_size);
src_pointer += copy_size;
dst_pointer += copy_size;
remaining_input -= copy_size;
remaining_output -= copy_size;
remaining_size -= copy_size;
if (remaining_input == 0) {
remaining_input = input_width;
src_pointer += input_gap;
}
if (remaining_output == 0) {
remaining_output = output_width;
dst_pointer += output_gap;
}
if (config.input_height == 0) {
LOG_CRITICAL(HW_GPU, "zero input height");
return;
}
LOG_TRACE(
HW_GPU,
"TextureCopy: 0x%X bytes from 0x%08X(%u+%u)-> 0x%08X(%u+%u), flags 0x%08X",
config.texture_copy.size, config.GetPhysicalInputAddress(), input_width,
input_gap, config.GetPhysicalOutputAddress(), output_width, output_gap,
config.flags);
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PPF);
break;
if (config.output_width == 0) {
LOG_CRITICAL(HW_GPU, "zero output width");
return;
}
if (config.output_height == 0) {
LOG_CRITICAL(HW_GPU, "zero output height");
return;
}
if (VideoCore::g_renderer->Rasterizer()->AccelerateDisplayTransfer(config))
return;
u8* src_pointer = Memory::GetPhysicalPointer(src_addr);
u8* dst_pointer = Memory::GetPhysicalPointer(dst_addr);
if (config.scaling > config.ScaleXY) {
LOG_CRITICAL(HW_GPU, "Unimplemented display transfer scaling mode %u",
config.scaling.Value());
UNIMPLEMENTED();
break;
return;
}
if (config.input_linear && config.scaling != config.NoScale) {
LOG_CRITICAL(HW_GPU, "Scaling is only implemented on tiled input");
UNIMPLEMENTED();
break;
return;
}
int horizontal_scale = config.scaling != config.NoScale ? 1 : 0;
@ -247,14 +201,12 @@ inline void Write(u32 addr, const T data) {
u32 output_width = config.output_width >> horizontal_scale;
u32 output_height = config.output_height >> vertical_scale;
u32 input_size = config.input_width * config.input_height *
GPU::Regs::BytesPerPixel(config.input_format);
u32 output_size =
output_width * output_height * GPU::Regs::BytesPerPixel(config.output_format);
u32 input_size =
config.input_width * config.input_height * GPU::Regs::BytesPerPixel(config.input_format);
u32 output_size = output_width * output_height * GPU::Regs::BytesPerPixel(config.output_format);
Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(), input_size);
Memory::RasterizerFlushAndInvalidateRegion(config.GetPhysicalOutputAddress(),
output_size);
Memory::RasterizerFlushAndInvalidateRegion(config.GetPhysicalOutputAddress(), output_size);
for (u32 y = 0; y < output_height; ++y) {
for (u32 x = 0; x < output_width; ++x) {
@ -265,11 +217,14 @@ inline void Write(u32 addr, const T data) {
u32 input_x = x << horizontal_scale;
u32 input_y = y << vertical_scale;
u32 output_y;
if (config.flip_vertically) {
// Flip the y value of the output data,
// we do this after calculating the [x,y] position of the input image
// to account for the scaling options.
y = output_height - y - 1;
output_y = output_height - y - 1;
} else {
output_y = y;
}
u32 dst_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.output_format);
@ -280,18 +235,16 @@ inline void Write(u32 addr, const T data) {
if (config.input_linear) {
if (!config.dont_swizzle) {
// Interpret the input as linear and the output as tiled
u32 coarse_y = y & ~7;
u32 coarse_y = output_y & ~7;
u32 stride = output_width * dst_bytes_per_pixel;
src_offset =
(input_x + input_y * config.input_width) * src_bytes_per_pixel;
dst_offset = VideoCore::GetMortonOffset(x, y, dst_bytes_per_pixel) +
src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
dst_offset = VideoCore::GetMortonOffset(x, output_y, dst_bytes_per_pixel) +
coarse_y * stride;
} else {
// Both input and output are linear
src_offset =
(input_x + input_y * config.input_width) * src_bytes_per_pixel;
dst_offset = (x + y * output_width) * dst_bytes_per_pixel;
src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
dst_offset = (x + output_y * output_width) * dst_bytes_per_pixel;
}
} else {
if (!config.dont_swizzle) {
@ -299,22 +252,20 @@ inline void Write(u32 addr, const T data) {
u32 coarse_y = input_y & ~7;
u32 stride = config.input_width * src_bytes_per_pixel;
src_offset = VideoCore::GetMortonOffset(input_x, input_y,
src_bytes_per_pixel) +
src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) +
coarse_y * stride;
dst_offset = (x + y * output_width) * dst_bytes_per_pixel;
dst_offset = (x + output_y * output_width) * dst_bytes_per_pixel;
} else {
// Both input and output are tiled
u32 out_coarse_y = y & ~7;
u32 out_coarse_y = output_y & ~7;
u32 out_stride = output_width * dst_bytes_per_pixel;
u32 in_coarse_y = input_y & ~7;
u32 in_stride = config.input_width * src_bytes_per_pixel;
src_offset = VideoCore::GetMortonOffset(input_x, input_y,
src_bytes_per_pixel) +
src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) +
in_coarse_y * in_stride;
dst_offset = VideoCore::GetMortonOffset(x, y, dst_bytes_per_pixel) +
dst_offset = VideoCore::GetMortonOffset(x, output_y, dst_bytes_per_pixel) +
out_coarse_y * out_stride;
}
}
@ -326,12 +277,12 @@ inline void Write(u32 addr, const T data) {
DecodePixel(config.input_format, src_pixel + src_bytes_per_pixel);
src_color = ((src_color + pixel) / 2).Cast<u8>();
} else if (config.scaling == config.ScaleXY) {
Math::Vec4<u8> pixel1 = DecodePixel(
config.input_format, src_pixel + 1 * src_bytes_per_pixel);
Math::Vec4<u8> pixel2 = DecodePixel(
config.input_format, src_pixel + 2 * src_bytes_per_pixel);
Math::Vec4<u8> pixel3 = DecodePixel(
config.input_format, src_pixel + 3 * src_bytes_per_pixel);
Math::Vec4<u8> pixel1 =
DecodePixel(config.input_format, src_pixel + 1 * src_bytes_per_pixel);
Math::Vec4<u8> pixel2 =
DecodePixel(config.input_format, src_pixel + 2 * src_bytes_per_pixel);
Math::Vec4<u8> pixel3 =
DecodePixel(config.input_format, src_pixel + 3 * src_bytes_per_pixel);
src_color = (((src_color + pixel1) + (pixel2 + pixel3)) / 4).Cast<u8>();
}
@ -364,14 +315,154 @@ inline void Write(u32 addr, const T data) {
}
}
}
}
LOG_TRACE(HW_GPU, "DisplayTriggerTransfer: 0x%08x bytes from 0x%08x(%ux%u)-> "
static void TextureCopy(const Regs::DisplayTransferConfig& config) {
const PAddr src_addr = config.GetPhysicalInputAddress();
const PAddr dst_addr = config.GetPhysicalOutputAddress();
// TODO: do hwtest with these cases
if (!Memory::IsValidPhysicalAddress(src_addr)) {
LOG_CRITICAL(HW_GPU, "invalid input address 0x%08X", src_addr);
return;
}
if (!Memory::IsValidPhysicalAddress(dst_addr)) {
LOG_CRITICAL(HW_GPU, "invalid output address 0x%08X", dst_addr);
return;
}
if (config.texture_copy.input_width == 0) {
LOG_CRITICAL(HW_GPU, "zero input width");
return;
}
if (config.texture_copy.output_width == 0) {
LOG_CRITICAL(HW_GPU, "zero output width");
return;
}
if (config.texture_copy.size == 0) {
LOG_CRITICAL(HW_GPU, "zero size");
return;
}
if (VideoCore::g_renderer->Rasterizer()->AccelerateTextureCopy(config))
return;
u8* src_pointer = Memory::GetPhysicalPointer(src_addr);
u8* dst_pointer = Memory::GetPhysicalPointer(dst_addr);
u32 input_width = config.texture_copy.input_width * 16;
u32 input_gap = config.texture_copy.input_gap * 16;
u32 output_width = config.texture_copy.output_width * 16;
u32 output_gap = config.texture_copy.output_gap * 16;
size_t contiguous_input_size =
config.texture_copy.size / input_width * (input_width + input_gap);
Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(),
static_cast<u32>(contiguous_input_size));
size_t contiguous_output_size =
config.texture_copy.size / output_width * (output_width + output_gap);
Memory::RasterizerFlushAndInvalidateRegion(config.GetPhysicalOutputAddress(),
static_cast<u32>(contiguous_output_size));
u32 remaining_size = config.texture_copy.size;
u32 remaining_input = input_width;
u32 remaining_output = output_width;
while (remaining_size > 0) {
u32 copy_size = std::min({remaining_input, remaining_output, remaining_size});
std::memcpy(dst_pointer, src_pointer, copy_size);
src_pointer += copy_size;
dst_pointer += copy_size;
remaining_input -= copy_size;
remaining_output -= copy_size;
remaining_size -= copy_size;
if (remaining_input == 0) {
remaining_input = input_width;
src_pointer += input_gap;
}
if (remaining_output == 0) {
remaining_output = output_width;
dst_pointer += output_gap;
}
}
}
template <typename T>
inline void Write(u32 addr, const T data) {
addr -= HW::VADDR_GPU;
u32 index = addr / 4;
// Writes other than u32 are untested, so I'd rather have them abort than silently fail
if (index >= Regs::NumIds() || !std::is_same<T, u32>::value) {
LOG_ERROR(HW_GPU, "unknown Write%lu 0x%08X @ 0x%08X", sizeof(data) * 8, (u32)data, addr);
return;
}
g_regs[index] = static_cast<u32>(data);
switch (index) {
// Memory fills are triggered once the fill value is written.
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[0].trigger, 0x00004 + 0x3):
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[1].trigger, 0x00008 + 0x3): {
const bool is_second_filler = (index != GPU_REG_INDEX(memory_fill_config[0].trigger));
auto& config = g_regs.memory_fill_config[is_second_filler];
if (config.trigger) {
MemoryFill(config);
LOG_TRACE(HW_GPU, "MemoryFill from 0x%08x to 0x%08x", config.GetStartAddress(),
config.GetEndAddress());
// It seems that it won't signal interrupt if "address_start" is zero.
// TODO: hwtest this
if (config.GetStartAddress() != 0) {
if (!is_second_filler) {
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC0);
} else {
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC1);
}
}
// Reset "trigger" flag and set the "finish" flag
// NOTE: This was confirmed to happen on hardware even if "address_start" is zero.
config.trigger.Assign(0);
config.finished.Assign(1);
}
break;
}
case GPU_REG_INDEX(display_transfer_config.trigger): {
MICROPROFILE_SCOPE(GPU_DisplayTransfer);
const auto& config = g_regs.display_transfer_config;
if (config.trigger & 1) {
if (Pica::g_debug_context)
Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::IncomingDisplayTransfer,
nullptr);
if (config.is_texture_copy) {
TextureCopy(config);
LOG_TRACE(HW_GPU, "TextureCopy: 0x%X bytes from 0x%08X(%u+%u)-> "
"0x%08X(%u+%u), flags 0x%08X",
config.texture_copy.size, config.GetPhysicalInputAddress(),
config.texture_copy.input_width * 16, config.texture_copy.input_gap * 16,
config.GetPhysicalOutputAddress(), config.texture_copy.output_width * 16,
config.texture_copy.output_gap * 16, config.flags);
} else {
DisplayTransfer(config);
LOG_TRACE(HW_GPU, "DisplayTransfer: 0x%08x(%ux%u)-> "
"0x%08x(%ux%u), dst format %x, flags 0x%08X",
config.output_height * output_width *
GPU::Regs::BytesPerPixel(config.output_format),
config.GetPhysicalInputAddress(), config.input_width.Value(),
config.input_height.Value(), config.GetPhysicalOutputAddress(),
output_width, output_height, config.output_format.Value(), config.flags);
config.output_width.Value(), config.output_height.Value(),
config.output_format.Value(), config.flags);
}
g_regs.display_transfer_config.trigger = 0;

3
src/core/memory.cpp
View File

@ -251,6 +251,9 @@ bool IsValidVirtualAddress(const VAddr vaddr) {
if (page_pointer)
return true;
if (current_page_table->attributes[vaddr >> PAGE_BITS] == PageType::RasterizerCachedMemory)
return true;
if (current_page_table->attributes[vaddr >> PAGE_BITS] != PageType::Special)
return false;

7
src/video_core/rasterizer_interface.h
View File

@ -42,11 +42,16 @@ public:
/// and invalidated
virtual void FlushAndInvalidateRegion(PAddr addr, u32 size) = 0;
/// Attempt to use a faster method to perform a display transfer
/// Attempt to use a faster method to perform a display transfer with is_texture_copy = 0
virtual bool AccelerateDisplayTransfer(const GPU::Regs::DisplayTransferConfig& config) {
return false;
}
/// Attempt to use a faster method to perform a display transfer with is_texture_copy = 1
virtual bool AccelerateTextureCopy(const GPU::Regs::DisplayTransferConfig& config) {
return false;
}
/// Attempt to use a faster method to fill a region
virtual bool AccelerateFill(const GPU::Regs::MemoryFillConfig& config) {
return false;

10
src/video_core/renderer_opengl/gl_rasterizer.cpp
View File

@ -709,11 +709,6 @@ bool RasterizerOpenGL::AccelerateDisplayTransfer(const GPU::Regs::DisplayTransfe
using PixelFormat = CachedSurface::PixelFormat;
using SurfaceType = CachedSurface::SurfaceType;
if (config.is_texture_copy) {
// TODO(tfarley): Try to hardware accelerate this
return false;
}
CachedSurface src_params;
src_params.addr = config.GetPhysicalInputAddress();
src_params.width = config.output_width;
@ -768,6 +763,11 @@ bool RasterizerOpenGL::AccelerateDisplayTransfer(const GPU::Regs::DisplayTransfe
return true;
}
bool RasterizerOpenGL::AccelerateTextureCopy(const GPU::Regs::DisplayTransferConfig& config) {
// TODO(tfarley): Try to hardware accelerate this
return false;
}
bool RasterizerOpenGL::AccelerateFill(const GPU::Regs::MemoryFillConfig& config) {
using PixelFormat = CachedSurface::PixelFormat;
using SurfaceType = CachedSurface::SurfaceType;

1
src/video_core/renderer_opengl/gl_rasterizer.h
View File

@ -238,6 +238,7 @@ public:
void FlushRegion(PAddr addr, u32 size) override;
void FlushAndInvalidateRegion(PAddr addr, u32 size) override;
bool AccelerateDisplayTransfer(const GPU::Regs::DisplayTransferConfig& config) override;
bool AccelerateTextureCopy(const GPU::Regs::DisplayTransferConfig& config) override;
bool AccelerateFill(const GPU::Regs::MemoryFillConfig& config) override;
bool AccelerateDisplay(const GPU::Regs::FramebufferConfig& config, PAddr framebuffer_addr,
u32 pixel_stride, ScreenInfo& screen_info) override;