From bf16c73260787f80de2a0bc13b11737471695d0f Mon Sep 17 00:00:00 2001 From: wwylele Date: Tue, 27 Jun 2017 23:20:13 +0300 Subject: gpu: fix edge cases for TextureCopy --- src/core/hw/gpu.cpp | 41 +++++++++++++++++++++++------------------ 1 file changed, 23 insertions(+), 18 deletions(-) (limited to 'src') diff --git a/src/core/hw/gpu.cpp b/src/core/hw/gpu.cpp index 42809c731..a0a523822 100644 --- a/src/core/hw/gpu.cpp +++ b/src/core/hw/gpu.cpp @@ -5,6 +5,7 @@ #include #include #include +#include "common/alignment.h" #include "common/color.h" #include "common/common_types.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 dst_addr = config.GetPhysicalOutputAddress(); - // TODO: do hwtest with these cases + // TODO: do hwtest with invalid addresses if (!Memory::IsValidPhysicalAddress(src_addr)) { LOG_CRITICAL(HW_GPU, "invalid input address 0x%08X", src_addr); return; @@ -324,31 +325,36 @@ static void TextureCopy(const Regs::DisplayTransferConfig& config) { return; } - if (config.texture_copy.input_width == 0) { - LOG_CRITICAL(HW_GPU, "zero input width"); + if (VideoCore::g_renderer->Rasterizer()->AccelerateTextureCopy(config)) return; - } - if (config.texture_copy.output_width == 0) { - LOG_CRITICAL(HW_GPU, "zero output width"); - return; - } + u8* src_pointer = Memory::GetPhysicalPointer(src_addr); + u8* dst_pointer = Memory::GetPhysicalPointer(dst_addr); - if (config.texture_copy.size == 0) { + u32 remaining_size = Common::AlignDown(config.texture_copy.size, 16); + + if (remaining_size == 0) { + // Real hardware freeze on this 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 input_width = input_gap == 0 ? remaining_size : config.texture_copy.input_width * 16; u32 output_gap = config.texture_copy.output_gap * 16; + u32 output_width = output_gap == 0 ? remaining_size : config.texture_copy.output_width * 16; + + if (input_width == 0) { + // Real hardware freeze on this + LOG_CRITICAL(HW_GPU, "zero input width"); + return; + } + + if (output_width == 0) { + // Real hardware freeze on this + LOG_CRITICAL(HW_GPU, "zero output width"); + return; + } size_t contiguous_input_size = config.texture_copy.size / input_width * (input_width + input_gap); @@ -360,7 +366,6 @@ static void TextureCopy(const Regs::DisplayTransferConfig& config) { Memory::RasterizerFlushAndInvalidateRegion(config.GetPhysicalOutputAddress(), static_cast(contiguous_output_size)); - u32 remaining_size = config.texture_copy.size; u32 remaining_input = input_width; u32 remaining_output = output_width; while (remaining_size > 0) { -- cgit v1.2.3 From 85a448d40560a40d5fe2424b4c50b7bebe2a6064 Mon Sep 17 00:00:00 2001 From: wwylele Date: Thu, 29 Jun 2017 13:09:23 +0300 Subject: gpu: add comments for TextureCopy --- src/core/hw/gpu.cpp | 14 +++++++------- src/core/hw/gpu.h | 2 +- 2 files changed, 8 insertions(+), 8 deletions(-) (limited to 'src') diff --git a/src/core/hw/gpu.cpp b/src/core/hw/gpu.cpp index a0a523822..6838e449c 100644 --- a/src/core/hw/gpu.cpp +++ b/src/core/hw/gpu.cpp @@ -334,25 +334,25 @@ static void TextureCopy(const Regs::DisplayTransferConfig& config) { u32 remaining_size = Common::AlignDown(config.texture_copy.size, 16); if (remaining_size == 0) { - // Real hardware freeze on this - LOG_CRITICAL(HW_GPU, "zero size"); + LOG_CRITICAL(HW_GPU, "zero size. Real hardware freezes on this."); return; } u32 input_gap = config.texture_copy.input_gap * 16; - u32 input_width = input_gap == 0 ? remaining_size : config.texture_copy.input_width * 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) { - // Real hardware freeze on this - LOG_CRITICAL(HW_GPU, "zero input width"); + LOG_CRITICAL(HW_GPU, "zero input width. Real hardware freezes on this."); return; } if (output_width == 0) { - // Real hardware freeze on this - LOG_CRITICAL(HW_GPU, "zero output width"); + LOG_CRITICAL(HW_GPU, "zero output width. Real hardware freezes on this."); return; } diff --git a/src/core/hw/gpu.h b/src/core/hw/gpu.h index bdd997b2a..21b127fee 100644 --- a/src/core/hw/gpu.h +++ b/src/core/hw/gpu.h @@ -225,7 +225,7 @@ struct Regs { INSERT_PADDING_WORDS(0x1); struct { - u32 size; + u32 size; // The lower 4 bits are ignored union { u32 input_size; -- cgit v1.2.3