2 files changed, 48 insertions, 15 deletions

diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp
index 9b6b8527b..913f8ebcb 100644
--- a/src/video_core/textures/decoders.cpp
+++ b/src/video_core/textures/decoders.cpp
@@ -15,6 +15,24 @@
 
 namespace Tegra::Texture {
 namespace {
+template <u32 mask>
+constexpr u32 pdep(u32 value) {
+    u32 result = 0;
+    u32 m = mask;
+    for (u32 bit = 1; m; bit += bit) {
+        if (value & bit)
+            result |= m & -m;
+        m &= m - 1;
+    }
+    return result;
+}
+
+template <u32 mask, u32 incr_amount>
+void incrpdep(u32& value) {
+    constexpr u32 swizzled_incr = pdep<mask>(incr_amount);
+    value = ((value | ~mask) + swizzled_incr) & mask;
+}
+
 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) {
@@ -44,18 +62,20 @@ void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32
                              ((z & block_depth_mask) << (GOB_SIZE_SHIFT + block_height));
         for (u32 line = 0; line < height; ++line) {
             const u32 y = line + origin_y;
-            const auto& table = SWIZZLE_TABLE[y % GOB_SIZE_Y];
+            const u32 swizzled_y = pdep<SWIZZLE_Y_BITS>(y);
 
             const u32 block_y = y >> GOB_SIZE_Y_SHIFT;
             const u32 offset_y = (block_y >> block_height) * block_size +
                                  ((block_y & block_height_mask) << GOB_SIZE_SHIFT);
 
-            for (u32 column = 0; column < width; ++column) {
+            u32 swizzled_x = pdep<SWIZZLE_X_BITS>(origin_x * BYTES_PER_PIXEL);
+            for (u32 column = 0; column < width;
+                 ++column, incrpdep<SWIZZLE_X_BITS, BYTES_PER_PIXEL>(swizzled_x)) {
                 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 swizzled_offset = base_swizzled_offset + (swizzled_x | swizzled_y);
 
                 const u32 unswizzled_offset =
                     slice * pitch * height + line * pitch + column * BYTES_PER_PIXEL;
@@ -103,12 +123,15 @@ void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32
         const u32 gob_address_y =
             (dst_y / (GOB_SIZE_Y * block_height)) * GOB_SIZE * block_height * image_width_in_gobs +
             ((dst_y % (GOB_SIZE_Y * block_height)) / GOB_SIZE_Y) * GOB_SIZE;
-        const auto& table = SWIZZLE_TABLE[dst_y % GOB_SIZE_Y];
-        for (u32 x = 0; x < subrect_width; ++x) {
+
+        const u32 swizzled_y = pdep<SWIZZLE_Y_BITS>(dst_y);
+        u32 swizzled_x = pdep<SWIZZLE_X_BITS>(offset_x * BYTES_PER_PIXEL);
+        for (u32 x = 0; x < subrect_width;
+             ++x, incrpdep<SWIZZLE_X_BITS, BYTES_PER_PIXEL>(swizzled_x)) {
             const u32 dst_x = x + offset_x;
             const u32 gob_address =
                 gob_address_y + (dst_x * BYTES_PER_PIXEL / GOB_SIZE_X) * GOB_SIZE * block_height;
-            const u32 swizzled_offset = gob_address + table[(dst_x * BYTES_PER_PIXEL) % GOB_SIZE_X];
+            const u32 swizzled_offset = gob_address + (swizzled_x | swizzled_y);
             const u32 unswizzled_offset = line * source_pitch + x * BYTES_PER_PIXEL;
 
             const u8* const source_line = unswizzled_data + unswizzled_offset;
@@ -130,16 +153,19 @@ void UnswizzleSubrect(u32 line_length_in, u32 line_count, u32 pitch, u32 width,
 
     for (u32 line = 0; line < line_count; ++line) {
         const u32 src_y = line + origin_y;
-        const auto& table = SWIZZLE_TABLE[src_y % GOB_SIZE_Y];
+        const u32 swizzled_y = pdep<SWIZZLE_Y_BITS>(src_y);
 
         const u32 block_y = src_y >> GOB_SIZE_Y_SHIFT;
         const u32 src_offset_y = (block_y >> block_height) * block_size +
                                  ((block_y & block_height_mask) << GOB_SIZE_SHIFT);
-        for (u32 column = 0; column < line_length_in; ++column) {
+
+        u32 swizzled_x = pdep<SWIZZLE_X_BITS>(origin_x * BYTES_PER_PIXEL);
+        for (u32 column = 0; column < line_length_in;
+             ++column, incrpdep<SWIZZLE_X_BITS, BYTES_PER_PIXEL>(swizzled_x)) {
             const u32 src_x = (column + origin_x) * BYTES_PER_PIXEL;
             const u32 src_offset_x = (src_x >> GOB_SIZE_X_SHIFT) << x_shift;
 
-            const u32 swizzled_offset = src_offset_y + src_offset_x + table[src_x % GOB_SIZE_X];
+            const u32 swizzled_offset = src_offset_y + src_offset_x + (swizzled_x | swizzled_y);
             const u32 unswizzled_offset = line * pitch + column * BYTES_PER_PIXEL;
 
             std::memcpy(output + unswizzled_offset, input + swizzled_offset, BYTES_PER_PIXEL);
@@ -162,13 +188,15 @@ void SwizzleSliceToVoxel(u32 line_length_in, u32 line_count, u32 pitch, u32 widt
     const u32 x_shift = static_cast<u32>(GOB_SIZE_SHIFT) + block_height + block_depth;
 
     for (u32 line = 0; line < line_count; ++line) {
-        const auto& table = SWIZZLE_TABLE[line % GOB_SIZE_Y];
+        const u32 swizzled_y = pdep<SWIZZLE_Y_BITS>(line);
         const u32 block_y = line / GOB_SIZE_Y;
         const u32 dst_offset_y =
             (block_y >> block_height) * block_size + (block_y & block_height_mask) * GOB_SIZE;
-        for (u32 x = 0; x < line_length_in; ++x) {
+
+        u32 swizzled_x = 0;
+        for (u32 x = 0; x < line_length_in; ++x, incrpdep<SWIZZLE_X_BITS, 1>(swizzled_x)) {
             const u32 dst_offset =
-                ((x / GOB_SIZE_X) << x_shift) + dst_offset_y + table[x % GOB_SIZE_X];
+                ((x / GOB_SIZE_X) << x_shift) + dst_offset_y + (swizzled_x | swizzled_y);
             const u32 src_offset = x * BYTES_PER_PIXEL + line * pitch;
             std::memcpy(output + dst_offset, input + src_offset, BYTES_PER_PIXEL);
         }
@@ -267,11 +295,13 @@ void SwizzleKepler(const u32 width, const u32 height, const u32 dst_x, const u32
         const std::size_t gob_address_y =
             (y / (GOB_SIZE_Y * block_height)) * GOB_SIZE * block_height * image_width_in_gobs +
             ((y % (GOB_SIZE_Y * block_height)) / GOB_SIZE_Y) * GOB_SIZE;
-        const auto& table = SWIZZLE_TABLE[y % GOB_SIZE_Y];
-        for (std::size_t x = dst_x; x < width && count < copy_size; ++x) {
+        const u32 swizzled_y = pdep<SWIZZLE_Y_BITS>(static_cast<u32>(y));
+        u32 swizzled_x = pdep<SWIZZLE_X_BITS>(dst_x);
+        for (std::size_t x = dst_x; x < width && count < copy_size;
+             ++x, incrpdep<SWIZZLE_X_BITS, 1>(swizzled_x)) {
             const std::size_t gob_address =
                 gob_address_y + (x / GOB_SIZE_X) * GOB_SIZE * block_height;
-            const std::size_t swizzled_offset = gob_address + table[x % GOB_SIZE_X];
+            const std::size_t swizzled_offset = gob_address + (swizzled_x | swizzled_y);
             const u8* source_line = source_data + count;
             u8* dest_addr = swizzle_data + swizzled_offset;
             count++;
diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h
index 59dfd1621..31a11708f 100644
--- a/src/video_core/textures/decoders.h
+++ b/src/video_core/textures/decoders.h
@@ -20,6 +20,9 @@ constexpr u32 GOB_SIZE_Y_SHIFT = 3;
 constexpr u32 GOB_SIZE_Z_SHIFT = 0;
 constexpr u32 GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_SHIFT;
 
+constexpr u32 SWIZZLE_X_BITS = 0b100101111;
+constexpr u32 SWIZZLE_Y_BITS = 0b011010000;
+
 using SwizzleTable = std::array<std::array<u32, GOB_SIZE_X>, GOB_SIZE_Y>;
 
 /**