diff options
Diffstat (limited to 'src/video_core/renderer_opengl')
| -rw-r--r-- | src/video_core/renderer_opengl/gl_rasterizer_cache.cpp | 43 | ||||
| -rw-r--r-- | src/video_core/renderer_opengl/gl_shader_decompiler.cpp | 82 |
2 files changed, 97 insertions, 28 deletions
diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp index ae7152bd3..801d45144 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -331,8 +331,8 @@ static bool IsFormatBCn(PixelFormat format) { } template <bool morton_to_gl, PixelFormat format> -void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, std::size_t gl_buffer_size, - VAddr addr) { +void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth, u32 depth, u8* gl_buffer, + std::size_t gl_buffer_size, VAddr addr) { constexpr u32 bytes_per_pixel = SurfaceParams::GetFormatBpp(format) / CHAR_BIT; constexpr u32 gl_bytes_per_pixel = CachedSurface::GetGLBytesPerPixel(format); @@ -341,7 +341,7 @@ void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, std::si // pixel values. const u32 tile_size{IsFormatBCn(format) ? 4U : 1U}; const std::vector<u8> data = Tegra::Texture::UnswizzleTexture( - addr, tile_size, bytes_per_pixel, stride, height, block_height); + addr, tile_size, bytes_per_pixel, stride, height, depth, block_height, block_depth); const std::size_t size_to_copy{std::min(gl_buffer_size, data.size())}; memcpy(gl_buffer, data.data(), size_to_copy); } else { @@ -353,7 +353,7 @@ void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, std::si } } -static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr), +static constexpr std::array<void (*)(u32, u32, u32, u32, u32, u8*, std::size_t, VAddr), SurfaceParams::MaxPixelFormat> morton_to_gl_fns = { // clang-format off @@ -413,7 +413,7 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr), // clang-format on }; -static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr), +static constexpr std::array<void (*)(u32, u32, u32, u32, u32, u8*, std::size_t, VAddr), SurfaceParams::MaxPixelFormat> gl_to_morton_fns = { // clang-format off @@ -841,36 +841,23 @@ void CachedSurface::LoadGLBuffer() { if (params.is_tiled) { gl_buffer.resize(total_size); + u32 depth = params.depth; + u32 block_depth = params.block_depth; ASSERT_MSG(params.block_width == 1, "Block width is defined as {} on texture type {}", params.block_width, static_cast<u32>(params.target)); - ASSERT_MSG(params.block_depth == 1, "Block depth is defined as {} on texture type {}", - params.block_depth, static_cast<u32>(params.target)); - // TODO(bunnei): This only unswizzles and copies a 2D texture - we do not yet know how to do - // this for 3D textures, etc. - switch (params.target) { - case SurfaceParams::SurfaceTarget::Texture2D: - // Pass impl. to the fallback code below - break; - case SurfaceParams::SurfaceTarget::Texture2DArray: - case SurfaceParams::SurfaceTarget::TextureCubemap: - for (std::size_t index = 0; index < params.depth; ++index) { - const std::size_t offset{index * copy_size}; - morton_to_gl_fns[static_cast<std::size_t>(params.pixel_format)]( - params.width, params.block_height, params.height, gl_buffer.data() + offset, - copy_size, params.addr + offset); - } - break; - default: - LOG_CRITICAL(HW_GPU, "Unimplemented tiled load for target={}", - static_cast<u32>(params.target)); - UNREACHABLE(); + if (params.target == SurfaceParams::SurfaceTarget::Texture2D) { + // TODO(Blinkhawk): Eliminate this condition once all texture types are implemented. + depth = 1U; + block_depth = 1U; } + const std::size_t size = copy_size * depth; + morton_to_gl_fns[static_cast<std::size_t>(params.pixel_format)]( - params.width, params.block_height, params.height, gl_buffer.data(), copy_size, - params.addr); + params.width, params.block_height, params.height, block_depth, depth, gl_buffer.data(), + size, params.addr); } else { const u8* const texture_src_data_end{texture_src_data + total_size}; gl_buffer.assign(texture_src_data, texture_src_data_end); diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index c82a0dcfa..8dfb49507 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -2953,6 +2953,88 @@ private: LOG_WARNING(HW_GPU, "DEPBAR instruction is stubbed"); break; } + case OpCode::Id::VMAD: { + const bool signed_a = instr.vmad.signed_a == 1; + const bool signed_b = instr.vmad.signed_b == 1; + const bool result_signed = signed_a || signed_b; + boost::optional<std::string> forced_result; + + auto Unpack = [&](const std::string& op, bool is_chunk, bool is_signed, + Tegra::Shader::VmadType type, u64 byte_height) { + const std::string value = [&]() { + if (!is_chunk) { + const auto offset = static_cast<u32>(byte_height * 8); + return "((" + op + " >> " + std::to_string(offset) + ") & 0xff)"; + } + const std::string zero = "0"; + + switch (type) { + case Tegra::Shader::VmadType::Size16_Low: + return '(' + op + " & 0xffff)"; + case Tegra::Shader::VmadType::Size16_High: + return '(' + op + " >> 16)"; + case Tegra::Shader::VmadType::Size32: + // TODO(Rodrigo): From my hardware tests it becomes a bit "mad" when + // this type is used (1 * 1 + 0 == 0x5b800000). Until a better + // explanation is found: assert. + UNREACHABLE_MSG("Unimplemented"); + return zero; + case Tegra::Shader::VmadType::Invalid: + // Note(Rodrigo): This flag is invalid according to nvdisasm. From my + // testing (even though it's invalid) this makes the whole instruction + // assign zero to target register. + forced_result = boost::make_optional(zero); + return zero; + default: + UNREACHABLE(); + return zero; + } + }(); + + if (is_signed) { + return "int(" + value + ')'; + } + return value; + }; + + const std::string op_a = Unpack(regs.GetRegisterAsInteger(instr.gpr8, 0, false), + instr.vmad.is_byte_chunk_a != 0, signed_a, + instr.vmad.type_a, instr.vmad.byte_height_a); + + std::string op_b; + if (instr.vmad.use_register_b) { + op_b = Unpack(regs.GetRegisterAsInteger(instr.gpr20, 0, false), + instr.vmad.is_byte_chunk_b != 0, signed_b, instr.vmad.type_b, + instr.vmad.byte_height_b); + } else { + op_b = '(' + + std::to_string(signed_b ? static_cast<s16>(instr.alu.GetImm20_16()) + : instr.alu.GetImm20_16()) + + ')'; + } + + const std::string op_c = regs.GetRegisterAsInteger(instr.gpr39, 0, result_signed); + + std::string result; + if (forced_result) { + result = *forced_result; + } else { + result = '(' + op_a + " * " + op_b + " + " + op_c + ')'; + + switch (instr.vmad.shr) { + case Tegra::Shader::VmadShr::Shr7: + result = '(' + result + " >> 7)"; + break; + case Tegra::Shader::VmadShr::Shr15: + result = '(' + result + " >> 15)"; + break; + } + } + regs.SetRegisterToInteger(instr.gpr0, result_signed, 1, result, 1, 1, + instr.vmad.saturate == 1, 0, Register::Size::Word, + instr.vmad.cc); + break; + } default: { LOG_CRITICAL(HW_GPU, "Unhandled instruction: {}", opcode->GetName()); UNREACHABLE(); |