summaryrefslogtreecommitdiffstats
path: root/src/video_core/textures
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core/textures')
-rw-r--r--src/video_core/textures/astc.cpp156
-rw-r--r--src/video_core/textures/astc.h111
-rw-r--r--src/video_core/textures/decoders.cpp52
3 files changed, 176 insertions, 143 deletions
diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp
index 3ab500760..25161df1f 100644
--- a/src/video_core/textures/astc.cpp
+++ b/src/video_core/textures/astc.cpp
@@ -151,6 +151,76 @@ private:
const IntType& m_Bits;
};
+enum class IntegerEncoding { JustBits, Quint, Trit };
+
+struct IntegerEncodedValue {
+ constexpr IntegerEncodedValue() = default;
+
+ constexpr IntegerEncodedValue(IntegerEncoding encoding_, u32 num_bits_)
+ : encoding{encoding_}, num_bits{num_bits_} {}
+
+ constexpr bool MatchesEncoding(const IntegerEncodedValue& other) const {
+ return encoding == other.encoding && num_bits == other.num_bits;
+ }
+
+ // Returns the number of bits required to encode num_vals values.
+ u32 GetBitLength(u32 num_vals) const {
+ u32 total_bits = num_bits * num_vals;
+ if (encoding == IntegerEncoding::Trit) {
+ total_bits += (num_vals * 8 + 4) / 5;
+ } else if (encoding == IntegerEncoding::Quint) {
+ total_bits += (num_vals * 7 + 2) / 3;
+ }
+ return total_bits;
+ }
+
+ IntegerEncoding encoding{};
+ u32 num_bits = 0;
+ u32 bit_value = 0;
+ union {
+ u32 quint_value = 0;
+ u32 trit_value;
+ };
+};
+
+// Returns a new instance of this struct that corresponds to the
+// can take no more than mav_value values
+static constexpr IntegerEncodedValue CreateEncoding(u32 mav_value) {
+ while (mav_value > 0) {
+ u32 check = mav_value + 1;
+
+ // Is mav_value a power of two?
+ if (!(check & (check - 1))) {
+ return IntegerEncodedValue(IntegerEncoding::JustBits, std::popcount(mav_value));
+ }
+
+ // Is mav_value of the type 3*2^n - 1?
+ if ((check % 3 == 0) && !((check / 3) & ((check / 3) - 1))) {
+ return IntegerEncodedValue(IntegerEncoding::Trit, std::popcount(check / 3 - 1));
+ }
+
+ // Is mav_value of the type 5*2^n - 1?
+ if ((check % 5 == 0) && !((check / 5) & ((check / 5) - 1))) {
+ return IntegerEncodedValue(IntegerEncoding::Quint, std::popcount(check / 5 - 1));
+ }
+
+ // Apparently it can't be represented with a bounded integer sequence...
+ // just iterate.
+ mav_value--;
+ }
+ return IntegerEncodedValue(IntegerEncoding::JustBits, 0);
+}
+
+static constexpr std::array<IntegerEncodedValue, 256> MakeEncodedValues() {
+ std::array<IntegerEncodedValue, 256> encodings{};
+ for (std::size_t i = 0; i < encodings.size(); ++i) {
+ encodings[i] = CreateEncoding(static_cast<u32>(i));
+ }
+ return encodings;
+}
+
+static constexpr std::array<IntegerEncodedValue, 256> ASTC_ENCODINGS_VALUES = MakeEncodedValues();
+
namespace Tegra::Texture::ASTC {
using IntegerEncodedVector = boost::container::static_vector<
IntegerEncodedValue, 256,
@@ -521,35 +591,41 @@ static TexelWeightParams DecodeBlockInfo(InputBitStream& strm) {
return params;
}
-static void FillVoidExtentLDR(InputBitStream& strm, std::span<u32> outBuf, u32 blockWidth,
- u32 blockHeight) {
- // Don't actually care about the void extent, just read the bits...
- for (s32 i = 0; i < 4; ++i) {
- strm.ReadBits<13>();
+// Replicates low num_bits such that [(to_bit - 1):(to_bit - 1 - from_bit)]
+// is the same as [(num_bits - 1):0] and repeats all the way down.
+template <typename IntType>
+static constexpr IntType Replicate(IntType val, u32 num_bits, u32 to_bit) {
+ if (num_bits == 0 || to_bit == 0) {
+ return 0;
}
-
- // Decode the RGBA components and renormalize them to the range [0, 255]
- u16 r = static_cast<u16>(strm.ReadBits<16>());
- u16 g = static_cast<u16>(strm.ReadBits<16>());
- u16 b = static_cast<u16>(strm.ReadBits<16>());
- u16 a = static_cast<u16>(strm.ReadBits<16>());
-
- u32 rgba = (r >> 8) | (g & 0xFF00) | (static_cast<u32>(b) & 0xFF00) << 8 |
- (static_cast<u32>(a) & 0xFF00) << 16;
-
- for (u32 j = 0; j < blockHeight; j++) {
- for (u32 i = 0; i < blockWidth; i++) {
- outBuf[j * blockWidth + i] = rgba;
+ const IntType v = val & static_cast<IntType>((1 << num_bits) - 1);
+ IntType res = v;
+ u32 reslen = num_bits;
+ while (reslen < to_bit) {
+ u32 comp = 0;
+ if (num_bits > to_bit - reslen) {
+ u32 newshift = to_bit - reslen;
+ comp = num_bits - newshift;
+ num_bits = newshift;
}
+ res = static_cast<IntType>(res << num_bits);
+ res = static_cast<IntType>(res | (v >> comp));
+ reslen += num_bits;
}
+ return res;
}
-static void FillError(std::span<u32> outBuf, u32 blockWidth, u32 blockHeight) {
- for (u32 j = 0; j < blockHeight; j++) {
- for (u32 i = 0; i < blockWidth; i++) {
- outBuf[j * blockWidth + i] = 0xFFFF00FF;
- }
+static constexpr std::size_t NumReplicateEntries(u32 num_bits) {
+ return std::size_t(1) << num_bits;
+}
+
+template <typename IntType, u32 num_bits, u32 to_bit>
+static constexpr auto MakeReplicateTable() {
+ std::array<IntType, NumReplicateEntries(num_bits)> table{};
+ for (IntType value = 0; value < static_cast<IntType>(std::size(table)); ++value) {
+ table[value] = Replicate(value, num_bits, to_bit);
}
+ return table;
}
static constexpr auto REPLICATE_BYTE_TO_16_TABLE = MakeReplicateTable<u32, 8, 16>();
@@ -572,6 +648,9 @@ static constexpr auto REPLICATE_2_BIT_TO_8_TABLE = MakeReplicateTable<u32, 2, 8>
static constexpr auto REPLICATE_3_BIT_TO_8_TABLE = MakeReplicateTable<u32, 3, 8>();
static constexpr auto REPLICATE_4_BIT_TO_8_TABLE = MakeReplicateTable<u32, 4, 8>();
static constexpr auto REPLICATE_5_BIT_TO_8_TABLE = MakeReplicateTable<u32, 5, 8>();
+static constexpr auto REPLICATE_6_BIT_TO_8_TABLE = MakeReplicateTable<u32, 6, 8>();
+static constexpr auto REPLICATE_7_BIT_TO_8_TABLE = MakeReplicateTable<u32, 7, 8>();
+static constexpr auto REPLICATE_8_BIT_TO_8_TABLE = MakeReplicateTable<u32, 8, 8>();
/// Use a precompiled table with the most common usages, if it's not in the expected range, fallback
/// to the runtime implementation
static constexpr u32 FastReplicateTo8(u32 value, u32 num_bits) {
@@ -1316,6 +1395,37 @@ static void ComputeEndpoints(Pixel& ep1, Pixel& ep2, const u32*& colorValues,
#undef READ_INT_VALUES
}
+static void FillVoidExtentLDR(InputBitStream& strm, std::span<u32> outBuf, u32 blockWidth,
+ u32 blockHeight) {
+ // Don't actually care about the void extent, just read the bits...
+ for (s32 i = 0; i < 4; ++i) {
+ strm.ReadBits<13>();
+ }
+
+ // Decode the RGBA components and renormalize them to the range [0, 255]
+ u16 r = static_cast<u16>(strm.ReadBits<16>());
+ u16 g = static_cast<u16>(strm.ReadBits<16>());
+ u16 b = static_cast<u16>(strm.ReadBits<16>());
+ u16 a = static_cast<u16>(strm.ReadBits<16>());
+
+ u32 rgba = (r >> 8) | (g & 0xFF00) | (static_cast<u32>(b) & 0xFF00) << 8 |
+ (static_cast<u32>(a) & 0xFF00) << 16;
+
+ for (u32 j = 0; j < blockHeight; j++) {
+ for (u32 i = 0; i < blockWidth; i++) {
+ outBuf[j * blockWidth + i] = rgba;
+ }
+ }
+}
+
+static void FillError(std::span<u32> outBuf, u32 blockWidth, u32 blockHeight) {
+ for (u32 j = 0; j < blockHeight; j++) {
+ for (u32 i = 0; i < blockWidth; i++) {
+ outBuf[j * blockWidth + i] = 0xFFFF00FF;
+ }
+ }
+}
+
static void DecompressBlock(std::span<const u8, 16> inBuf, const u32 blockWidth,
const u32 blockHeight, std::span<u32, 12 * 12> outBuf) {
InputBitStream strm(inBuf);
diff --git a/src/video_core/textures/astc.h b/src/video_core/textures/astc.h
index 0229ae122..14d2beec0 100644
--- a/src/video_core/textures/astc.h
+++ b/src/video_core/textures/astc.h
@@ -9,117 +9,6 @@
namespace Tegra::Texture::ASTC {
-enum class IntegerEncoding { JustBits, Quint, Trit };
-
-struct IntegerEncodedValue {
- constexpr IntegerEncodedValue() = default;
-
- constexpr IntegerEncodedValue(IntegerEncoding encoding_, u32 num_bits_)
- : encoding{encoding_}, num_bits{num_bits_} {}
-
- constexpr bool MatchesEncoding(const IntegerEncodedValue& other) const {
- return encoding == other.encoding && num_bits == other.num_bits;
- }
-
- // Returns the number of bits required to encode num_vals values.
- u32 GetBitLength(u32 num_vals) const {
- u32 total_bits = num_bits * num_vals;
- if (encoding == IntegerEncoding::Trit) {
- total_bits += (num_vals * 8 + 4) / 5;
- } else if (encoding == IntegerEncoding::Quint) {
- total_bits += (num_vals * 7 + 2) / 3;
- }
- return total_bits;
- }
-
- IntegerEncoding encoding{};
- u32 num_bits = 0;
- u32 bit_value = 0;
- union {
- u32 quint_value = 0;
- u32 trit_value;
- };
-};
-
-// Returns a new instance of this struct that corresponds to the
-// can take no more than mav_value values
-constexpr IntegerEncodedValue CreateEncoding(u32 mav_value) {
- while (mav_value > 0) {
- u32 check = mav_value + 1;
-
- // Is mav_value a power of two?
- if (!(check & (check - 1))) {
- return IntegerEncodedValue(IntegerEncoding::JustBits, std::popcount(mav_value));
- }
-
- // Is mav_value of the type 3*2^n - 1?
- if ((check % 3 == 0) && !((check / 3) & ((check / 3) - 1))) {
- return IntegerEncodedValue(IntegerEncoding::Trit, std::popcount(check / 3 - 1));
- }
-
- // Is mav_value of the type 5*2^n - 1?
- if ((check % 5 == 0) && !((check / 5) & ((check / 5) - 1))) {
- return IntegerEncodedValue(IntegerEncoding::Quint, std::popcount(check / 5 - 1));
- }
-
- // Apparently it can't be represented with a bounded integer sequence...
- // just iterate.
- mav_value--;
- }
- return IntegerEncodedValue(IntegerEncoding::JustBits, 0);
-}
-
-constexpr std::array<IntegerEncodedValue, 256> MakeEncodedValues() {
- std::array<IntegerEncodedValue, 256> encodings{};
- for (std::size_t i = 0; i < encodings.size(); ++i) {
- encodings[i] = CreateEncoding(static_cast<u32>(i));
- }
- return encodings;
-}
-
-constexpr std::array<IntegerEncodedValue, 256> ASTC_ENCODINGS_VALUES = MakeEncodedValues();
-
-// Replicates low num_bits such that [(to_bit - 1):(to_bit - 1 - from_bit)]
-// is the same as [(num_bits - 1):0] and repeats all the way down.
-template <typename IntType>
-constexpr IntType Replicate(IntType val, u32 num_bits, u32 to_bit) {
- if (num_bits == 0 || to_bit == 0) {
- return 0;
- }
- const IntType v = val & static_cast<IntType>((1 << num_bits) - 1);
- IntType res = v;
- u32 reslen = num_bits;
- while (reslen < to_bit) {
- u32 comp = 0;
- if (num_bits > to_bit - reslen) {
- u32 newshift = to_bit - reslen;
- comp = num_bits - newshift;
- num_bits = newshift;
- }
- res = static_cast<IntType>(res << num_bits);
- res = static_cast<IntType>(res | (v >> comp));
- reslen += num_bits;
- }
- return res;
-}
-
-constexpr std::size_t NumReplicateEntries(u32 num_bits) {
- return std::size_t(1) << num_bits;
-}
-
-template <typename IntType, u32 num_bits, u32 to_bit>
-constexpr auto MakeReplicateTable() {
- std::array<IntType, NumReplicateEntries(num_bits)> table{};
- for (IntType value = 0; value < static_cast<IntType>(std::size(table)); ++value) {
- table[value] = Replicate(value, num_bits, to_bit);
- }
- return table;
-}
-
-constexpr auto REPLICATE_6_BIT_TO_8_TABLE = MakeReplicateTable<u32, 6, 8>();
-constexpr auto REPLICATE_7_BIT_TO_8_TABLE = MakeReplicateTable<u32, 7, 8>();
-constexpr auto REPLICATE_8_BIT_TO_8_TABLE = MakeReplicateTable<u32, 8, 8>();
-
void Decompress(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
uint32_t block_width, uint32_t block_height, std::span<uint8_t> output);
diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp
index f1f523ad1..c32ae956a 100644
--- a/src/video_core/textures/decoders.cpp
+++ b/src/video_core/textures/decoders.cpp
@@ -18,9 +18,9 @@
namespace Tegra::Texture {
namespace {
-template <bool TO_LINEAR>
-void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width,
- u32 height, u32 depth, u32 block_height, u32 block_depth, u32 stride_alignment) {
+template <bool TO_LINEAR, u32 BYTES_PER_PIXEL>
+void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 height, u32 depth,
+ u32 block_height, u32 block_depth, u32 stride_alignment) {
// The origin of the transformation can be configured here, leave it as zero as the current API
// doesn't expose it.
static constexpr u32 origin_x = 0;
@@ -28,9 +28,9 @@ void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixe
static constexpr u32 origin_z = 0;
// We can configure here a custom pitch
- // As it's not exposed 'width * bpp' will be the expected pitch.
- const u32 pitch = width * bytes_per_pixel;
- const u32 stride = Common::AlignUpLog2(width, stride_alignment) * bytes_per_pixel;
+ // As it's not exposed 'width * BYTES_PER_PIXEL' will be the expected pitch.
+ const u32 pitch = width * BYTES_PER_PIXEL;
+ const u32 stride = Common::AlignUpLog2(width, stride_alignment) * BYTES_PER_PIXEL;
const u32 gobs_in_x = Common::DivCeilLog2(stride, GOB_SIZE_X_SHIFT);
const u32 block_size = gobs_in_x << (GOB_SIZE_SHIFT + block_height + block_depth);
@@ -54,14 +54,14 @@ void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixe
((block_y & block_height_mask) << GOB_SIZE_SHIFT);
for (u32 column = 0; column < width; ++column) {
- const u32 x = (column + origin_x) * bytes_per_pixel;
+ const u32 x = (column + origin_x) * BYTES_PER_PIXEL;
const u32 offset_x = (x >> GOB_SIZE_X_SHIFT) << x_shift;
const u32 base_swizzled_offset = offset_z + offset_y + offset_x;
const u32 swizzled_offset = base_swizzled_offset + table[x % GOB_SIZE_X];
const u32 unswizzled_offset =
- slice * pitch * height + line * pitch + column * bytes_per_pixel;
+ slice * pitch * height + line * pitch + column * BYTES_PER_PIXEL;
if (const auto offset = (TO_LINEAR ? unswizzled_offset : swizzled_offset);
offset >= input.size()) {
@@ -73,11 +73,45 @@ void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixe
u8* const dst = &output[TO_LINEAR ? swizzled_offset : unswizzled_offset];
const u8* const src = &input[TO_LINEAR ? unswizzled_offset : swizzled_offset];
- std::memcpy(dst, src, bytes_per_pixel);
+
+ std::memcpy(dst, src, BYTES_PER_PIXEL);
}
}
}
}
+
+template <bool TO_LINEAR>
+void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width,
+ u32 height, u32 depth, u32 block_height, u32 block_depth, u32 stride_alignment) {
+ switch (bytes_per_pixel) {
+ case 1:
+ return SwizzleImpl<TO_LINEAR, 1>(output, input, width, height, depth, block_height,
+ block_depth, stride_alignment);
+ case 2:
+ return SwizzleImpl<TO_LINEAR, 2>(output, input, width, height, depth, block_height,
+ block_depth, stride_alignment);
+ case 3:
+ return SwizzleImpl<TO_LINEAR, 3>(output, input, width, height, depth, block_height,
+ block_depth, stride_alignment);
+ case 4:
+ return SwizzleImpl<TO_LINEAR, 4>(output, input, width, height, depth, block_height,
+ block_depth, stride_alignment);
+ case 6:
+ return SwizzleImpl<TO_LINEAR, 6>(output, input, width, height, depth, block_height,
+ block_depth, stride_alignment);
+ case 8:
+ return SwizzleImpl<TO_LINEAR, 8>(output, input, width, height, depth, block_height,
+ block_depth, stride_alignment);
+ case 12:
+ return SwizzleImpl<TO_LINEAR, 12>(output, input, width, height, depth, block_height,
+ block_depth, stride_alignment);
+ case 16:
+ return SwizzleImpl<TO_LINEAR, 16>(output, input, width, height, depth, block_height,
+ block_depth, stride_alignment);
+ default:
+ UNREACHABLE_MSG("Invalid bytes_per_pixel={}", bytes_per_pixel);
+ }
+}
} // Anonymous namespace
void UnswizzleTexture(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel,
generated by cgit v1.2.3 (git 2.25.1) at 2024-10-02 12:42:24 +0200