aboutsummaryrefslogtreecommitdiff
path: root/src/core/hw/gpu.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/hw/gpu.cpp')
-rw-r--r--src/core/hw/gpu.cpp178
1 files changed, 121 insertions, 57 deletions
diff --git a/src/core/hw/gpu.cpp b/src/core/hw/gpu.cpp
index aad0e5d0d..424ce2ca7 100644
--- a/src/core/hw/gpu.cpp
+++ b/src/core/hw/gpu.cpp
@@ -18,8 +18,9 @@
#include "core/hw/gpu.h"
#include "video_core/command_processor.h"
+#include "video_core/utils.h"
#include "video_core/video_core.h"
-
+#include <video_core/color.h>
namespace GPU {
@@ -67,23 +68,38 @@ inline void Write(u32 addr, const T data) {
switch (index) {
// Memory fills are triggered once the fill value is written.
- // NOTE: This is not verified.
- case GPU_REG_INDEX_WORKAROUND(memory_fill_config[0].value, 0x00004 + 0x3):
- case GPU_REG_INDEX_WORKAROUND(memory_fill_config[1].value, 0x00008 + 0x3):
+ 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].value));
- const auto& config = g_regs.memory_fill_config[is_second_filler];
-
- // TODO: Not sure if this check should be done at GSP level instead
- if (config.address_start) {
- // TODO: Not sure if this algorithm is correct, particularly because it doesn't use the size member at all
- u32* start = (u32*)Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetStartAddress()));
- u32* end = (u32*)Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetEndAddress()));
- for (u32* ptr = start; ptr < end; ++ptr)
- *ptr = bswap32(config.value); // TODO: This is just a workaround to missing framebuffer format emulation
+ 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.address_start && config.trigger) {
+ u8* start = Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetStartAddress()));
+ u8* end = Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetEndAddress()));
+
+ if (config.fill_24bit) {
+ // fill with 24-bit values
+ for (u8* ptr = start; ptr < end; ptr += 3) {
+ ptr[0] = config.value_24bit_b;
+ ptr[1] = config.value_24bit_g;
+ ptr[2] = config.value_24bit_r;
+ }
+ } else if (config.fill_32bit) {
+ // fill with 32-bit values
+ for (u32* ptr = (u32*)start; ptr < (u32*)end; ++ptr)
+ *ptr = config.value_32bit;
+ } else {
+ // fill with 16-bit values
+ for (u16* ptr = (u16*)start; ptr < (u16*)end; ++ptr)
+ *ptr = config.value_16bit;
+ }
LOG_TRACE(HW_GPU, "MemoryFill from 0x%08x to 0x%08x", config.GetStartAddress(), config.GetEndAddress());
+ config.trigger = 0;
+ config.finished = 1;
+
if (!is_second_filler) {
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC0);
} else {
@@ -97,61 +113,109 @@ inline void Write(u32 addr, const T data) {
{
const auto& config = g_regs.display_transfer_config;
if (config.trigger & 1) {
- u8* source_pointer = Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetPhysicalInputAddress()));
- u8* dest_pointer = Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetPhysicalOutputAddress()));
-
- // Cheap emulation of horizontal scaling: Just skip each second pixel of the
- // input framebuffer. We keep track of this in the pixel_skip variable.
- unsigned pixel_skip = (config.scale_horizontally != 0) ? 2 : 1;
-
- u32 output_width = config.output_width / pixel_skip;
-
- for (u32 y = 0; y < config.output_height; ++y) {
- // TODO: Why does the register seem to hold twice the framebuffer width?
+ u8* src_pointer = Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetPhysicalInputAddress()));
+ u8* dst_pointer = Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetPhysicalOutputAddress()));
+
+ unsigned horizontal_scale = (config.scale_horizontally != 0) ? 2 : 1;
+ unsigned vertical_scale = (config.scale_vertically != 0) ? 2 : 1;
+
+ u32 output_width = config.output_width / horizontal_scale;
+ u32 output_height = config.output_height / vertical_scale;
+
+ if (config.raw_copy) {
+ // Raw copies do not perform color conversion nor tiled->linear / linear->tiled conversions
+ // TODO(Subv): Verify if raw copies perform scaling
+ memcpy(dst_pointer, src_pointer, config.output_width * config.output_height *
+ GPU::Regs::BytesPerPixel(config.output_format));
+
+ LOG_TRACE(HW_GPU, "DisplayTriggerTransfer: 0x%08x bytes from 0x%08x(%ux%u)-> 0x%08x(%ux%u), flags 0x%08X, Raw copy",
+ config.output_height * output_width * GPU::Regs::BytesPerPixel(config.output_format),
+ config.GetPhysicalInputAddress(), config.input_width.Value(), config.input_height.Value(),
+ config.GetPhysicalOutputAddress(), config.output_width.Value(), config.output_height.Value(),
+ config.output_format.Value(), config.flags);
+
+ GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PPF);
+ break;
+ }
+ // TODO(Subv): Blend the pixels when horizontal / vertical scaling is enabled,
+ // right now we're just skipping the extra pixels.
+ for (u32 y = 0; y < output_height; ++y) {
for (u32 x = 0; x < output_width; ++x) {
- struct {
- int r, g, b, a;
- } source_color = { 0, 0, 0, 0 };
+ Math::Vec4<u8> src_color = { 0, 0, 0, 0 };
+
+ u32 scaled_x = x * horizontal_scale;
+ u32 scaled_y = y * vertical_scale;
+
+ 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.output_tiled) {
+ // 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 = (scaled_x + scaled_y * config.input_width) * src_bytes_per_pixel;
+ dst_offset = VideoCore::GetMortonOffset(x, y, dst_bytes_per_pixel) + coarse_y * stride;
+ } else {
+ // Interpret the input as tiled and the output as linear
+ u32 coarse_y = scaled_y & ~7;
+ u32 stride = config.input_width * src_bytes_per_pixel;
+
+ src_offset = VideoCore::GetMortonOffset(scaled_x, scaled_y, src_bytes_per_pixel) + coarse_y * stride;
+ dst_offset = (x + y * output_width) * dst_bytes_per_pixel;
+ }
+ const u8* src_pixel = src_pointer + src_offset;
switch (config.input_format) {
case Regs::PixelFormat::RGBA8:
- {
- // TODO: Most likely got the component order messed up.
- u8* srcptr = source_pointer + (x * pixel_skip + y * config.input_width) * 4;
- source_color.r = srcptr[0]; // blue
- source_color.g = srcptr[1]; // green
- source_color.b = srcptr[2]; // red
- source_color.a = srcptr[3]; // alpha
+ src_color = Color::DecodeRGBA8(src_pixel);
+ break;
+
+ case Regs::PixelFormat::RGB8:
+ src_color = Color::DecodeRGB8(src_pixel);
+ break;
+
+ case Regs::PixelFormat::RGB565:
+ src_color = Color::DecodeRGB565(src_pixel);
+ break;
+
+ case Regs::PixelFormat::RGB5A1:
+ src_color = Color::DecodeRGB5A1(src_pixel);
+ break;
+
+ case Regs::PixelFormat::RGBA4:
+ src_color = Color::DecodeRGBA4(src_pixel);
break;
- }
default:
LOG_ERROR(HW_GPU, "Unknown source framebuffer format %x", config.input_format.Value());
break;
}
+ u8* dst_pixel = dst_pointer + dst_offset;
switch (config.output_format) {
- /*case Regs::PixelFormat::RGBA8:
- {
- // TODO: Untested
- u8* dstptr = (u32*)(dest_pointer + x * 4 + y * config.output_width * 4);
- dstptr[0] = source_color.r;
- dstptr[1] = source_color.g;
- dstptr[2] = source_color.b;
- dstptr[3] = source_color.a;
+ case Regs::PixelFormat::RGBA8:
+ Color::EncodeRGBA8(src_color, dst_pixel);
break;
- }*/
case Regs::PixelFormat::RGB8:
- {
- // TODO: Most likely got the component order messed up.
- u8* dstptr = dest_pointer + (x + y * output_width) * 3;
- dstptr[0] = source_color.r; // blue
- dstptr[1] = source_color.g; // green
- dstptr[2] = source_color.b; // red
+ 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());
@@ -160,11 +224,11 @@ inline void Write(u32 addr, const T data) {
}
}
- LOG_TRACE(HW_GPU, "DisplayTriggerTransfer: 0x%08x bytes from 0x%08x(%ux%u)-> 0x%08x(%ux%u), dst format %x",
- config.output_height * output_width * 4,
- config.GetPhysicalInputAddress(), (u32)config.input_width, (u32)config.input_height,
- config.GetPhysicalOutputAddress(), (u32)output_width, (u32)config.output_height,
- config.output_format.Value());
+ LOG_TRACE(HW_GPU, "DisplayTriggerTransfer: 0x%08x bytes from 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);
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PPF);
}
generated by cgit v1.2.3 (git 2.25.1) at 2025-11-12 00:28:27 +0000