gpu: move MemoryFill, TextureCopy and DisplayTransfer into functions

The old code indented too much to read. Split into functions and do general cleanup.
This commit is contained in:
wwylele 2016-09-27 18:38:42 +08:00
parent d2419570b9
commit c88cdc9a2b

View file

@ -80,6 +80,234 @@ 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_DisplayTransfer, "GPU", "DisplayTransfer", MP_RGB(100, 100, 255));
MICROPROFILE_DEFINE(GPU_CmdlistProcessing, "GPU", "Cmdlist Processing", MP_RGB(100, 255, 100)); MICROPROFILE_DEFINE(GPU_CmdlistProcessing, "GPU", "Cmdlist Processing", MP_RGB(100, 255, 100));
static void MemoryFill(const Regs::MemoryFillConfig& config) {
u8* start = Memory::GetPhysicalPointer(config.GetStartAddress());
u8* end = Memory::GetPhysicalPointer(config.GetEndAddress());
// TODO: Consider always accelerating and returning vector of
// regions that the accelerated fill did not cover to
// reduce/eliminate the fill that the cpu has to do.
// This would also mean that the flush below is not needed.
// Fill should first flush all surfaces that touch but are
// not completely within the fill range.
// 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))
return;
Memory::RasterizerFlushAndInvalidateRegion(config.GetStartAddress(),
config.GetEndAddress() - config.GetStartAddress());
if (config.fill_24bit) {
// fill with 24-bit values
for (u8* ptr = start; ptr < end; ptr += 3) {
ptr[0] = config.value_24bit_r;
ptr[1] = config.value_24bit_g;
ptr[2] = config.value_24bit_b;
}
} else if (config.fill_32bit) {
// fill with 32-bit values
if (end > start) {
u32 value = config.value_32bit;
size_t len = (end - start) / sizeof(u32);
for (size_t i = 0; i < len; ++i)
memcpy(&start[i * sizeof(u32)], &value, sizeof(u32));
}
} else {
// fill with 16-bit values
u16 value_16bit = config.value_16bit.Value();
for (u8* ptr = start; ptr < end; ptr += sizeof(u16))
memcpy(ptr, &value_16bit, sizeof(u16));
}
}
static void DisplayTransfer(const Regs::DisplayTransferConfig& config) {
if (VideoCore::g_renderer->Rasterizer()->AccelerateDisplayTransfer(config))
return;
u8* src_pointer = Memory::GetPhysicalPointer(config.GetPhysicalInputAddress());
u8* dst_pointer = Memory::GetPhysicalPointer(config.GetPhysicalOutputAddress());
if (config.scaling > config.ScaleXY) {
LOG_CRITICAL(HW_GPU, "Unimplemented display transfer scaling mode %u",
config.scaling.Value());
UNIMPLEMENTED();
return;
}
if (config.input_linear && config.scaling != config.NoScale) {
LOG_CRITICAL(HW_GPU, "Scaling is only implemented on tiled input");
UNIMPLEMENTED();
return;
}
int horizontal_scale = config.scaling != config.NoScale ? 1 : 0;
int vertical_scale = config.scaling == config.ScaleXY ? 1 : 0;
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);
Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(), input_size);
Memory::RasterizerFlushAndInvalidateRegion(config.GetPhysicalOutputAddress(), output_size);
for (u32 y = 0; y < output_height; ++y) {
for (u32 x = 0; x < output_width; ++x) {
Math::Vec4<u8> src_color;
// Calculate the [x,y] position of the input image
// based on the current output position and the scale
u32 input_x = x << horizontal_scale;
u32 input_y = y << vertical_scale;
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;
}
u32 dst_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.output_format);
u32 src_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.input_format);
u32 src_offset;
u32 dst_offset;
if (config.input_linear) {
if (!config.dont_swizzle) {
// Interpret the input as linear and the output as tiled
u32 coarse_y = 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) + 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;
}
} else {
if (!config.dont_swizzle) {
// Interpret the input as tiled and the output as linear
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) +
coarse_y * stride;
dst_offset = (x + y * output_width) * dst_bytes_per_pixel;
} else {
// Both input and output are tiled
u32 out_coarse_y = 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) +
in_coarse_y * in_stride;
dst_offset = VideoCore::GetMortonOffset(x, y, dst_bytes_per_pixel) +
out_coarse_y * out_stride;
}
}
const u8* src_pixel = src_pointer + src_offset;
src_color = DecodePixel(config.input_format, src_pixel);
if (config.scaling == config.ScaleX) {
Math::Vec4<u8> pixel =
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);
src_color = (((src_color + pixel1) + (pixel2 + pixel3)) / 4).Cast<u8>();
}
u8* dst_pixel = dst_pointer + dst_offset;
switch (config.output_format) {
case Regs::PixelFormat::RGBA8:
Color::EncodeRGBA8(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB8:
Color::EncodeRGB8(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB565:
Color::EncodeRGB565(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB5A1:
Color::EncodeRGB5A1(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGBA4:
Color::EncodeRGBA4(src_color, dst_pixel);
break;
default:
LOG_ERROR(HW_GPU, "Unknown destination framebuffer format %x",
config.output_format.Value());
break;
}
}
}
}
static void TextureCopy(const Regs::DisplayTransferConfig& config) {
if (VideoCore::g_renderer->Rasterizer()->AccelerateTextureCopy(config))
return;
u8* src_pointer = Memory::GetPhysicalPointer(config.GetPhysicalInputAddress());
u8* dst_pointer = Memory::GetPhysicalPointer(config.GetPhysicalOutputAddress());
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> template <typename T>
inline void Write(u32 addr, const T data) { inline void Write(u32 addr, const T data) {
addr -= HW::VADDR_GPU; addr -= HW::VADDR_GPU;
@ -102,55 +330,14 @@ inline void Write(u32 addr, const T data) {
auto& config = g_regs.memory_fill_config[is_second_filler]; auto& config = g_regs.memory_fill_config[is_second_filler];
if (config.trigger) { if (config.trigger) {
if (config.address_start) { // Some games pass invalid values here MemoryFill(config);
u8* start = Memory::GetPhysicalPointer(config.GetStartAddress()); LOG_TRACE(HW_GPU, "MemoryFill from 0x%08x to 0x%08x", config.GetStartAddress(),
u8* end = Memory::GetPhysicalPointer(config.GetEndAddress()); config.GetEndAddress());
// TODO: Consider always accelerating and returning vector of if (!is_second_filler) {
// regions that the accelerated fill did not cover to GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC0);
// reduce/eliminate the fill that the cpu has to do. } else {
// This would also mean that the flush below is not needed. GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC1);
// Fill should first flush all surfaces that touch but are
// not completely within the fill range.
// 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)) {
Memory::RasterizerFlushAndInvalidateRegion(config.GetStartAddress(),
config.GetEndAddress() -
config.GetStartAddress());
if (config.fill_24bit) {
// fill with 24-bit values
for (u8* ptr = start; ptr < end; ptr += 3) {
ptr[0] = config.value_24bit_r;
ptr[1] = config.value_24bit_g;
ptr[2] = config.value_24bit_b;
}
} else if (config.fill_32bit) {
// fill with 32-bit values
if (end > start) {
u32 value = config.value_32bit;
size_t len = (end - start) / sizeof(u32);
for (size_t i = 0; i < len; ++i)
memcpy(&start[i * sizeof(u32)], &value, sizeof(u32));
}
} else {
// fill with 16-bit values
u16 value_16bit = config.value_16bit.Value();
for (u8* ptr = start; ptr < end; ptr += sizeof(u16))
memcpy(ptr, &value_16bit, sizeof(u16));
}
}
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);
}
} }
// Reset "trigger" flag and set the "finish" flag // Reset "trigger" flag and set the "finish" flag
@ -171,207 +358,22 @@ inline void Write(u32 addr, const T data) {
Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::IncomingDisplayTransfer, Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::IncomingDisplayTransfer,
nullptr); nullptr);
if (!VideoCore::g_renderer->Rasterizer()->AccelerateDisplayTransfer(config)) { if (config.is_texture_copy) {
u8* src_pointer = Memory::GetPhysicalPointer(config.GetPhysicalInputAddress()); TextureCopy(config);
u8* dst_pointer = Memory::GetPhysicalPointer(config.GetPhysicalOutputAddress()); LOG_TRACE(HW_GPU, "TextureCopy: 0x%X bytes from 0x%08X(%u+%u)-> "
"0x%08X(%u+%u), flags 0x%08X",
if (config.is_texture_copy) { config.texture_copy.size, config.GetPhysicalInputAddress(),
u32 input_width = config.texture_copy.input_width * 16; config.texture_copy.input_width * 16, config.texture_copy.input_gap * 16,
u32 input_gap = config.texture_copy.input_gap * 16; config.GetPhysicalOutputAddress(), config.texture_copy.output_width * 16,
u32 output_width = config.texture_copy.output_width * 16; config.texture_copy.output_gap * 16, config.flags);
u32 output_gap = config.texture_copy.output_gap * 16; } else {
DisplayTransfer(config);
size_t contiguous_input_size = LOG_TRACE(HW_GPU, "DisplayTransfer: 0x%08x(%ux%u)-> "
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;
}
}
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.scaling > config.ScaleXY) {
LOG_CRITICAL(HW_GPU, "Unimplemented display transfer scaling mode %u",
config.scaling.Value());
UNIMPLEMENTED();
break;
}
if (config.input_linear && config.scaling != config.NoScale) {
LOG_CRITICAL(HW_GPU, "Scaling is only implemented on tiled input");
UNIMPLEMENTED();
break;
}
int horizontal_scale = config.scaling != config.NoScale ? 1 : 0;
int vertical_scale = config.scaling == config.ScaleXY ? 1 : 0;
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);
Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(), input_size);
Memory::RasterizerFlushAndInvalidateRegion(config.GetPhysicalOutputAddress(),
output_size);
for (u32 y = 0; y < output_height; ++y) {
for (u32 x = 0; x < output_width; ++x) {
Math::Vec4<u8> src_color;
// Calculate the [x,y] position of the input image
// based on the current output position and the scale
u32 input_x = x << horizontal_scale;
u32 input_y = y << vertical_scale;
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;
}
u32 dst_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.output_format);
u32 src_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.input_format);
u32 src_offset;
u32 dst_offset;
if (config.input_linear) {
if (!config.dont_swizzle) {
// Interpret the input as linear and the output as tiled
u32 coarse_y = 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) +
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;
}
} else {
if (!config.dont_swizzle) {
// Interpret the input as tiled and the output as linear
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) +
coarse_y * stride;
dst_offset = (x + y * output_width) * dst_bytes_per_pixel;
} else {
// Both input and output are tiled
u32 out_coarse_y = 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) +
in_coarse_y * in_stride;
dst_offset = VideoCore::GetMortonOffset(x, y, dst_bytes_per_pixel) +
out_coarse_y * out_stride;
}
}
const u8* src_pixel = src_pointer + src_offset;
src_color = DecodePixel(config.input_format, src_pixel);
if (config.scaling == config.ScaleX) {
Math::Vec4<u8> pixel =
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);
src_color = (((src_color + pixel1) + (pixel2 + pixel3)) / 4).Cast<u8>();
}
u8* dst_pixel = dst_pointer + dst_offset;
switch (config.output_format) {
case Regs::PixelFormat::RGBA8:
Color::EncodeRGBA8(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB8:
Color::EncodeRGB8(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB565:
Color::EncodeRGB565(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB5A1:
Color::EncodeRGB5A1(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGBA4:
Color::EncodeRGBA4(src_color, dst_pixel);
break;
default:
LOG_ERROR(HW_GPU, "Unknown destination framebuffer format %x",
config.output_format.Value());
break;
}
}
}
LOG_TRACE(HW_GPU, "DisplayTriggerTransfer: 0x%08x bytes from 0x%08x(%ux%u)-> "
"0x%08x(%ux%u), dst format %x, flags 0x%08X", "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.GetPhysicalInputAddress(), config.input_width.Value(),
config.input_height.Value(), config.GetPhysicalOutputAddress(), 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; g_regs.display_transfer_config.trigger = 0;