MaxwellDMA: Fixes, corrections and relaxations.

This commit fixes offsets on Linear -> Tiled copies, corrects z pos
fortiled->linear copies, corrects bytes_per_pixel calculation in tiled
-> linear copies and relaxes some limitations set by latest dma fixes
refactors.
This commit is contained in:
Fernando Sahmkow 2019-07-24 19:18:17 -04:00 committed by FernandoS27
parent 9be9600bdc
commit a452ff983d
3 changed files with 36 additions and 23 deletions

View file

@ -5,6 +5,7 @@
#include "common/assert.h" #include "common/assert.h"
#include "common/logging/log.h" #include "common/logging/log.h"
#include "core/core.h" #include "core/core.h"
#include "core/settings.h"
#include "video_core/engines/maxwell_3d.h" #include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/maxwell_dma.h" #include "video_core/engines/maxwell_dma.h"
#include "video_core/memory_manager.h" #include "video_core/memory_manager.h"
@ -84,13 +85,17 @@ void MaxwellDMA::HandleCopy() {
ASSERT(regs.exec.enable_2d == 1); ASSERT(regs.exec.enable_2d == 1);
if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) { if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) {
ASSERT(regs.src_params.size_z == 1); ASSERT(regs.src_params.BlockDepth() == 0);
// If the input is tiled and the output is linear, deswizzle the input and copy it over. // If the input is tiled and the output is linear, deswizzle the input and copy it over.
const u32 src_bytes_per_pixel = regs.src_pitch / regs.src_params.size_x; const u32 bytes_per_pixel = regs.dst_pitch / regs.x_count;
const std::size_t src_size = Texture::CalculateSize( const std::size_t src_size = Texture::CalculateSize(
true, src_bytes_per_pixel, regs.src_params.size_x, regs.src_params.size_y, true, bytes_per_pixel, regs.src_params.size_x, regs.src_params.size_y,
regs.src_params.size_z, regs.src_params.BlockHeight(), regs.src_params.BlockDepth()); regs.src_params.size_z, regs.src_params.BlockHeight(), regs.src_params.BlockDepth());
const std::size_t src_layer_size = Texture::CalculateSize(
true, bytes_per_pixel, regs.src_params.size_x, regs.src_params.size_y, 1,
regs.src_params.BlockHeight(), regs.src_params.BlockDepth());
const std::size_t dst_size = regs.dst_pitch * regs.y_count; const std::size_t dst_size = regs.dst_pitch * regs.y_count;
if (read_buffer.size() < src_size) { if (read_buffer.size() < src_size) {
@ -104,23 +109,23 @@ void MaxwellDMA::HandleCopy() {
memory_manager.ReadBlock(source, read_buffer.data(), src_size); memory_manager.ReadBlock(source, read_buffer.data(), src_size);
memory_manager.ReadBlock(dest, write_buffer.data(), dst_size); memory_manager.ReadBlock(dest, write_buffer.data(), dst_size);
Texture::UnswizzleSubrect(regs.x_count, regs.y_count, regs.dst_pitch, Texture::UnswizzleSubrect(
regs.src_params.size_x, src_bytes_per_pixel, read_buffer.data(), regs.x_count, regs.y_count, regs.dst_pitch, regs.src_params.size_x, bytes_per_pixel,
write_buffer.data(), regs.src_params.BlockHeight(), read_buffer.data() + src_layer_size * regs.src_params.pos_z, write_buffer.data(),
regs.src_params.pos_x, regs.src_params.pos_y); regs.src_params.BlockHeight(), regs.src_params.pos_x, regs.src_params.pos_y);
memory_manager.WriteBlock(dest, write_buffer.data(), dst_size); memory_manager.WriteBlock(dest, write_buffer.data(), dst_size);
} else { } else {
ASSERT(regs.dst_params.BlockDepth() == 0); ASSERT(regs.dst_params.BlockDepth() == 0);
const u32 src_bytes_per_pixel = regs.src_pitch / regs.x_count; const u32 bytes_per_pixel = regs.src_pitch / regs.x_count;
const std::size_t dst_size = Texture::CalculateSize( const std::size_t dst_size = Texture::CalculateSize(
true, src_bytes_per_pixel, regs.dst_params.size_x, regs.dst_params.size_y, true, bytes_per_pixel, regs.dst_params.size_x, regs.dst_params.size_y,
regs.dst_params.size_z, regs.dst_params.BlockHeight(), regs.dst_params.BlockDepth()); regs.dst_params.size_z, regs.dst_params.BlockHeight(), regs.dst_params.BlockDepth());
const std::size_t dst_layer_size = Texture::CalculateSize( const std::size_t dst_layer_size = Texture::CalculateSize(
true, src_bytes_per_pixel, regs.dst_params.size_x, regs.dst_params.size_y, 1, true, bytes_per_pixel, regs.dst_params.size_x, regs.dst_params.size_y, 1,
regs.dst_params.BlockHeight(), regs.dst_params.BlockDepth()); regs.dst_params.BlockHeight(), regs.dst_params.BlockDepth());
const std::size_t src_size = regs.src_pitch * regs.y_count; const std::size_t src_size = regs.src_pitch * regs.y_count;
@ -133,14 +138,19 @@ void MaxwellDMA::HandleCopy() {
write_buffer.resize(dst_size); write_buffer.resize(dst_size);
} }
memory_manager.ReadBlock(source, read_buffer.data(), src_size); if (Settings::values.use_accurate_gpu_emulation) {
memory_manager.ReadBlock(dest, write_buffer.data(), dst_size); memory_manager.ReadBlock(source, read_buffer.data(), src_size);
memory_manager.ReadBlock(dest, write_buffer.data(), dst_size);
} else {
memory_manager.ReadBlockUnsafe(source, read_buffer.data(), src_size);
memory_manager.ReadBlockUnsafe(dest, write_buffer.data(), dst_size);
}
// If the input is linear and the output is tiled, swizzle the input and copy it over. // If the input is linear and the output is tiled, swizzle the input and copy it over.
Texture::SwizzleSubrect(regs.x_count, regs.y_count, regs.src_pitch, regs.dst_params.size_x, Texture::SwizzleSubrect(
src_bytes_per_pixel, regs.x_count, regs.y_count, regs.src_pitch, regs.dst_params.size_x, bytes_per_pixel,
write_buffer.data() + dst_layer_size * regs.dst_params.pos_z, write_buffer.data() + dst_layer_size * regs.dst_params.pos_z, read_buffer.data(),
read_buffer.data(), regs.dst_params.BlockHeight()); regs.dst_params.BlockHeight(), regs.dst_params.pos_x, regs.dst_params.pos_y);
memory_manager.WriteBlock(dest, write_buffer.data(), dst_size); memory_manager.WriteBlock(dest, write_buffer.data(), dst_size);
} }

View file

@ -257,19 +257,21 @@ std::vector<u8> UnswizzleTexture(u8* address, u32 tile_size_x, u32 tile_size_y,
void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width, void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width,
u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data,
u32 block_height_bit) { u32 block_height_bit, u32 offset_x, u32 offset_y) {
const u32 block_height = 1U << block_height_bit; const u32 block_height = 1U << block_height_bit;
const u32 image_width_in_gobs{(swizzled_width * bytes_per_pixel + (gob_size_x - 1)) / const u32 image_width_in_gobs{(swizzled_width * bytes_per_pixel + (gob_size_x - 1)) /
gob_size_x}; gob_size_x};
for (u32 line = 0; line < subrect_height; ++line) { for (u32 line = 0; line < subrect_height; ++line) {
const u32 dst_y = line + offset_y;
const u32 gob_address_y = const u32 gob_address_y =
(line / (gob_size_y * block_height)) * gob_size * block_height * image_width_in_gobs + (dst_y / (gob_size_y * block_height)) * gob_size * block_height * image_width_in_gobs +
((line % (gob_size_y * block_height)) / gob_size_y) * gob_size; ((dst_y % (gob_size_y * block_height)) / gob_size_y) * gob_size;
const auto& table = legacy_swizzle_table[line % gob_size_y]; const auto& table = legacy_swizzle_table[dst_y % gob_size_y];
for (u32 x = 0; x < subrect_width; ++x) { for (u32 x = 0; x < subrect_width; ++x) {
const u32 dst_x = x + offset_x;
const u32 gob_address = const u32 gob_address =
gob_address_y + (x * bytes_per_pixel / gob_size_x) * gob_size * block_height; gob_address_y + (dst_x * bytes_per_pixel / gob_size_x) * gob_size * block_height;
const u32 swizzled_offset = gob_address + table[(x * bytes_per_pixel) % gob_size_x]; const u32 swizzled_offset = gob_address + table[(dst_x * bytes_per_pixel) % gob_size_x];
u8* source_line = unswizzled_data + line * source_pitch + x * bytes_per_pixel; u8* source_line = unswizzled_data + line * source_pitch + x * bytes_per_pixel;
u8* dest_addr = swizzled_data + swizzled_offset; u8* dest_addr = swizzled_data + swizzled_offset;

View file

@ -44,7 +44,8 @@ std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height
/// Copies an untiled subrectangle into a tiled surface. /// Copies an untiled subrectangle into a tiled surface.
void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width, void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width,
u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height); u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height,
u32 offset_x, u32 offset_y);
/// Copies a tiled subrectangle into a linear surface. /// Copies a tiled subrectangle into a linear surface.
void UnswizzleSubrect(u32 subrect_width, u32 subrect_height, u32 dest_pitch, u32 swizzled_width, void UnswizzleSubrect(u32 subrect_width, u32 subrect_height, u32 dest_pitch, u32 swizzled_width,