Merge pull request from wwylele/texture-copy-fix

gpu: fix edge cases for TextureCopy
This commit is contained in:
Yuri Kunde Schlesner 2017-06-29 09:29:40 -07:00 committed by GitHub
commit 686fde7e52
2 changed files with 25 additions and 20 deletions
src/core/hw

View file

@ -5,6 +5,7 @@
#include <cstring> #include <cstring>
#include <numeric> #include <numeric>
#include <type_traits> #include <type_traits>
#include "common/alignment.h"
#include "common/color.h" #include "common/color.h"
#include "common/common_types.h" #include "common/common_types.h"
#include "common/logging/log.h" #include "common/logging/log.h"
@ -313,7 +314,7 @@ static void TextureCopy(const Regs::DisplayTransferConfig& config) {
const PAddr src_addr = config.GetPhysicalInputAddress(); const PAddr src_addr = config.GetPhysicalInputAddress();
const PAddr dst_addr = config.GetPhysicalOutputAddress(); const PAddr dst_addr = config.GetPhysicalOutputAddress();
// TODO: do hwtest with these cases // TODO: do hwtest with invalid addresses
if (!Memory::IsValidPhysicalAddress(src_addr)) { if (!Memory::IsValidPhysicalAddress(src_addr)) {
LOG_CRITICAL(HW_GPU, "invalid input address 0x%08X", src_addr); LOG_CRITICAL(HW_GPU, "invalid input address 0x%08X", src_addr);
return; return;
@ -324,32 +325,37 @@ static void TextureCopy(const Regs::DisplayTransferConfig& config) {
return; 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)) if (VideoCore::g_renderer->Rasterizer()->AccelerateTextureCopy(config))
return; return;
u8* src_pointer = Memory::GetPhysicalPointer(src_addr); u8* src_pointer = Memory::GetPhysicalPointer(src_addr);
u8* dst_pointer = Memory::GetPhysicalPointer(dst_addr); u8* dst_pointer = Memory::GetPhysicalPointer(dst_addr);
u32 input_width = config.texture_copy.input_width * 16; 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 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; 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 = size_t contiguous_input_size =
config.texture_copy.size / input_width * (input_width + input_gap); config.texture_copy.size / input_width * (input_width + input_gap);
Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(), Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(),
@ -360,7 +366,6 @@ static void TextureCopy(const Regs::DisplayTransferConfig& config) {
Memory::RasterizerFlushAndInvalidateRegion(config.GetPhysicalOutputAddress(), Memory::RasterizerFlushAndInvalidateRegion(config.GetPhysicalOutputAddress(),
static_cast<u32>(contiguous_output_size)); static_cast<u32>(contiguous_output_size));
u32 remaining_size = config.texture_copy.size;
u32 remaining_input = input_width; u32 remaining_input = input_width;
u32 remaining_output = output_width; u32 remaining_output = output_width;
while (remaining_size > 0) { while (remaining_size > 0) {

View file

@ -225,7 +225,7 @@ struct Regs {
INSERT_PADDING_WORDS(0x1); INSERT_PADDING_WORDS(0x1);
struct { struct {
u32 size; u32 size; // The lower 4 bits are ignored
union { union {
u32 input_size; u32 input_size;