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;
}
if (!Memory::IsValidPhysicalAddress(dst_addr)) {
LOG_CRITICAL(HW_GPU, "invalid output address 0x%08X", dst_addr);
return;
}
if (config.input_width == 0) {
LOG_CRITICAL(HW_GPU, "zero input width");
return;
}
if (config.input_height == 0) {
LOG_CRITICAL(HW_GPU, "zero input height");
return;
}
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();
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;
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.
output_y = output_height - y - 1;
} else {
output_y = y;
}
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 = 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, 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 + output_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 + output_y * output_width) * dst_bytes_per_pixel;
} else {
// Both input and output are tiled
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) +
in_coarse_y * in_stride;
dst_offset = VideoCore::GetMortonOffset(x, output_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) {
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;
}
}
}
static void TextureCopy(const Regs::DisplayTransferConfig& config) {
const PAddr src_addr = config.GetPhysicalInputAddress();
const PAddr dst_addr = config.GetPhysicalOutputAddress();
// TODO: do hwtest with invalid addresses
if (!Memory::IsValidPhysicalAddress(src_addr)) {
LOG_CRITICAL(HW_GPU, "invalid input address {:#010X}", src_addr);
return;
}
if (!Memory::IsValidPhysicalAddress(dst_addr)) {
LOG_CRITICAL(HW_GPU, "invalid output address {:#010X}", dst_addr);
return;
}
if (VideoCore::g_renderer->Rasterizer()->AccelerateTextureCopy(config))
return;
u8* src_pointer = Memory::GetPhysicalPointer(src_addr);
u8* dst_pointer = Memory::GetPhysicalPointer(dst_addr);
u32 remaining_size = Common::AlignDown(config.texture_copy.size, 16);
if (remaining_size == 0) {
LOG_CRITICAL(HW_GPU, "zero size. Real hardware freezes on this.");
return;
}
u32 input_gap = config.texture_copy.input_gap * 16;
u32 output_gap = config.texture_copy.output_gap * 16;
// Zero gap means contiguous input/output even if width = 0. To avoid infinite loop below, width
// is assigned with the total size if gap = 0.
u32 input_width = input_gap == 0 ? remaining_size : config.texture_copy.input_width * 16;
u32 output_width = output_gap == 0 ? remaining_size : config.texture_copy.output_width * 16;
if (input_width == 0) {
LOG_CRITICAL(HW_GPU, "zero input width. Real hardware freezes on this.");
return;
}
if (output_width == 0) {
LOG_CRITICAL(HW_GPU, "zero output width. Real hardware freezes on this.");
return;
}
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);
// Only need to flush output if it has a gap
const auto FlushInvalidate_fn = (output_gap != 0) ? Memory::RasterizerFlushAndInvalidateRegion
: Memory::RasterizerInvalidateRegion;
FlushInvalidate_fn(config.GetPhysicalOutputAddress(), static_cast<u32>(contiguous_output_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;
}
}
}
