summaryrefslogtreecommitdiffstats
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core')
-rw-r--r--src/video_core/engines/fermi_2d.cpp14
-rw-r--r--src/video_core/engines/maxwell_dma.cpp10
-rw-r--r--src/video_core/engines/maxwell_dma.h4
-rw-r--r--src/video_core/engines/shader_bytecode.h36
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer_cache.cpp43
-rw-r--r--src/video_core/renderer_opengl/gl_shader_decompiler.cpp82
-rw-r--r--src/video_core/textures/decoders.cpp207
-rw-r--r--src/video_core/textures/decoders.h21
-rw-r--r--src/video_core/textures/texture.h1
9 files changed, 313 insertions, 105 deletions
diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp
index 912e785b9..597b279b9 100644
--- a/src/video_core/engines/fermi_2d.cpp
+++ b/src/video_core/engines/fermi_2d.cpp
@@ -62,14 +62,16 @@ void Fermi2D::HandleSurfaceCopy() {
u8* dst_buffer = Memory::GetPointer(dest_cpu);
if (!regs.src.linear && regs.dst.linear) {
// If the input is tiled and the output is linear, deswizzle the input and copy it over.
- Texture::CopySwizzledData(regs.src.width, regs.src.height, src_bytes_per_pixel,
- dst_bytes_per_pixel, src_buffer, dst_buffer, true,
- regs.src.BlockHeight());
+ Texture::CopySwizzledData(regs.src.width, regs.src.height, regs.src.depth,
+ src_bytes_per_pixel, dst_bytes_per_pixel, src_buffer,
+ dst_buffer, true, regs.src.BlockHeight(),
+ regs.src.BlockDepth());
} else {
// If the input is linear and the output is tiled, swizzle the input and copy it over.
- Texture::CopySwizzledData(regs.src.width, regs.src.height, src_bytes_per_pixel,
- dst_bytes_per_pixel, dst_buffer, src_buffer, false,
- regs.dst.BlockHeight());
+ Texture::CopySwizzledData(regs.src.width, regs.src.height, regs.src.depth,
+ src_bytes_per_pixel, dst_bytes_per_pixel, dst_buffer,
+ src_buffer, false, regs.dst.BlockHeight(),
+ regs.dst.BlockDepth());
}
}
}
diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp
index aa7481b8c..bf2a21bb6 100644
--- a/src/video_core/engines/maxwell_dma.cpp
+++ b/src/video_core/engines/maxwell_dma.cpp
@@ -68,12 +68,14 @@ void MaxwellDMA::HandleCopy() {
if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) {
// If the input is tiled and the output is linear, deswizzle the input and copy it over.
- Texture::CopySwizzledData(regs.src_params.size_x, regs.src_params.size_y, 1, 1, src_buffer,
- dst_buffer, true, regs.src_params.BlockHeight());
+ Texture::CopySwizzledData(regs.src_params.size_x, regs.src_params.size_y,
+ regs.src_params.size_z, 1, 1, src_buffer, dst_buffer, true,
+ regs.src_params.BlockHeight(), regs.src_params.BlockDepth());
} else {
// If the input is linear and the output is tiled, swizzle the input and copy it over.
- Texture::CopySwizzledData(regs.dst_params.size_x, regs.dst_params.size_y, 1, 1, dst_buffer,
- src_buffer, false, regs.dst_params.BlockHeight());
+ Texture::CopySwizzledData(regs.dst_params.size_x, regs.dst_params.size_y,
+ regs.dst_params.size_z, 1, 1, dst_buffer, src_buffer, false,
+ regs.dst_params.BlockHeight(), regs.dst_params.BlockDepth());
}
}
diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h
index 311ccb616..df19e02e2 100644
--- a/src/video_core/engines/maxwell_dma.h
+++ b/src/video_core/engines/maxwell_dma.h
@@ -43,6 +43,10 @@ public:
u32 BlockHeight() const {
return 1 << block_height;
}
+
+ u32 BlockDepth() const {
+ return 1 << block_depth;
+ }
};
static_assert(sizeof(Parameters) == 24, "Parameters has wrong size");
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index 550ab1148..9a59b65b3 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -214,6 +214,18 @@ enum class IMinMaxExchange : u64 {
XHi = 3,
};
+enum class VmadType : u64 {
+ Size16_Low = 0,
+ Size16_High = 1,
+ Size32 = 2,
+ Invalid = 3,
+};
+
+enum class VmadShr : u64 {
+ Shr7 = 1,
+ Shr15 = 2,
+};
+
enum class XmadMode : u64 {
None = 0,
CLo = 1,
@@ -452,6 +464,7 @@ union Instruction {
BitField<48, 16, u64> opcode;
union {
+ BitField<20, 16, u64> imm20_16;
BitField<20, 19, u64> imm20_19;
BitField<20, 32, s64> imm20_32;
BitField<45, 1, u64> negate_b;
@@ -493,6 +506,10 @@ union Instruction {
}
} lop3;
+ u16 GetImm20_16() const {
+ return static_cast<u16>(imm20_16);
+ }
+
u32 GetImm20_19() const {
u32 imm{static_cast<u32>(imm20_19)};
imm <<= 12;
@@ -1017,6 +1034,23 @@ union Instruction {
} isberd;
union {
+ BitField<48, 1, u64> signed_a;
+ BitField<38, 1, u64> is_byte_chunk_a;
+ BitField<36, 2, VmadType> type_a;
+ BitField<36, 2, u64> byte_height_a;
+
+ BitField<49, 1, u64> signed_b;
+ BitField<50, 1, u64> use_register_b;
+ BitField<30, 1, u64> is_byte_chunk_b;
+ BitField<28, 2, VmadType> type_b;
+ BitField<28, 2, u64> byte_height_b;
+
+ BitField<51, 2, VmadShr> shr;
+ BitField<55, 1, u64> saturate; // Saturates the result (a * b + c)
+ BitField<47, 1, u64> cc;
+ } vmad;
+
+ union {
BitField<20, 16, u64> imm20_16;
BitField<36, 1, u64> product_shift_left;
BitField<37, 1, u64> merge_37;
@@ -1083,6 +1117,7 @@ public:
IPA,
OUT_R, // Emit vertex/primitive
ISBERD,
+ VMAD,
FFMA_IMM, // Fused Multiply and Add
FFMA_CR,
FFMA_RC,
@@ -1320,6 +1355,7 @@ private:
INST("11100000--------", Id::IPA, Type::Trivial, "IPA"),
INST("1111101111100---", Id::OUT_R, Type::Trivial, "OUT_R"),
INST("1110111111010---", Id::ISBERD, Type::Trivial, "ISBERD"),
+ INST("01011111--------", Id::VMAD, Type::Trivial, "VMAD"),
INST("0011001-1-------", Id::FFMA_IMM, Type::Ffma, "FFMA_IMM"),
INST("010010011-------", Id::FFMA_CR, Type::Ffma, "FFMA_CR"),
INST("010100011-------", Id::FFMA_RC, Type::Ffma, "FFMA_RC"),
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();
diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp
index 8d2daa284..18ab723f7 100644
--- a/src/video_core/textures/decoders.cpp
+++ b/src/video_core/textures/decoders.cpp
@@ -4,6 +4,7 @@
#include <cmath>
#include <cstring>
+#include "common/alignment.h"
#include "common/assert.h"
#include "core/memory.h"
#include "video_core/gpu.h"
@@ -39,72 +40,146 @@ struct alignas(64) SwizzleTable {
constexpr auto legacy_swizzle_table = SwizzleTable<8, 64, 1>();
constexpr auto fast_swizzle_table = SwizzleTable<8, 4, 16>();
-static void LegacySwizzleData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel,
- u8* swizzled_data, u8* unswizzled_data, bool unswizzle,
- u32 block_height) {
+/**
+ * This function manages ALL the GOBs(Group of Bytes) Inside a single block.
+ * Instead of going gob by gob, we map the coordinates inside a block and manage from
+ * those. Block_Width is assumed to be 1.
+ */
+void PreciseProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle,
+ const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end,
+ const u32 y_end, const u32 z_end, const u32 tile_offset,
+ const u32 xy_block_size, const u32 layer_z, const u32 stride_x,
+ const u32 bytes_per_pixel, const u32 out_bytes_per_pixel) {
std::array<u8*, 2> data_ptrs;
- const std::size_t stride = width * bytes_per_pixel;
- const std::size_t gobs_in_x = 64;
- const std::size_t gobs_in_y = 8;
- const std::size_t gobs_size = gobs_in_x * gobs_in_y;
- const std::size_t image_width_in_gobs{(stride + gobs_in_x - 1) / gobs_in_x};
- for (std::size_t y = 0; y < height; ++y) {
- const std::size_t gob_y_address =
- (y / (gobs_in_y * block_height)) * gobs_size * block_height * image_width_in_gobs +
- (y % (gobs_in_y * block_height) / gobs_in_y) * gobs_size;
- const auto& table = legacy_swizzle_table[y % gobs_in_y];
- for (std::size_t x = 0; x < width; ++x) {
- const std::size_t gob_address =
- gob_y_address + (x * bytes_per_pixel / gobs_in_x) * gobs_size * block_height;
- const std::size_t x2 = x * bytes_per_pixel;
- const std::size_t swizzle_offset = gob_address + table[x2 % gobs_in_x];
- const std::size_t pixel_index = (x + y * width) * out_bytes_per_pixel;
-
- data_ptrs[unswizzle] = swizzled_data + swizzle_offset;
- data_ptrs[!unswizzle] = unswizzled_data + pixel_index;
-
- std::memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel);
+ u32 z_address = tile_offset;
+ const u32 gob_size_x = 64;
+ const u32 gob_size_y = 8;
+ const u32 gob_size_z = 1;
+ const u32 gob_size = gob_size_x * gob_size_y * gob_size_z;
+ for (u32 z = z_start; z < z_end; z++) {
+ u32 y_address = z_address;
+ u32 pixel_base = layer_z * z + y_start * stride_x;
+ for (u32 y = y_start; y < y_end; y++) {
+ const auto& table = legacy_swizzle_table[y % gob_size_y];
+ for (u32 x = x_start; x < x_end; x++) {
+ const u32 swizzle_offset{y_address + table[x * bytes_per_pixel % gob_size_x]};
+ const u32 pixel_index{x * out_bytes_per_pixel + pixel_base};
+ data_ptrs[unswizzle] = swizzled_data + swizzle_offset;
+ data_ptrs[!unswizzle] = unswizzled_data + pixel_index;
+ std::memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel);
+ }
+ pixel_base += stride_x;
+ if ((y + 1) % gob_size_y == 0)
+ y_address += gob_size;
}
+ z_address += xy_block_size;
}
}
-static void FastSwizzleData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel,
- u8* swizzled_data, u8* unswizzled_data, bool unswizzle,
- u32 block_height) {
+/**
+ * This function manages ALL the GOBs(Group of Bytes) Inside a single block.
+ * Instead of going gob by gob, we map the coordinates inside a block and manage from
+ * those. Block_Width is assumed to be 1.
+ */
+void FastProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle,
+ const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end,
+ const u32 y_end, const u32 z_end, const u32 tile_offset,
+ const u32 xy_block_size, const u32 layer_z, const u32 stride_x,
+ const u32 bytes_per_pixel, const u32 out_bytes_per_pixel) {
std::array<u8*, 2> data_ptrs;
- const std::size_t stride{width * bytes_per_pixel};
- const std::size_t gobs_in_x = 64;
- const std::size_t gobs_in_y = 8;
- const std::size_t gobs_size = gobs_in_x * gobs_in_y;
- const std::size_t image_width_in_gobs{(stride + gobs_in_x - 1) / gobs_in_x};
- const std::size_t copy_size{16};
- for (std::size_t y = 0; y < height; ++y) {
- const std::size_t initial_gob =
- (y / (gobs_in_y * block_height)) * gobs_size * block_height * image_width_in_gobs +
- (y % (gobs_in_y * block_height) / gobs_in_y) * gobs_size;
- const std::size_t pixel_base{y * width * out_bytes_per_pixel};
- const auto& table = fast_swizzle_table[y % gobs_in_y];
- for (std::size_t xb = 0; xb < stride; xb += copy_size) {
- const std::size_t gob_address{initial_gob +
- (xb / gobs_in_x) * gobs_size * block_height};
- const std::size_t swizzle_offset{gob_address + table[(xb / 16) % 4]};
- const std::size_t out_x = xb * out_bytes_per_pixel / bytes_per_pixel;
- const std::size_t pixel_index{out_x + pixel_base};
- data_ptrs[unswizzle] = swizzled_data + swizzle_offset;
- data_ptrs[!unswizzle] = unswizzled_data + pixel_index;
- std::memcpy(data_ptrs[0], data_ptrs[1], copy_size);
+ u32 z_address = tile_offset;
+ const u32 x_startb = x_start * bytes_per_pixel;
+ const u32 x_endb = x_end * bytes_per_pixel;
+ const u32 copy_size = 16;
+ const u32 gob_size_x = 64;
+ const u32 gob_size_y = 8;
+ const u32 gob_size_z = 1;
+ const u32 gob_size = gob_size_x * gob_size_y * gob_size_z;
+ for (u32 z = z_start; z < z_end; z++) {
+ u32 y_address = z_address;
+ u32 pixel_base = layer_z * z + y_start * stride_x;
+ for (u32 y = y_start; y < y_end; y++) {
+ const auto& table = fast_swizzle_table[y % gob_size_y];
+ for (u32 xb = x_startb; xb < x_endb; xb += copy_size) {
+ const u32 swizzle_offset{y_address + table[(xb / copy_size) % 4]};
+ const u32 out_x = xb * out_bytes_per_pixel / bytes_per_pixel;
+ const u32 pixel_index{out_x + pixel_base};
+ data_ptrs[unswizzle] = swizzled_data + swizzle_offset;
+ data_ptrs[!unswizzle] = unswizzled_data + pixel_index;
+ std::memcpy(data_ptrs[0], data_ptrs[1], copy_size);
+ }
+ pixel_base += stride_x;
+ if ((y + 1) % gob_size_y == 0)
+ y_address += gob_size;
}
+ z_address += xy_block_size;
}
}
-void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel,
- u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 block_height) {
+/**
+ * This function unswizzles or swizzles a texture by mapping Linear to BlockLinear Textue.
+ * The body of this function takes care of splitting the swizzled texture into blocks,
+ * and managing the extents of it. Once all the parameters of a single block are obtained,
+ * the function calls 'ProcessBlock' to process that particular Block.
+ *
+ * Documentation for the memory layout and decoding can be found at:
+ * https://envytools.readthedocs.io/en/latest/hw/memory/g80-surface.html#blocklinear-surfaces
+ */
+template <bool fast>
+void SwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, const u32 width,
+ const u32 height, const u32 depth, const u32 bytes_per_pixel,
+ const u32 out_bytes_per_pixel, const u32 block_height, const u32 block_depth) {
+ auto div_ceil = [](const u32 x, const u32 y) { return ((x + y - 1) / y); };
+ const u32 stride_x = width * out_bytes_per_pixel;
+ const u32 layer_z = height * stride_x;
+ const u32 gob_x_bytes = 64;
+ const u32 gob_elements_x = gob_x_bytes / bytes_per_pixel;
+ const u32 gob_elements_y = 8;
+ const u32 gob_elements_z = 1;
+ const u32 block_x_elements = gob_elements_x;
+ const u32 block_y_elements = gob_elements_y * block_height;
+ const u32 block_z_elements = gob_elements_z * block_depth;
+ const u32 blocks_on_x = div_ceil(width, block_x_elements);
+ const u32 blocks_on_y = div_ceil(height, block_y_elements);
+ const u32 blocks_on_z = div_ceil(depth, block_z_elements);
+ const u32 blocks = blocks_on_x * blocks_on_y * blocks_on_z;
+ const u32 gob_size = gob_x_bytes * gob_elements_y * gob_elements_z;
+ const u32 xy_block_size = gob_size * block_height;
+ const u32 block_size = xy_block_size * block_depth;
+ u32 tile_offset = 0;
+ for (u32 zb = 0; zb < blocks_on_z; zb++) {
+ const u32 z_start = zb * block_z_elements;
+ const u32 z_end = std::min(depth, z_start + block_z_elements);
+ for (u32 yb = 0; yb < blocks_on_y; yb++) {
+ const u32 y_start = yb * block_y_elements;
+ const u32 y_end = std::min(height, y_start + block_y_elements);
+ for (u32 xb = 0; xb < blocks_on_x; xb++) {
+ const u32 x_start = xb * block_x_elements;
+ const u32 x_end = std::min(width, x_start + block_x_elements);
+ if (fast) {
+ FastProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start,
+ z_start, x_end, y_end, z_end, tile_offset, xy_block_size,
+ layer_z, stride_x, bytes_per_pixel, out_bytes_per_pixel);
+ } else {
+ PreciseProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start,
+ z_start, x_end, y_end, z_end, tile_offset, xy_block_size,
+ layer_z, stride_x, bytes_per_pixel, out_bytes_per_pixel);
+ }
+ tile_offset += block_size;
+ }
+ }
+ }
+}
+
+void CopySwizzledData(u32 width, u32 height, u32 depth, u32 bytes_per_pixel,
+ u32 out_bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data,
+ bool unswizzle, u32 block_height, u32 block_depth) {
if (bytes_per_pixel % 3 != 0 && (width * bytes_per_pixel) % 16 == 0) {
- FastSwizzleData(width, height, bytes_per_pixel, out_bytes_per_pixel, swizzled_data,
- unswizzled_data, unswizzle, block_height);
+ SwizzledData<true>(swizzled_data, unswizzled_data, unswizzle, width, height, depth,
+ bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth);
} else {
- LegacySwizzleData(width, height, bytes_per_pixel, out_bytes_per_pixel, swizzled_data,
- unswizzled_data, unswizzle, block_height);
+ SwizzledData<false>(swizzled_data, unswizzled_data, unswizzle, width, height, depth,
+ bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth);
}
}
@@ -154,10 +229,11 @@ u32 BytesPerPixel(TextureFormat format) {
}
std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size, u32 bytes_per_pixel, u32 width,
- u32 height, u32 block_height) {
- std::vector<u8> unswizzled_data(width * height * bytes_per_pixel);
- CopySwizzledData(width / tile_size, height / tile_size, bytes_per_pixel, bytes_per_pixel,
- Memory::GetPointer(address), unswizzled_data.data(), true, block_height);
+ u32 height, u32 depth, u32 block_height, u32 block_depth) {
+ std::vector<u8> unswizzled_data(width * height * depth * bytes_per_pixel);
+ CopySwizzledData(width / tile_size, height / tile_size, depth, bytes_per_pixel, bytes_per_pixel,
+ Memory::GetPointer(address), unswizzled_data.data(), true, block_height,
+ block_depth);
return unswizzled_data;
}
@@ -201,4 +277,19 @@ std::vector<u8> DecodeTexture(const std::vector<u8>& texture_data, TextureFormat
return rgba_data;
}
+std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height, u32 depth,
+ u32 block_height, u32 block_depth) {
+ if (tiled) {
+ const u32 gobs_in_x = 64 / bytes_per_pixel;
+ const u32 gobs_in_y = 8;
+ const u32 gobs_in_z = 1;
+ const u32 aligned_width = Common::AlignUp(width, gobs_in_x);
+ const u32 aligned_height = Common::AlignUp(height, gobs_in_y * block_height);
+ const u32 aligned_depth = Common::AlignUp(depth, gobs_in_z * block_depth);
+ return aligned_width * aligned_height * aligned_depth * bytes_per_pixel;
+ } else {
+ return width * height * depth * bytes_per_pixel;
+ }
+}
+
} // namespace Tegra::Texture
diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h
index 1f7b731be..aaf316947 100644
--- a/src/video_core/textures/decoders.h
+++ b/src/video_core/textures/decoders.h
@@ -14,17 +14,14 @@ namespace Tegra::Texture {
* Unswizzles a swizzled texture without changing its format.
*/
std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size, u32 bytes_per_pixel, u32 width,
- u32 height, u32 block_height = TICEntry::DefaultBlockHeight);
-
-/**
- * Unswizzles a swizzled depth texture without changing its format.
- */
-std::vector<u8> UnswizzleDepthTexture(VAddr address, DepthFormat format, u32 width, u32 height,
- u32 block_height = TICEntry::DefaultBlockHeight);
+ u32 height, u32 depth,
+ u32 block_height = TICEntry::DefaultBlockHeight,
+ u32 block_depth = TICEntry::DefaultBlockHeight);
/// Copies texture data from a buffer and performs swizzling/unswizzling as necessary.
-void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel,
- u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 block_height);
+void CopySwizzledData(u32 width, u32 height, u32 depth, u32 bytes_per_pixel,
+ u32 out_bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data,
+ bool unswizzle, u32 block_height, u32 block_depth);
/**
* Decodes an unswizzled texture into a A8R8G8B8 texture.
@@ -32,4 +29,10 @@ void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_
std::vector<u8> DecodeTexture(const std::vector<u8>& texture_data, TextureFormat format, u32 width,
u32 height);
+/**
+ * This function calculates the correct size of a texture depending if it's tiled or not.
+ */
+std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height, u32 depth,
+ u32 block_height, u32 block_depth);
+
} // namespace Tegra::Texture
diff --git a/src/video_core/textures/texture.h b/src/video_core/textures/texture.h
index 58d17abcb..5947bd2b9 100644
--- a/src/video_core/textures/texture.h
+++ b/src/video_core/textures/texture.h
@@ -141,6 +141,7 @@ static_assert(sizeof(TextureHandle) == 4, "TextureHandle has wrong size");
struct TICEntry {
static constexpr u32 DefaultBlockHeight = 16;
+ static constexpr u32 DefaultBlockDepth = 1;
union {
u32 raw;
generated by cgit v1.2.3 (git 2.25.1) at 2024-08-21 04:47:16 +0200