From 7ef879b29684510298a13b02942bf4fe960dea77 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Sun, 30 Jul 2023 12:09:02 -0400
Subject: [PATCH 01/33] restrict

---
 src/video_core/host_shaders/astc_decoder.comp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index bf2693559..a622d89a9 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -48,11 +48,11 @@ struct TexelWeightParams {
     bool void_extent_hdr;
 };
 
-layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU32 {
+layout(binding = BINDING_INPUT_BUFFER, std430) readonly restrict buffer InputBufferU32 {
     uvec4 astc_data[];
 };
 
-layout(binding = BINDING_OUTPUT_IMAGE, rgba8) uniform writeonly image2DArray dest_image;
+layout(binding = BINDING_OUTPUT_IMAGE, rgba8) uniform writeonly restrict image2DArray dest_image;
 
 const uint GOB_SIZE_X_SHIFT = 6;
 const uint GOB_SIZE_Y_SHIFT = 3;

From e0c59c7b0b1f082a884414acc5006b243df29e57 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Sun, 30 Jul 2023 12:11:55 -0400
Subject: [PATCH 02/33] skip bits

---
 src/video_core/host_shaders/astc_decoder.comp | 23 +++++++++++--------
 1 file changed, 14 insertions(+), 9 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index a622d89a9..55b23201a 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -360,6 +360,11 @@ uint StreamBits(uint num_bits) {
     return ret;
 }
 
+void SkipBits(uint num_bits) {
+    const int int_bits = int(num_bits);
+    total_bitsread += int_bits;
+}
+
 uint StreamColorBits(uint num_bits) {
     uint ret = 0;
     int int_bits = int(num_bits);
@@ -1071,15 +1076,15 @@ void FillError(ivec3 coord) {
 }
 
 void FillVoidExtentLDR(ivec3 coord) {
-    StreamBits(52);
-    uint r_u = StreamBits(16);
-    uint g_u = StreamBits(16);
-    uint b_u = StreamBits(16);
-    uint a_u = StreamBits(16);
-    float a = float(a_u) / 65535.0f;
-    float r = float(r_u) / 65535.0f;
-    float g = float(g_u) / 65535.0f;
-    float b = float(b_u) / 65535.0f;
+    SkipBits(52);
+    const uint r_u = StreamBits(16);
+    const uint g_u = StreamBits(16);
+    const uint b_u = StreamBits(16);
+    const uint a_u = StreamBits(16);
+    const float a = float(a_u) / 65535.0f;
+    const float r = float(r_u) / 65535.0f;
+    const float g = float(g_u) / 65535.0f;
+    const float b = float(b_u) / 65535.0f;
     for (uint j = 0; j < block_dims.y; j++) {
         for (uint i = 0; i < block_dims.x; i++) {
             imageStore(dest_image, coord + ivec3(i, j, 0), vec4(r, g, b, a));

From 6ff65abd62c4a4f7cc83a79e1ba30d0d99f26ff3 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Sun, 30 Jul 2023 12:13:03 -0400
Subject: [PATCH 03/33] params.max_weight

---
 src/video_core/host_shaders/astc_decoder.comp | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index 55b23201a..a814ef483 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -1058,12 +1058,9 @@ TexelWeightParams DecodeBlockInfo() {
     }
     weight_index -= 2;
     if ((mode_layout != 9) && ((mode & 0x200) != 0)) {
-        const int max_weights[6] = int[6](7, 8, 9, 10, 11, 12);
-        params.max_weight = max_weights[weight_index];
-    } else {
-        const int max_weights[6] = int[6](1, 2, 3, 4, 5, 6);
-        params.max_weight = max_weights[weight_index];
+        weight_index += 6;
     }
+    params.max_weight = weight_index + 1;
     return params;
 }
 

From ac09cc3504fd9c2e256377f75e7ecb187c4bb6f7 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Sun, 30 Jul 2023 12:26:48 -0400
Subject: [PATCH 04/33] weights refactor

---
 src/video_core/host_shaders/astc_decoder.comp | 48 +++++++++----------
 1 file changed, 22 insertions(+), 26 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index a814ef483..b84ddd67d 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -116,8 +116,6 @@ const uint REPLICATE_7_BIT_TO_8_TABLE[128] =
            237, 239, 241, 243, 245, 247, 249, 251, 253, 255);
 
 // Input ASTC texture globals
-uint current_index = 0;
-int bitsread = 0;
 int total_bitsread = 0;
 uvec4 local_buff;
 
@@ -144,13 +142,6 @@ int texel_vector_index = 0;
 
 uint unquantized_texel_weights[2][144];
 
-uint SwizzleOffset(uvec2 pos) {
-    uint x = pos.x;
-    uint y = pos.y;
-    return ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + ((x % 32) / 16) * 32 +
-                          (y % 2) * 16 + (x % 16);
-}
-
 // 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.
 uint Replicate(uint val, uint num_bits, uint to_bit) {
@@ -1224,33 +1215,40 @@ void DecompressBlock(ivec3 coord) {
             uint local_partition = 0;
             if (num_partitions > 1) {
                 local_partition = Select2DPartition(partition_index, i, j, num_partitions,
-                                                     (block_dims.y * block_dims.x) < 32);
+                                                    (block_dims.y * block_dims.x) < 32);
             }
-            vec4 p;
-            uvec4 C0 = ReplicateByteTo16(endpoints[local_partition][0]);
-            uvec4 C1 = ReplicateByteTo16(endpoints[local_partition][1]);
-            uvec4 plane_vec = uvec4(0);
-            uvec4 weight_vec = uvec4(0);
-            for (uint c = 0; c < 4; c++) {
-                if (params.dual_plane && (((plane_index + 1) & 3) == c)) {
-                    plane_vec[c] = 1;
+            const uvec4 C0 = ReplicateByteTo16(endpoints[local_partition][0]);
+            const uvec4 C1 = ReplicateByteTo16(endpoints[local_partition][1]);
+            const uint weight_offset = (j * block_dims.x + i);
+            const uint primary_weight = unquantized_texel_weights[weight_offset][0];
+            uvec4 weight_vec = uvec4(primary_weight);
+            if (params.dual_plane) {
+                const uint secondary_weight = unquantized_texel_weights[weight_offset][1];
+                for (uint c = 0; c < 4; c++) {
+                    const bool is_secondary = ((plane_index + 1u) & 3u) == c;
+                    weight_vec[c] = is_secondary ? secondary_weight : primary_weight;
                 }
-                weight_vec[c] = unquantized_texel_weights[plane_vec[c]][j * block_dims.x + i];
             }
-            vec4 Cf = vec4((C0 * (uvec4(64) - weight_vec) + C1 * weight_vec + uvec4(32)) / 64);
-            p = (Cf / 65535.0);
+            const vec4 Cf =
+                vec4((C0 * (uvec4(64) - weight_vec) + C1 * weight_vec + uvec4(32)) / 64);
+            const vec4 p = (Cf / 65535.0);
             imageStore(dest_image, coord + ivec3(i, j, 0), p.gbar);
         }
     }
 }
 
+
+uint SwizzleOffset(uvec2 pos) {
+    uint x = pos.x;
+    uint y = pos.y;
+    return ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + ((x % 32) / 16) * 32 +
+                          (y % 2) * 16 + (x % 16);
+}
+
 void main() {
     uvec3 pos = gl_GlobalInvocationID;
     pos.x <<= BYTES_PER_BLOCK_LOG2;
-
-    // Read as soon as possible due to its latency
     const uint swizzle = SwizzleOffset(pos.xy);
-
     const uint block_y = pos.y >> GOB_SIZE_Y_SHIFT;
 
     uint offset = 0;
@@ -1264,8 +1262,6 @@ void main() {
     if (any(greaterThanEqual(coord, imageSize(dest_image)))) {
         return;
     }
-    current_index = 0;
-    bitsread = 0;
     local_buff = astc_data[offset / 16];
     DecompressBlock(coord);
 }

From 27c8bb9615039eefc837ade5d91d5733c738f4a0 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Sun, 30 Jul 2023 12:41:52 -0400
Subject: [PATCH 05/33] flattening

---
 src/video_core/host_shaders/astc_decoder.comp | 87 ++++++++++---------
 1 file changed, 44 insertions(+), 43 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index b84ddd67d..f720df6d2 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -140,8 +140,6 @@ int result_index = 0;
 EncodingData texel_vector[144];
 int texel_vector_index = 0;
 
-uint unquantized_texel_weights[2][144];
-
 // 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.
 uint Replicate(uint val, uint num_bits, uint to_bit) {
@@ -879,58 +877,60 @@ uint UnquantizeTexelWeight(EncodingData val) {
     return result;
 }
 
-void UnquantizeTexelWeights(bool dual_plane, uvec2 size) {
-    uint weight_idx = 0;
-    uint unquantized[2][144];
-    uint area = size.x * size.y;
-    for (uint itr = 0; itr < texel_vector_index; itr++) {
-        unquantized[0][weight_idx] = UnquantizeTexelWeight(texel_vector[itr]);
-        if (dual_plane) {
-            ++itr;
-            unquantized[1][weight_idx] = UnquantizeTexelWeight(texel_vector[itr]);
-            if (itr == texel_vector_index) {
-                break;
-            }
-        }
-        if (++weight_idx >= (area))
-            break;
-    }
-
+void UnquantizeTexelWeights(bool is_dual_plane, uvec2 size, out uint unquantized_texel_weights[2 * 144]) {
     const uint Ds = uint((block_dims.x * 0.5f + 1024) / (block_dims.x - 1));
     const uint Dt = uint((block_dims.y * 0.5f + 1024) / (block_dims.y - 1));
-    const uint k_plane_scale = dual_plane ? 2 : 1;
-    for (uint plane = 0; plane < k_plane_scale; plane++) {
+    const uint num_planes = is_dual_plane ? 2 : 1;
+    const uint area = size.x * size.y;
+    const uint loop_count = min(result_index, area * num_planes);
+    uint unquantized[2 * 144];
+    for (uint itr = 0; itr < loop_count; ++itr) {
+        unquantized[itr] = UnquantizeTexelWeight(texel_vector[itr]);
+    }
+    for (uint plane = 0; plane < num_planes; ++plane) {
         for (uint t = 0; t < block_dims.y; t++) {
             for (uint s = 0; s < block_dims.x; s++) {
-                uint cs = Ds * s;
-                uint ct = Dt * t;
-                uint gs = (cs * (size.x - 1) + 32) >> 6;
-                uint gt = (ct * (size.y - 1) + 32) >> 6;
-                uint js = gs >> 4;
-                uint fs = gs & 0xF;
-                uint jt = gt >> 4;
-                uint ft = gt & 0x0F;
-                uint w11 = (fs * ft + 8) >> 4;
-                uint w10 = ft - w11;
-                uint w01 = fs - w11;
-                uint w00 = 16 - fs - ft + w11;
-                uvec4 w = uvec4(w00, w01, w10, w11);
-                uint v0 = jt * size.x + js;
+                const uint cs = Ds * s;
+                const uint ct = Dt * t;
+                const uint gs = (cs * (size.x - 1) + 32) >> 6;
+                const uint gt = (ct * (size.y - 1) + 32) >> 6;
+                const uint js = gs >> 4;
+                const uint fs = gs & 0xF;
+                const uint jt = gt >> 4;
+                const uint ft = gt & 0x0F;
+                const uint w11 = (fs * ft + 8) >> 4;
+                const uint w10 = ft - w11;
+                const uint w01 = fs - w11;
+                const uint w00 = 16 - fs - ft + w11;
+                const uvec4 w = uvec4(w00, w01, w10, w11);
+                const uint v0 = jt * size.x + js;
 
                 uvec4 p = uvec4(0);
+
+#define VectorIndicesFromBase(offset_base)                                                         \
+    const uint offset = is_dual_plane ? 2 * offset_base + plane : offset_base;                     \
+
                 if (v0 < area) {
-                    p.x = unquantized[plane][v0];
+                    const uint offset_base = v0;
+                    VectorIndicesFromBase(offset_base);
+                    p.x = unquantized[offset];
                 }
                 if ((v0 + 1) < (area)) {
-                    p.y = unquantized[plane][v0 + 1];
+                    const uint offset_base = v0 + 1;
+                    VectorIndicesFromBase(offset_base);
+                    p.y = unquantized[offset];
                 }
                 if ((v0 + size.x) < (area)) {
-                    p.z = unquantized[plane][(v0 + size.x)];
+                    const uint offset_base = v0 + size.x;
+                    VectorIndicesFromBase(offset_base);
+                    p.z = unquantized[offset];
                 }
                 if ((v0 + size.x + 1) < (area)) {
-                    p.w = unquantized[plane][(v0 + size.x + 1)];
+                    const uint offset_base = v0 + size.x + 1;
+                    VectorIndicesFromBase(offset_base);
+                    p.w = unquantized[offset];
                 }
-                unquantized_texel_weights[plane][t * block_dims.x + s] = (uint(dot(p, w)) + 8) >> 4;
+                unquantized_texel_weights[plane * 144 + t * block_dims.x + s] = (uint(dot(p, w)) + 8) >> 4;
             }
         }
     }
@@ -1208,7 +1208,8 @@ void DecompressBlock(ivec3 coord) {
     texel_flag = true; // use texel "vector" and bit stream in integer decoding
     DecodeIntegerSequence(params.max_weight, GetNumWeightValues(params.size, params.dual_plane));
 
-    UnquantizeTexelWeights(params.dual_plane, params.size);
+    uint unquantized_texel_weights[2 * 144];
+    UnquantizeTexelWeights(params.dual_plane, params.size, unquantized_texel_weights);
 
     for (uint j = 0; j < block_dims.y; j++) {
         for (uint i = 0; i < block_dims.x; i++) {
@@ -1220,10 +1221,10 @@ void DecompressBlock(ivec3 coord) {
             const uvec4 C0 = ReplicateByteTo16(endpoints[local_partition][0]);
             const uvec4 C1 = ReplicateByteTo16(endpoints[local_partition][1]);
             const uint weight_offset = (j * block_dims.x + i);
-            const uint primary_weight = unquantized_texel_weights[weight_offset][0];
+            const uint primary_weight = unquantized_texel_weights[weight_offset];
             uvec4 weight_vec = uvec4(primary_weight);
             if (params.dual_plane) {
-                const uint secondary_weight = unquantized_texel_weights[weight_offset][1];
+                const uint secondary_weight = unquantized_texel_weights[weight_offset + 144];
                 for (uint c = 0; c < 4; c++) {
                     const bool is_secondary = ((plane_index + 1u) & 3u) == c;
                     weight_vec[c] = is_secondary ? secondary_weight : primary_weight;

From b8ca47e094738689e4e8470b0f1f00d60ee26f4c Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Sun, 30 Jul 2023 12:56:32 -0400
Subject: [PATCH 06/33] EncodingData pack

---
 src/video_core/host_shaders/astc_decoder.comp | 113 +++++++++++-------
 1 file changed, 69 insertions(+), 44 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index f720df6d2..37b502324 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -33,10 +33,7 @@ UNIFORM(6) uint block_height_mask;
 END_PUSH_CONSTANTS
 
 struct EncodingData {
-    uint encoding;
-    uint num_bits;
-    uint bit_value;
-    uint quint_trit_value;
+    uint data;
 };
 
 struct TexelWeightParams {
@@ -66,17 +63,47 @@ const int TRIT = 2;
 
 // ASTC Encodings data, sorted in ascending order based on their BitLength value
 // (see GetBitLength() function)
-EncodingData encoding_values[22] = EncodingData[](
-    EncodingData(JUST_BITS, 0, 0, 0), EncodingData(JUST_BITS, 1, 0, 0), EncodingData(TRIT, 0, 0, 0),
-    EncodingData(JUST_BITS, 2, 0, 0), EncodingData(QUINT, 0, 0, 0), EncodingData(TRIT, 1, 0, 0),
-    EncodingData(JUST_BITS, 3, 0, 0), EncodingData(QUINT, 1, 0, 0), EncodingData(TRIT, 2, 0, 0),
-    EncodingData(JUST_BITS, 4, 0, 0), EncodingData(QUINT, 2, 0, 0), EncodingData(TRIT, 3, 0, 0),
-    EncodingData(JUST_BITS, 5, 0, 0), EncodingData(QUINT, 3, 0, 0), EncodingData(TRIT, 4, 0, 0),
-    EncodingData(JUST_BITS, 6, 0, 0), EncodingData(QUINT, 4, 0, 0), EncodingData(TRIT, 5, 0, 0),
-    EncodingData(JUST_BITS, 7, 0, 0), EncodingData(QUINT, 5, 0, 0), EncodingData(TRIT, 6, 0, 0),
-    EncodingData(JUST_BITS, 8, 0, 0)
+const EncodingData encoding_values[22] = EncodingData[](
+    EncodingData(JUST_BITS),              EncodingData(JUST_BITS | (1u << 8u)), EncodingData(TRIT),                   EncodingData(JUST_BITS | (2u << 8u)),
+    EncodingData(QUINT),                  EncodingData(TRIT | (1u << 8u)),      EncodingData(JUST_BITS | (3u << 8u)), EncodingData(QUINT | (1u << 8u)),
+    EncodingData(TRIT | (2u << 8u)),      EncodingData(JUST_BITS | (4u << 8u)), EncodingData(QUINT | (2u << 8u)),     EncodingData(TRIT | (3u << 8u)),
+    EncodingData(JUST_BITS | (5u << 8u)), EncodingData(QUINT | (3u << 8u)),     EncodingData(TRIT | (4u << 8u)),      EncodingData(JUST_BITS | (6u << 8u)),
+    EncodingData(QUINT | (4u << 8u)),     EncodingData(TRIT | (5u << 8u)),      EncodingData(JUST_BITS | (7u << 8u)), EncodingData(QUINT | (5u << 8u)),
+    EncodingData(TRIT | (6u << 8u)),      EncodingData(JUST_BITS | (8u << 8u))
 );
 
+// EncodingData helpers
+uint Encoding(EncodingData val) {
+    return bitfieldExtract(val.data, 0, 8);
+}
+uint NumBits(EncodingData val) {
+    return bitfieldExtract(val.data, 8, 8);
+}
+uint BitValue(EncodingData val) {
+    return bitfieldExtract(val.data, 16, 8);
+}
+uint QuintTritValue(EncodingData val) {
+    return bitfieldExtract(val.data, 24, 8);
+}
+
+void Encoding(inout EncodingData val, uint v) {
+    val.data = bitfieldInsert(val.data, v, 0, 8);
+}
+void NumBits(inout EncodingData val, uint v) {
+    val.data = bitfieldInsert(val.data, v, 8, 8);
+}
+void BitValue(inout EncodingData val, uint v) {
+    val.data = bitfieldInsert(val.data, v, 16, 8);
+}
+void QuintTritValue(inout EncodingData val, uint v) {
+    val.data = bitfieldInsert(val.data, v, 24, 8);
+}
+
+EncodingData CreateEncodingData(uint encoding, uint num_bits, uint bit_val, uint quint_trit_val) {
+    return EncodingData(((encoding) << 0u) | ((num_bits) << 8u) |
+                        ((bit_val) << 16u) | ((quint_trit_val) << 24u));
+}
+
 // The following constants are expanded variants of the Replicate()
 // function calls corresponding to the following arguments:
 // value: index into the generated table
@@ -379,10 +406,12 @@ void ResultEmplaceBack(EncodingData val) {
 
 // Returns the number of bits required to encode n_vals values.
 uint GetBitLength(uint n_vals, uint encoding_index) {
-    uint total_bits = encoding_values[encoding_index].num_bits * n_vals;
-    if (encoding_values[encoding_index].encoding == TRIT) {
+    const EncodingData encoding_value = encoding_values[encoding_index];
+    const uint encoding = Encoding(encoding_value);
+    uint total_bits = NumBits(encoding_value) * n_vals;
+    if (encoding == TRIT) {
         total_bits += Div5Ceil(n_vals * 8);
-    } else if (encoding_values[encoding_index].encoding == QUINT) {
+    } else if (encoding == QUINT) {
         total_bits += Div3Ceil(n_vals * 7);
     }
     return total_bits;
@@ -451,11 +480,7 @@ void DecodeQuintBlock(uint num_bits) {
         }
     }
     for (uint i = 0; i < 3; i++) {
-        EncodingData val;
-        val.encoding = QUINT;
-        val.num_bits = num_bits;
-        val.bit_value = m[i];
-        val.quint_trit_value = q[i];
+        const EncodingData val = CreateEncodingData(QUINT, num_bits, m[i], q[i]);
         ResultEmplaceBack(val);
     }
 }
@@ -503,30 +528,28 @@ void DecodeTritBlock(uint num_bits) {
         t[0] = (BitsBracket(C, 1) << 1) | (BitsBracket(C, 0) & ~BitsBracket(C, 1));
     }
     for (uint i = 0; i < 5; i++) {
-        EncodingData val;
-        val.encoding = TRIT;
-        val.num_bits = num_bits;
-        val.bit_value = m[i];
-        val.quint_trit_value = t[i];
+        const EncodingData val = CreateEncodingData(TRIT, num_bits, m[i], t[i]);
         ResultEmplaceBack(val);
     }
 }
 
 void DecodeIntegerSequence(uint max_range, uint num_values) {
     EncodingData val = encoding_values[max_range];
+    const uint encoding = Encoding(val);
+    const uint num_bits = NumBits(val);
     uint vals_decoded = 0;
     while (vals_decoded < num_values) {
-        switch (val.encoding) {
+        switch (encoding) {
         case QUINT:
-            DecodeQuintBlock(val.num_bits);
+            DecodeQuintBlock(num_bits);
             vals_decoded += 3;
             break;
         case TRIT:
-            DecodeTritBlock(val.num_bits);
+            DecodeTritBlock(num_bits);
             vals_decoded += 5;
             break;
         case JUST_BITS:
-            val.bit_value = StreamColorBits(val.num_bits);
+            BitValue(val, StreamColorBits(num_bits));
             ResultEmplaceBack(val);
             vals_decoded++;
             break;
@@ -554,17 +577,18 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
         if (out_index >= num_values) {
             break;
         }
-        EncodingData val = result_vector[itr];
-        uint bitlen = val.num_bits;
-        uint bitval = val.bit_value;
+        const EncodingData val = result_vector[itr];
+        const uint encoding = Encoding(val);
+        const uint bitlen = NumBits(val);
+        const uint bitval = BitValue(val);
         uint A = 0, B = 0, C = 0, D = 0;
         A = ReplicateBitTo9((bitval & 1));
-        switch (val.encoding) {
+        switch (encoding) {
         case JUST_BITS:
             color_values[out_index++] = FastReplicateTo8(bitval, bitlen);
             break;
         case TRIT: {
-            D = val.quint_trit_value;
+            D = QuintTritValue(val);
             switch (bitlen) {
             case 1:
                 C = 204;
@@ -603,7 +627,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
             break;
         }
         case QUINT: {
-            D = val.quint_trit_value;
+            D = QuintTritValue(val);
             switch (bitlen) {
             case 1:
                 C = 113;
@@ -636,7 +660,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
             break;
         }
         }
-        if (val.encoding != JUST_BITS) {
+        if (encoding != JUST_BITS) {
             uint T = (D * C) + B;
             T ^= A;
             T = (A & 0x80) | (T >> 2);
@@ -806,17 +830,18 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode) {
 }
 
 uint UnquantizeTexelWeight(EncodingData val) {
-    uint bitval = val.bit_value;
-    uint bitlen = val.num_bits;
-    uint A = ReplicateBitTo7((bitval & 1));
+    const uint encoding = Encoding(val);
+    const uint bitlen = NumBits(val);
+    const uint bitval = BitValue(val);
+    const uint A = ReplicateBitTo7((bitval & 1));
     uint B = 0, C = 0, D = 0;
     uint result = 0;
-    switch (val.encoding) {
+    switch (encoding) {
     case JUST_BITS:
         result = FastReplicateTo6(bitval, bitlen);
         break;
     case TRIT: {
-        D = val.quint_trit_value;
+        D = QuintTritValue(val);
         switch (bitlen) {
         case 0: {
             uint results[3] = {0, 32, 63};
@@ -845,7 +870,7 @@ uint UnquantizeTexelWeight(EncodingData val) {
         break;
     }
     case QUINT: {
-        D = val.quint_trit_value;
+        D = QuintTritValue(val);
         switch (bitlen) {
         case 0: {
             uint results[5] = {0, 16, 32, 47, 63};
@@ -866,7 +891,7 @@ uint UnquantizeTexelWeight(EncodingData val) {
         break;
     }
     }
-    if (val.encoding != JUST_BITS && bitlen > 0) {
+    if (encoding != JUST_BITS && bitlen > 0) {
         result = D * C + B;
         result ^= A;
         result = (A & 0x20) | (result >> 2);

From 0078e5a33822d0e15cc7fab2809e5bc4883cff26 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Sun, 30 Jul 2023 13:04:41 -0400
Subject: [PATCH 07/33] reuse vectors memory

---
 src/video_core/host_shaders/astc_decoder.comp | 50 +++++++------------
 1 file changed, 17 insertions(+), 33 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index 37b502324..4277b0756 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -154,19 +154,10 @@ int color_bitsread = 0;
 uint color_values[32];
 int colvals_index = 0;
 
-// Weight data globals
-uvec4 texel_weight_data;
-int texel_bitsread = 0;
-
-bool texel_flag = false;
-
 // Global "vectors" to be pushed into when decoding
 EncodingData result_vector[144];
 int result_index = 0;
 
-EncodingData texel_vector[144];
-int texel_vector_index = 0;
-
 // 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.
 uint Replicate(uint val, uint num_bits, uint to_bit) {
@@ -382,26 +373,15 @@ void SkipBits(uint num_bits) {
 }
 
 uint StreamColorBits(uint num_bits) {
-    uint ret = 0;
-    int int_bits = int(num_bits);
-    if (texel_flag) {
-        ret = ExtractBits(texel_weight_data, texel_bitsread, int_bits);
-        texel_bitsread += int_bits;
-    } else {
-        ret = ExtractBits(color_endpoint_data, color_bitsread, int_bits);
-        color_bitsread += int_bits;
-    }
+    const int int_bits = int(num_bits);
+    const uint ret = ExtractBits(color_endpoint_data, color_bitsread, int_bits);
+    color_bitsread += int_bits;
     return ret;
 }
 
 void ResultEmplaceBack(EncodingData val) {
-    if (texel_flag) {
-        texel_vector[texel_vector_index] = val;
-        ++texel_vector_index;
-    } else {
-        result_vector[result_index] = val;
-        ++result_index;
-    }
+    result_vector[result_index] = val;
+    ++result_index;
 }
 
 // Returns the number of bits required to encode n_vals values.
@@ -910,7 +890,7 @@ void UnquantizeTexelWeights(bool is_dual_plane, uvec2 size, out uint unquantized
     const uint loop_count = min(result_index, area * num_planes);
     uint unquantized[2 * 144];
     for (uint itr = 0; itr < loop_count; ++itr) {
-        unquantized[itr] = UnquantizeTexelWeight(texel_vector[itr]);
+        unquantized[itr] = UnquantizeTexelWeight(result_vector[itr]);
     }
     for (uint plane = 0; plane < num_planes; ++plane) {
         for (uint t = 0; t < block_dims.y; t++) {
@@ -1215,22 +1195,26 @@ void DecompressBlock(ivec3 coord) {
         ComputeEndpoints(endpoints[i][0], endpoints[i][1], color_endpoint_mode[i]);
     }
 
-    texel_weight_data = local_buff;
-    texel_weight_data = bitfieldReverse(texel_weight_data).wzyx;
+    color_endpoint_data = local_buff;
+    color_endpoint_data = bitfieldReverse(color_endpoint_data).wzyx;
     uint clear_byte_start =
         (GetPackedBitSize(params.size, params.dual_plane, params.max_weight) >> 3) + 1;
 
-    uint byte_insert = ExtractBits(texel_weight_data, int(clear_byte_start - 1) * 8, 8) &
+    uint byte_insert = ExtractBits(color_endpoint_data, int(clear_byte_start - 1) * 8, 8) &
         uint(
             ((1 << (GetPackedBitSize(params.size, params.dual_plane, params.max_weight) % 8)) - 1));
     uint vec_index = (clear_byte_start - 1) >> 2;
-    texel_weight_data[vec_index] =
-        bitfieldInsert(texel_weight_data[vec_index], byte_insert, int((clear_byte_start - 1) % 4) * 8, 8);
+    color_endpoint_data[vec_index] =
+        bitfieldInsert(color_endpoint_data[vec_index], byte_insert, int((clear_byte_start - 1) % 4) * 8, 8);
     for (uint i = clear_byte_start; i < 16; ++i) {
         uint idx = i >> 2;
-        texel_weight_data[idx] = bitfieldInsert(texel_weight_data[idx], 0, int(i % 4) * 8, 8);
+        color_endpoint_data[idx] = bitfieldInsert(color_endpoint_data[idx], 0, int(i % 4) * 8, 8);
     }
-    texel_flag = true; // use texel "vector" and bit stream in integer decoding
+
+    // Re-init vector variables for next decode phase
+    result_index = 0;
+    color_bitsread = 0;
+
     DecodeIntegerSequence(params.max_weight, GetNumWeightValues(params.size, params.dual_plane));
 
     uint unquantized_texel_weights[2 * 144];

From d17a51bc59462b136635515e40019c8ae4e05b5e Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Sun, 30 Jul 2023 13:05:45 -0400
Subject: [PATCH 08/33] extractbits robustness

---
 src/video_core/host_shaders/astc_decoder.comp | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index 4277b0756..f65e1d1b9 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -349,14 +349,17 @@ uint ExtractBits(uvec4 payload, int offset, int bits) {
     if (bits <= 0) {
         return 0;
     }
-    int last_offset = offset + bits - 1;
-    int shifted_offset = offset >> 5;
+    if (bits > 32) {
+        return 0;
+    }
+    const int last_offset = offset + bits - 1;
+    const int shifted_offset = offset >> 5;
     if ((last_offset >> 5) == shifted_offset) {
         return bitfieldExtract(payload[shifted_offset], offset & 31, bits);
     }
-    int first_bits = 32 - (offset & 31);
-    int result_first = int(bitfieldExtract(payload[shifted_offset], offset & 31, first_bits));
-    int result_second = int(bitfieldExtract(payload[shifted_offset + 1], 0, bits - first_bits));
+    const int first_bits = 32 - (offset & 31);
+    const int result_first = int(bitfieldExtract(payload[shifted_offset], offset & 31, first_bits));
+    const int result_second = int(bitfieldExtract(payload[shifted_offset + 1], 0, bits - first_bits));
     return result_first | (result_second << first_bits);
 }
 

From 998246efc27950e3fbebc00a8ca3442dfabb9904 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Sun, 30 Jul 2023 13:12:12 -0400
Subject: [PATCH 09/33] minor redundancy cleanup

---
 src/video_core/host_shaders/astc_decoder.comp | 14 ++------------
 1 file changed, 2 insertions(+), 12 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index f65e1d1b9..90b40c55f 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -418,14 +418,6 @@ uint BitsBracket(uint bits, uint pos) {
 }
 
 uint BitsOp(uint bits, uint start, uint end) {
-    if (start == end) {
-        return BitsBracket(bits, start);
-    } else if (start > end) {
-        uint t = start;
-        start = end;
-        end = t;
-    }
-
     uint mask = (1 << (end - start + 1)) - 1;
     return ((bits >> start) & mask);
 }
@@ -665,10 +657,8 @@ ivec2 BitTransferSigned(int a, int b) {
 }
 
 uvec4 ClampByte(ivec4 color) {
-    for (uint i = 0; i < 4; ++i) {
-        color[i] = (color[i] < 0) ? 0 : ((color[i] > 255) ? 255 : color[i]);
-    }
-    return uvec4(color);
+    const uvec4 clamped = uvec4(clamp(color, 0, 255));
+    return clamped;
 }
 
 ivec4 BlueContract(int a, int r, int g, int b) {

From 5248fa926dd53948b0df4f93c50107dc30ae2305 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Tue, 1 Aug 2023 17:22:03 -0400
Subject: [PATCH 10/33] const, pack result_vector and replicate tables,

undo amd opts
---
 src/video_core/host_shaders/astc_decoder.comp | 491 ++++++++++--------
 1 file changed, 262 insertions(+), 229 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index 90b40c55f..e8801b0ff 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -57,20 +57,40 @@ const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT;
 
 const uint BYTES_PER_BLOCK_LOG2 = 4;
 
-const int JUST_BITS = 0;
-const int QUINT = 1;
-const int TRIT = 2;
+const uint JUST_BITS = 0u;
+const uint QUINT = 1u;
+const uint TRIT = 2u;
 
 // ASTC Encodings data, sorted in ascending order based on their BitLength value
 // (see GetBitLength() function)
-const EncodingData encoding_values[22] = EncodingData[](
-    EncodingData(JUST_BITS),              EncodingData(JUST_BITS | (1u << 8u)), EncodingData(TRIT),                   EncodingData(JUST_BITS | (2u << 8u)),
-    EncodingData(QUINT),                  EncodingData(TRIT | (1u << 8u)),      EncodingData(JUST_BITS | (3u << 8u)), EncodingData(QUINT | (1u << 8u)),
-    EncodingData(TRIT | (2u << 8u)),      EncodingData(JUST_BITS | (4u << 8u)), EncodingData(QUINT | (2u << 8u)),     EncodingData(TRIT | (3u << 8u)),
-    EncodingData(JUST_BITS | (5u << 8u)), EncodingData(QUINT | (3u << 8u)),     EncodingData(TRIT | (4u << 8u)),      EncodingData(JUST_BITS | (6u << 8u)),
-    EncodingData(QUINT | (4u << 8u)),     EncodingData(TRIT | (5u << 8u)),      EncodingData(JUST_BITS | (7u << 8u)), EncodingData(QUINT | (5u << 8u)),
-    EncodingData(TRIT | (6u << 8u)),      EncodingData(JUST_BITS | (8u << 8u))
-);
+const uvec4 encoding_values[6] = uvec4[](
+    uvec4((JUST_BITS), (JUST_BITS | (1u << 8u)), (TRIT), (JUST_BITS | (2u << 8u))),
+    uvec4((QUINT), (TRIT | (1u << 8u)), (JUST_BITS | (3u << 8u)), (QUINT | (1u << 8u))),
+    uvec4((TRIT | (2u << 8u)), (JUST_BITS | (4u << 8u)), (QUINT | (2u << 8u)), (TRIT | (3u << 8u))),
+    uvec4((JUST_BITS | (5u << 8u)), (QUINT | (3u << 8u)), (TRIT | (4u << 8u)), (JUST_BITS | (6u << 8u))),
+    uvec4((QUINT | (4u << 8u)), (TRIT | (5u << 8u)), (JUST_BITS | (7u << 8u)), (QUINT | (5u << 8u))),
+    uvec4((TRIT | (6u << 8u)), (JUST_BITS | (8u << 8u)), 0u, 0u));
+
+// Input ASTC texture globals
+int total_bitsread = 0;
+uvec4 local_buff;
+
+// Color data globals
+uvec4 color_endpoint_data;
+int color_bitsread = 0;
+
+// Global "vector" to be pushed into when decoding
+// At most will require BLOCK_WIDTH x BLOCK_HEIGHT in single plane mode
+// At most will require BLOCK_WIDTH x BLOCK_HEIGHT x 2 in dual plane mode
+// So the maximum would be 144 (12 x 12) elements, x 2 for two planes
+#define DIVCEIL(number, divisor) (number + divisor - 1) / divisor
+#define ARRAY_NUM_ELEMENTS 144
+#define VECTOR_ARRAY_SIZE DIVCEIL(ARRAY_NUM_ELEMENTS * 2, 4)
+uvec4 result_vector[VECTOR_ARRAY_SIZE];
+
+int result_index = 0;
+uint result_vector_max_index;
+bool result_limit_reached = false;
 
 // EncodingData helpers
 uint Encoding(EncodingData val) {
@@ -104,78 +124,17 @@ EncodingData CreateEncodingData(uint encoding, uint num_bits, uint bit_val, uint
                         ((bit_val) << 16u) | ((quint_trit_val) << 24u));
 }
 
-// The following constants are expanded variants of the Replicate()
-// function calls corresponding to the following arguments:
-// value: index into the generated table
-// num_bits: the after "REPLICATE" in the table name. i.e. 4 is num_bits in REPLICATE_4.
-// to_bit: the integer after "TO_"
-const uint REPLICATE_BIT_TO_7_TABLE[2] = uint[](0, 127);
-const uint REPLICATE_1_BIT_TO_9_TABLE[2] = uint[](0, 511);
 
-const uint REPLICATE_1_BIT_TO_8_TABLE[2] = uint[](0, 255);
-const uint REPLICATE_2_BIT_TO_8_TABLE[4] = uint[](0, 85, 170, 255);
-const uint REPLICATE_3_BIT_TO_8_TABLE[8] = uint[](0, 36, 73, 109, 146, 182, 219, 255);
-const uint REPLICATE_4_BIT_TO_8_TABLE[16] =
-    uint[](0, 17, 34, 51, 68, 85, 102, 119, 136, 153, 170, 187, 204, 221, 238, 255);
-const uint REPLICATE_5_BIT_TO_8_TABLE[32] =
-    uint[](0, 8, 16, 24, 33, 41, 49, 57, 66, 74, 82, 90, 99, 107, 115, 123, 132, 140, 148, 156, 165,
-           173, 181, 189, 198, 206, 214, 222, 231, 239, 247, 255);
-const uint REPLICATE_1_BIT_TO_6_TABLE[2] = uint[](0, 63);
-const uint REPLICATE_2_BIT_TO_6_TABLE[4] = uint[](0, 21, 42, 63);
-const uint REPLICATE_3_BIT_TO_6_TABLE[8] = uint[](0, 9, 18, 27, 36, 45, 54, 63);
-const uint REPLICATE_4_BIT_TO_6_TABLE[16] =
-    uint[](0, 4, 8, 12, 17, 21, 25, 29, 34, 38, 42, 46, 51, 55, 59, 63);
-const uint REPLICATE_5_BIT_TO_6_TABLE[32] =
-    uint[](0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30, 33, 35, 37, 39, 41, 43, 45,
-           47, 49, 51, 53, 55, 57, 59, 61, 63);
-const uint REPLICATE_6_BIT_TO_8_TABLE[64] =
-    uint[](0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60, 65, 69, 73, 77, 81, 85, 89,
-           93, 97, 101, 105, 109, 113, 117, 121, 125, 130, 134, 138, 142, 146, 150, 154, 158, 162,
-           166, 170, 174, 178, 182, 186, 190, 195, 199, 203, 207, 211, 215, 219, 223, 227, 231, 235,
-           239, 243, 247, 251, 255);
-const uint REPLICATE_7_BIT_TO_8_TABLE[128] =
-    uint[](0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30, 32, 34, 36, 38, 40, 42, 44,
-           46, 48, 50, 52, 54, 56, 58, 60, 62, 64, 66, 68, 70, 72, 74, 76, 78, 80, 82, 84, 86, 88,
-           90, 92, 94, 96, 98, 100, 102, 104, 106, 108, 110, 112, 114, 116, 118, 120, 122, 124, 126,
-           129, 131, 133, 135, 137, 139, 141, 143, 145, 147, 149, 151, 153, 155, 157, 159, 161, 163,
-           165, 167, 169, 171, 173, 175, 177, 179, 181, 183, 185, 187, 189, 191, 193, 195, 197, 199,
-           201, 203, 205, 207, 209, 211, 213, 215, 217, 219, 221, 223, 225, 227, 229, 231, 233, 235,
-           237, 239, 241, 243, 245, 247, 249, 251, 253, 255);
-
-// Input ASTC texture globals
-int total_bitsread = 0;
-uvec4 local_buff;
-
-// Color data globals
-uvec4 color_endpoint_data;
-int color_bitsread = 0;
-
-// Four values, two endpoints, four maximum partitions
-uint color_values[32];
-int colvals_index = 0;
-
-// Global "vectors" to be pushed into when decoding
-EncodingData result_vector[144];
-int result_index = 0;
-
-// 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.
-uint Replicate(uint val, uint num_bits, uint to_bit) {
-    const uint v = val & uint((1 << num_bits) - 1);
-    uint res = v;
-    uint reslen = num_bits;
-    while (reslen < to_bit) {
-        uint comp = 0;
-        if (num_bits > to_bit - reslen) {
-            uint newshift = to_bit - reslen;
-            comp = num_bits - newshift;
-            num_bits = newshift;
-        }
-        res = uint(res << num_bits);
-        res = uint(res | (v >> comp));
-        reslen += num_bits;
+void ResultEmplaceBack(EncodingData val) {
+    if (result_index >= result_vector_max_index) {
+        // Alert callers to avoid decoding more than needed by this phase
+        result_limit_reached = true;
+        return;
     }
-    return res;
+    const uint array_index = result_index / 4;
+    const uint vector_index = result_index % 4;
+    result_vector[array_index][vector_index] = val.data;
+    ++result_index;
 }
 
 uvec4 ReplicateByteTo16(uvec4 value) {
@@ -183,64 +142,105 @@ uvec4 ReplicateByteTo16(uvec4 value) {
 }
 
 uint ReplicateBitTo7(uint value) {
-    return REPLICATE_BIT_TO_7_TABLE[value];
+    return value * 127;
 }
 
 uint ReplicateBitTo9(uint value) {
-    return REPLICATE_1_BIT_TO_9_TABLE[value];
-}
-
-uint FastReplicate(uint value, uint num_bits, uint to_bit) {
-    if (num_bits == 0) {
-        return 0;
-    }
-    if (num_bits == to_bit) {
-        return value;
-    }
-    if (to_bit == 6) {
-        switch (num_bits) {
-        case 1:
-            return REPLICATE_1_BIT_TO_6_TABLE[value];
-        case 2:
-            return REPLICATE_2_BIT_TO_6_TABLE[value];
-        case 3:
-            return REPLICATE_3_BIT_TO_6_TABLE[value];
-        case 4:
-            return REPLICATE_4_BIT_TO_6_TABLE[value];
-        case 5:
-            return REPLICATE_5_BIT_TO_6_TABLE[value];
-        default:
-            break;
-        }
-    } else { /* if (to_bit == 8) */
-        switch (num_bits) {
-        case 1:
-            return REPLICATE_1_BIT_TO_8_TABLE[value];
-        case 2:
-            return REPLICATE_2_BIT_TO_8_TABLE[value];
-        case 3:
-            return REPLICATE_3_BIT_TO_8_TABLE[value];
-        case 4:
-            return REPLICATE_4_BIT_TO_8_TABLE[value];
-        case 5:
-            return REPLICATE_5_BIT_TO_8_TABLE[value];
-        case 6:
-            return REPLICATE_6_BIT_TO_8_TABLE[value];
-        case 7:
-            return REPLICATE_7_BIT_TO_8_TABLE[value];
-        default:
-            break;
-        }
-    }
-    return Replicate(value, num_bits, to_bit);
+    return value * 511;
 }
 
 uint FastReplicateTo8(uint value, uint num_bits) {
-    return FastReplicate(value, num_bits, 8);
+    if (value == 0) {
+        return 0;
+    }
+    const uint array_index = value / 4;
+    const uint vector_index = value % 4;
+    switch (num_bits) {
+    case 1:
+        return 255;
+    case 2: {
+        const uvec4 REPLICATE_2_BIT_TO_8_TABLE = (uvec4(0, 85, 170, 255));
+        return REPLICATE_2_BIT_TO_8_TABLE[vector_index];
+    }
+    case 3: {
+        const uvec4 REPLICATE_3_BIT_TO_8_TABLE[2] =
+            uvec4[](uvec4(0, 36, 73, 109), uvec4(146, 182, 219, 255));
+        return REPLICATE_3_BIT_TO_8_TABLE[array_index][vector_index];
+    }
+    case 4: {
+        const uvec4 REPLICATE_4_BIT_TO_8_TABLE[4] =
+            uvec4[](uvec4(0, 17, 34, 51), uvec4(68, 85, 102, 119), uvec4(136, 153, 170, 187),
+                    uvec4(204, 221, 238, 255));
+        return REPLICATE_4_BIT_TO_8_TABLE[array_index][vector_index];
+    }
+    case 5: {
+        const uvec4 REPLICATE_5_BIT_TO_8_TABLE[8] =
+            uvec4[](uvec4(0, 8, 16, 24), uvec4(33, 41, 49, 57), uvec4(66, 74, 82, 90),
+                    uvec4(99, 107, 115, 123), uvec4(132, 140, 148, 156), uvec4(165, 173, 181, 189),
+                    uvec4(198, 206, 214, 222), uvec4(231, 239, 247, 255));
+        return REPLICATE_5_BIT_TO_8_TABLE[array_index][vector_index];
+    }
+    case 6: {
+        const uvec4 REPLICATE_6_BIT_TO_8_TABLE[16] = uvec4[](
+            uvec4(0, 4, 8, 12), uvec4(16, 20, 24, 28), uvec4(32, 36, 40, 44), uvec4(48, 52, 56, 60),
+            uvec4(65, 69, 73, 77), uvec4(81, 85, 89, 93), uvec4(97, 101, 105, 109),
+            uvec4(113, 117, 121, 125), uvec4(130, 134, 138, 142), uvec4(146, 150, 154, 158),
+            uvec4(162, 166, 170, 174), uvec4(178, 182, 186, 190), uvec4(195, 199, 203, 207),
+            uvec4(211, 215, 219, 223), uvec4(227, 231, 235, 239), uvec4(243, 247, 251, 255));
+        return REPLICATE_6_BIT_TO_8_TABLE[array_index][vector_index];
+    }
+    case 7: {
+        const uvec4 REPLICATE_7_BIT_TO_8_TABLE[32] =
+            uvec4[](uvec4(0, 2, 4, 6), uvec4(8, 10, 12, 14), uvec4(16, 18, 20, 22),
+                    uvec4(24, 26, 28, 30), uvec4(32, 34, 36, 38), uvec4(40, 42, 44, 46),
+                    uvec4(48, 50, 52, 54), uvec4(56, 58, 60, 62), uvec4(64, 66, 68, 70),
+                    uvec4(72, 74, 76, 78), uvec4(80, 82, 84, 86), uvec4(88, 90, 92, 94),
+                    uvec4(96, 98, 100, 102), uvec4(104, 106, 108, 110), uvec4(112, 114, 116, 118),
+                    uvec4(120, 122, 124, 126), uvec4(129, 131, 133, 135), uvec4(137, 139, 141, 143),
+                    uvec4(145, 147, 149, 151), uvec4(153, 155, 157, 159), uvec4(161, 163, 165, 167),
+                    uvec4(169, 171, 173, 175), uvec4(177, 179, 181, 183), uvec4(185, 187, 189, 191),
+                    uvec4(193, 195, 197, 199), uvec4(201, 203, 205, 207), uvec4(209, 211, 213, 215),
+                    uvec4(217, 219, 221, 223), uvec4(225, 227, 229, 231), uvec4(233, 235, 237, 239),
+                    uvec4(241, 243, 245, 247), uvec4(249, 251, 253, 255));
+        return REPLICATE_7_BIT_TO_8_TABLE[array_index][vector_index];
+    }
+    }
+    return value;
 }
 
 uint FastReplicateTo6(uint value, uint num_bits) {
-    return FastReplicate(value, num_bits, 6);
+    if (value == 0) {
+        return 0;
+    }
+    const uint array_index = value / 4;
+    const uint vector_index = value % 4;
+    switch (num_bits) {
+    case 1:
+        return 63;
+    case 2: {
+        const uvec4 REPLICATE_2_BIT_TO_6_TABLE = uvec4(0, 21, 42, 63);
+        return REPLICATE_2_BIT_TO_6_TABLE[vector_index];
+    }
+    case 3: {
+        const uvec4 REPLICATE_3_BIT_TO_6_TABLE[2] =
+            uvec4[](uvec4(0, 9, 18, 27), uvec4(36, 45, 54, 63));
+        return REPLICATE_3_BIT_TO_6_TABLE[array_index][vector_index];
+    }
+    case 4: {
+        const uvec4 REPLICATE_4_BIT_TO_6_TABLE[4] =
+            uvec4[](uvec4(0, 4, 8, 12), uvec4(17, 21, 25, 29), uvec4(34, 38, 42, 46),
+                    uvec4(51, 55, 59, 63));
+        return REPLICATE_4_BIT_TO_6_TABLE[array_index][vector_index];
+    }
+    case 5: {
+        const uvec4 REPLICATE_5_BIT_TO_6_TABLE[8] =
+            uvec4[](uvec4(0, 2, 4, 6), uvec4(8, 10, 12, 14), uvec4(16, 18, 20, 22),
+                    uvec4(24, 26, 28, 30), uvec4(33, 35, 37, 39), uvec4(41, 43, 45, 47),
+                    uvec4(49, 51, 53, 55), uvec4(57, 59, 61, 63));
+        return REPLICATE_5_BIT_TO_6_TABLE[array_index][vector_index];
+    }
+    }
+    return value;
 }
 
 uint Div3Floor(uint v) {
@@ -281,7 +281,7 @@ uint Select2DPartition(uint seed, uint x, uint y, uint partition_count, bool sma
 
     seed += (partition_count - 1) * 1024;
 
-    uint rnum = Hash52(uint(seed));
+    const uint rnum = Hash52(uint(seed));
     uint seed1 = uint(rnum & 0xF);
     uint seed2 = uint((rnum >> 4) & 0xF);
     uint seed3 = uint((rnum >> 8) & 0xF);
@@ -364,8 +364,8 @@ uint ExtractBits(uvec4 payload, int offset, int bits) {
 }
 
 uint StreamBits(uint num_bits) {
-    int int_bits = int(num_bits);
-    uint ret = ExtractBits(local_buff, total_bitsread, int_bits);
+    const int int_bits = int(num_bits);
+    const uint ret = ExtractBits(local_buff, total_bitsread, int_bits);
     total_bitsread += int_bits;
     return ret;
 }
@@ -382,14 +382,18 @@ uint StreamColorBits(uint num_bits) {
     return ret;
 }
 
-void ResultEmplaceBack(EncodingData val) {
-    result_vector[result_index] = val;
-    ++result_index;
+EncodingData GetEncodingFromVector(uint index) {
+    const uint array_index = index / 4;
+    const uint vector_index = index % 4;
+
+    const uint data = result_vector[array_index][vector_index];
+    return EncodingData(data);
 }
 
 // Returns the number of bits required to encode n_vals values.
 uint GetBitLength(uint n_vals, uint encoding_index) {
-    const EncodingData encoding_value = encoding_values[encoding_index];
+    const EncodingData encoding_value =
+        EncodingData(encoding_values[encoding_index / 4][encoding_index % 4]);
     const uint encoding = Encoding(encoding_value);
     uint total_bits = NumBits(encoding_value) * n_vals;
     if (encoding == TRIT) {
@@ -409,7 +413,7 @@ uint GetNumWeightValues(uvec2 size, bool dual_plane) {
 }
 
 uint GetPackedBitSize(uvec2 size, bool dual_plane, uint max_weight) {
-    uint n_vals = GetNumWeightValues(size, dual_plane);
+    const uint n_vals = GetNumWeightValues(size, dual_plane);
     return GetBitLength(n_vals, max_weight);
 }
 
@@ -418,13 +422,13 @@ uint BitsBracket(uint bits, uint pos) {
 }
 
 uint BitsOp(uint bits, uint start, uint end) {
-    uint mask = (1 << (end - start + 1)) - 1;
+    const uint mask = (1 << (end - start + 1)) - 1;
     return ((bits >> start) & mask);
 }
 
 void DecodeQuintBlock(uint num_bits) {
-    uint m[3];
-    uint q[3];
+    uvec3 m;
+    uvec3 q;
     uint Q;
     m[0] = StreamColorBits(num_bits);
     Q = StreamColorBits(3);
@@ -433,25 +437,25 @@ void DecodeQuintBlock(uint num_bits) {
     m[2] = StreamColorBits(num_bits);
     Q |= StreamColorBits(2) << 5;
     if (BitsOp(Q, 1, 2) == 3 && BitsOp(Q, 5, 6) == 0) {
-        q[0] = 4;
-        q[1] = 4;
-        q[2] = (BitsBracket(Q, 0) << 2) | ((BitsBracket(Q, 4) & ~BitsBracket(Q, 0)) << 1) |
-               (BitsBracket(Q, 3) & ~BitsBracket(Q, 0));
+        q.x = 4;
+        q.y = 4;
+        q.z = (BitsBracket(Q, 0) << 2) | ((BitsBracket(Q, 4) & ~BitsBracket(Q, 0)) << 1) |
+              (BitsBracket(Q, 3) & ~BitsBracket(Q, 0));
     } else {
         uint C = 0;
         if (BitsOp(Q, 1, 2) == 3) {
-            q[2] = 4;
+            q.z = 4;
             C = (BitsOp(Q, 3, 4) << 3) | ((~BitsOp(Q, 5, 6) & 3) << 1) | BitsBracket(Q, 0);
         } else {
-            q[2] = BitsOp(Q, 5, 6);
+            q.z = BitsOp(Q, 5, 6);
             C = BitsOp(Q, 0, 4);
         }
         if (BitsOp(C, 0, 2) == 5) {
-            q[1] = 4;
-            q[0] = BitsOp(C, 3, 4);
+            q.y = 4;
+            q.x = BitsOp(C, 3, 4);
         } else {
-            q[1] = BitsOp(C, 3, 4);
-            q[0] = BitsOp(C, 0, 2);
+            q.y = BitsOp(C, 3, 4);
+            q.x = BitsOp(C, 0, 2);
         }
     }
     for (uint i = 0; i < 3; i++) {
@@ -509,11 +513,11 @@ void DecodeTritBlock(uint num_bits) {
 }
 
 void DecodeIntegerSequence(uint max_range, uint num_values) {
-    EncodingData val = encoding_values[max_range];
+    EncodingData val = EncodingData(encoding_values[max_range / 4][max_range % 4]);
     const uint encoding = Encoding(val);
     const uint num_bits = NumBits(val);
     uint vals_decoded = 0;
-    while (vals_decoded < num_values) {
+    while (vals_decoded < num_values && !result_limit_reached) {
         switch (encoding) {
         case QUINT:
             DecodeQuintBlock(num_bits);
@@ -532,7 +536,8 @@ void DecodeIntegerSequence(uint max_range, uint num_values) {
     }
 }
 
-void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
+void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits,
+                       out uvec4 color_values[8]) {
     uint num_values = 0;
     for (uint i = 0; i < num_partitions; i++) {
         num_values += ((modes[i] >> 2) + 1) << 1;
@@ -540,8 +545,8 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
     // Find the largest encoding that's within color_data_bits
     // TODO(ameerj): profile with binary search
     int range = 0;
-    while (++range < encoding_values.length()) {
-        uint bit_length = GetBitLength(num_values, range);
+    while (++range < ((encoding_values.length() * 4) - 2)) {
+        const uint bit_length = GetBitLength(num_values, range);
         if (bit_length > color_data_bits) {
             break;
         }
@@ -552,7 +557,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
         if (out_index >= num_values) {
             break;
         }
-        const EncodingData val = result_vector[itr];
+        const EncodingData val = GetEncodingFromVector(itr);
         const uint encoding = Encoding(val);
         const uint bitlen = NumBits(val);
         const uint bitval = BitValue(val);
@@ -560,7 +565,8 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
         A = ReplicateBitTo9((bitval & 1));
         switch (encoding) {
         case JUST_BITS:
-            color_values[out_index++] = FastReplicateTo8(bitval, bitlen);
+            color_values[out_index / 4][out_index % 4] = FastReplicateTo8(bitval, bitlen);
+            ++out_index;
             break;
         case TRIT: {
             D = QuintTritValue(val);
@@ -570,31 +576,31 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
                 break;
             case 2: {
                 C = 93;
-                uint b = (bitval >> 1) & 1;
+                const uint b = (bitval >> 1) & 1;
                 B = (b << 8) | (b << 4) | (b << 2) | (b << 1);
                 break;
             }
             case 3: {
                 C = 44;
-                uint cb = (bitval >> 1) & 3;
+                const uint cb = (bitval >> 1) & 3;
                 B = (cb << 7) | (cb << 2) | cb;
                 break;
             }
             case 4: {
                 C = 22;
-                uint dcb = (bitval >> 1) & 7;
+                const uint dcb = (bitval >> 1) & 7;
                 B = (dcb << 6) | dcb;
                 break;
             }
             case 5: {
                 C = 11;
-                uint edcb = (bitval >> 1) & 0xF;
+                const uint edcb = (bitval >> 1) & 0xF;
                 B = (edcb << 5) | (edcb >> 2);
                 break;
             }
             case 6: {
                 C = 5;
-                uint fedcb = (bitval >> 1) & 0x1F;
+                const uint fedcb = (bitval >> 1) & 0x1F;
                 B = (fedcb << 4) | (fedcb >> 4);
                 break;
             }
@@ -609,25 +615,25 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
                 break;
             case 2: {
                 C = 54;
-                uint b = (bitval >> 1) & 1;
+                const uint b = (bitval >> 1) & 1;
                 B = (b << 8) | (b << 3) | (b << 2);
                 break;
             }
             case 3: {
                 C = 26;
-                uint cb = (bitval >> 1) & 3;
+                const uint cb = (bitval >> 1) & 3;
                 B = (cb << 7) | (cb << 1) | (cb >> 1);
                 break;
             }
             case 4: {
                 C = 13;
-                uint dcb = (bitval >> 1) & 7;
+                const uint dcb = (bitval >> 1) & 7;
                 B = (dcb << 6) | (dcb >> 1);
                 break;
             }
             case 5: {
                 C = 6;
-                uint edcb = (bitval >> 1) & 0xF;
+                const uint edcb = (bitval >> 1) & 0xF;
                 B = (edcb << 5) | (edcb >> 3);
                 break;
             }
@@ -639,7 +645,8 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
             uint T = (D * C) + B;
             T ^= A;
             T = (A & 0x80) | (T >> 2);
-            color_values[out_index++] = T;
+            color_values[out_index / 4][out_index % 4] = T;
+            ++out_index;
         }
     }
 }
@@ -657,25 +664,30 @@ ivec2 BitTransferSigned(int a, int b) {
 }
 
 uvec4 ClampByte(ivec4 color) {
-    const uvec4 clamped = uvec4(clamp(color, 0, 255));
-    return clamped;
+    for (uint i = 0; i < 4; ++i) {
+        color[i] = clamp(color[i], 0, 255);
+    }
+    return uvec4(color);
 }
 
 ivec4 BlueContract(int a, int r, int g, int b) {
     return ivec4(a, (r + b) >> 1, (g + b) >> 1, b);
 }
 
-void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode) {
+void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode,
+                      in uvec4 color_values[8], inout uint colvals_index) {
 #define READ_UINT_VALUES(N)                                                                        \
     uint v[N];                                                                                     \
     for (uint i = 0; i < N; i++) {                                                                 \
-        v[i] = color_values[colvals_index++];                                                      \
+        v[i] = color_values[colvals_index / 4][colvals_index % 4];                                 \
+        ++colvals_index;                                                                           \
     }
 
 #define READ_INT_VALUES(N)                                                                         \
     int v[N];                                                                                      \
     for (uint i = 0; i < N; i++) {                                                                 \
-        v[i] = int(color_values[colvals_index++]);                                                 \
+        v[i] = int(color_values[colvals_index / 4][colvals_index % 4]);                            \
+        ++colvals_index;                                                                           \
     }
 
     switch (color_endpoint_mode) {
@@ -687,8 +699,8 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode) {
     }
     case 1: {
         READ_UINT_VALUES(2)
-        uint L0 = (v[0] >> 2) | (v[1] & 0xC0);
-        uint L1 = min(L0 + (v[1] & 0x3F), 0xFFU);
+        const uint L0 = (v[0] >> 2) | (v[1] & 0xC0);
+        const uint L1 = min(L0 + (v[1] & 0x3F), 0xFFU);
         ep1 = uvec4(0xFF, L0, L0, L0);
         ep2 = uvec4(0xFF, L1, L1, L1);
         break;
@@ -817,7 +829,7 @@ uint UnquantizeTexelWeight(EncodingData val) {
         D = QuintTritValue(val);
         switch (bitlen) {
         case 0: {
-            uint results[3] = {0, 32, 63};
+            const uint results[3] = {0, 32, 63};
             result = results[D];
             break;
         }
@@ -827,13 +839,13 @@ uint UnquantizeTexelWeight(EncodingData val) {
         }
         case 2: {
             C = 23;
-            uint b = (bitval >> 1) & 1;
+            const uint b = (bitval >> 1) & 1;
             B = (b << 6) | (b << 2) | b;
             break;
         }
         case 3: {
             C = 11;
-            uint cb = (bitval >> 1) & 3;
+            const uint cb = (bitval >> 1) & 3;
             B = (cb << 5) | cb;
             break;
         }
@@ -846,7 +858,7 @@ uint UnquantizeTexelWeight(EncodingData val) {
         D = QuintTritValue(val);
         switch (bitlen) {
         case 0: {
-            uint results[5] = {0, 16, 32, 47, 63};
+            const uint results[5] = {0, 16, 32, 47, 63};
             result = results[D];
             break;
         }
@@ -856,7 +868,7 @@ uint UnquantizeTexelWeight(EncodingData val) {
         }
         case 2: {
             C = 13;
-            uint b = (bitval >> 1) & 1;
+            const uint b = (bitval >> 1) & 1;
             B = (b << 6) | (b << 1);
             break;
         }
@@ -875,15 +887,18 @@ uint UnquantizeTexelWeight(EncodingData val) {
     return result;
 }
 
-void UnquantizeTexelWeights(bool is_dual_plane, uvec2 size, out uint unquantized_texel_weights[2 * 144]) {
+void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane,
+                            out uvec4 unquantized_texel_weights[VECTOR_ARRAY_SIZE]) {
     const uint Ds = uint((block_dims.x * 0.5f + 1024) / (block_dims.x - 1));
     const uint Dt = uint((block_dims.y * 0.5f + 1024) / (block_dims.y - 1));
     const uint num_planes = is_dual_plane ? 2 : 1;
     const uint area = size.x * size.y;
     const uint loop_count = min(result_index, area * num_planes);
-    uint unquantized[2 * 144];
     for (uint itr = 0; itr < loop_count; ++itr) {
-        unquantized[itr] = UnquantizeTexelWeight(result_vector[itr]);
+        const uint array_index = itr / 4;
+        const uint vector_index = itr % 4;
+        result_vector[array_index][vector_index] =
+            UnquantizeTexelWeight(GetEncodingFromVector(itr));
     }
     for (uint plane = 0; plane < num_planes; ++plane) {
         for (uint t = 0; t < block_dims.y; t++) {
@@ -907,28 +922,33 @@ void UnquantizeTexelWeights(bool is_dual_plane, uvec2 size, out uint unquantized
 
 #define VectorIndicesFromBase(offset_base)                                                         \
     const uint offset = is_dual_plane ? 2 * offset_base + plane : offset_base;                     \
+    const uint array_index = offset / 4;                                                           \
+    const uint vector_index = offset % 4;
 
                 if (v0 < area) {
                     const uint offset_base = v0;
                     VectorIndicesFromBase(offset_base);
-                    p.x = unquantized[offset];
+                    p.x = result_vector[array_index][vector_index];
                 }
                 if ((v0 + 1) < (area)) {
                     const uint offset_base = v0 + 1;
                     VectorIndicesFromBase(offset_base);
-                    p.y = unquantized[offset];
+                    p.y = result_vector[array_index][vector_index];
                 }
                 if ((v0 + size.x) < (area)) {
                     const uint offset_base = v0 + size.x;
                     VectorIndicesFromBase(offset_base);
-                    p.z = unquantized[offset];
+                    p.z = result_vector[array_index][vector_index];
                 }
                 if ((v0 + size.x + 1) < (area)) {
                     const uint offset_base = v0 + size.x + 1;
                     VectorIndicesFromBase(offset_base);
-                    p.w = unquantized[offset];
+                    p.w = result_vector[array_index][vector_index];
                 }
-                unquantized_texel_weights[plane * 144 + t * block_dims.x + s] = (uint(dot(p, w)) + 8) >> 4;
+                const uint offset = (t * block_dims.x + s) + ARRAY_NUM_ELEMENTS * plane;
+                const uint array_index = offset / 4;
+                const uint vector_index = offset % 4;
+                unquantized_texel_weights[array_index][vector_index] = (uint(dot(p, w)) + 8) >> 4;
             }
         }
     }
@@ -1050,6 +1070,7 @@ TexelWeightParams DecodeBlockInfo() {
         weight_index += 6;
     }
     params.max_weight = weight_index + 1;
+
     return params;
 }
 
@@ -1079,7 +1100,7 @@ void FillVoidExtentLDR(ivec3 coord) {
 }
 
 void DecompressBlock(ivec3 coord) {
-    TexelWeightParams params = DecodeBlockInfo();
+    const TexelWeightParams params = DecodeBlockInfo();
     if (params.error_state) {
         FillError(coord);
         return;
@@ -1096,12 +1117,11 @@ void DecompressBlock(ivec3 coord) {
         FillError(coord);
         return;
     }
-    uint num_partitions = StreamBits(2) + 1;
+    const uint num_partitions = StreamBits(2) + 1;
     if (num_partitions > 4 || (num_partitions == 4 && params.dual_plane)) {
         FillError(coord);
         return;
     }
-    int plane_index = -1;
     uint partition_index = 1;
     uvec4 color_endpoint_mode = uvec4(0);
     uint ced_pointer = 0;
@@ -1113,8 +1133,8 @@ void DecompressBlock(ivec3 coord) {
         partition_index = StreamBits(10);
         base_cem = StreamBits(6);
     }
-    uint base_mode = base_cem & 3;
-    uint weight_bits = GetPackedBitSize(params.size, params.dual_plane, params.max_weight);
+    const uint base_mode = base_cem & 3;
+    const uint weight_bits = GetPackedBitSize(params.size, params.dual_plane, params.max_weight);
     uint remaining_bits = 128 - weight_bits - total_bitsread;
     uint extra_cem_bits = 0;
     if (base_mode > 0) {
@@ -1133,10 +1153,7 @@ void DecompressBlock(ivec3 coord) {
         }
     }
     remaining_bits -= extra_cem_bits;
-    uint plane_selector_bits = 0;
-    if (params.dual_plane) {
-        plane_selector_bits = 2;
-    }
+    const uint plane_selector_bits = params.dual_plane ? 2 : 0;
     remaining_bits -= plane_selector_bits;
     if (remaining_bits > 128) {
         // Bad data, more remaining bits than 4 bytes
@@ -1144,17 +1161,17 @@ void DecompressBlock(ivec3 coord) {
         return;
     }
     // Read color data...
-    uint color_data_bits = remaining_bits;
+    const uint color_data_bits = remaining_bits;
     while (remaining_bits > 0) {
-        int nb = int(min(remaining_bits, 32U));
-        uint b = StreamBits(nb);
+        const int nb = int(min(remaining_bits, 32U));
+        const uint b = StreamBits(nb);
         color_endpoint_data[ced_pointer] = uint(bitfieldExtract(b, 0, nb));
         ++ced_pointer;
         remaining_bits -= nb;
     }
-    plane_index = int(StreamBits(plane_selector_bits));
+    const uint plane_index = uint(StreamBits(plane_selector_bits));
     if (base_mode > 0) {
-        uint extra_cem = StreamBits(extra_cem_bits);
+        const uint extra_cem = StreamBits(extra_cem_bits);
         uint cem = (extra_cem << 6) | base_cem;
         cem >>= 2;
         uvec4 C = uvec4(0);
@@ -1176,43 +1193,54 @@ void DecompressBlock(ivec3 coord) {
             color_endpoint_mode[i] |= M[i];
         }
     } else if (num_partitions > 1) {
-        uint cem = base_cem >> 2;
+        const uint cem = base_cem >> 2;
         for (uint i = 0; i < num_partitions; i++) {
             color_endpoint_mode[i] = cem;
         }
     }
-    DecodeColorValues(color_endpoint_mode, num_partitions, color_data_bits);
 
-    uvec4 endpoints[4][2];
-    for (uint i = 0; i < num_partitions; i++) {
-        ComputeEndpoints(endpoints[i][0], endpoints[i][1], color_endpoint_mode[i]);
+    uvec4 endpoints0[4];
+    uvec4 endpoints1[4];
+    {
+        // This decode phase should at most push 32 elements into the vector
+        result_vector_max_index = 32;
+
+        uvec4 color_values[8];
+        uint colvals_index = 0;
+        DecodeColorValues(color_endpoint_mode, num_partitions, color_data_bits, color_values);
+        for (uint i = 0; i < num_partitions; i++) {
+            ComputeEndpoints(endpoints0[i], endpoints1[i], color_endpoint_mode[i], color_values,
+                             colvals_index);
+        }
     }
-
     color_endpoint_data = local_buff;
     color_endpoint_data = bitfieldReverse(color_endpoint_data).wzyx;
-    uint clear_byte_start =
-        (GetPackedBitSize(params.size, params.dual_plane, params.max_weight) >> 3) + 1;
+    const uint clear_byte_start = (weight_bits >> 3) + 1;
 
-    uint byte_insert = ExtractBits(color_endpoint_data, int(clear_byte_start - 1) * 8, 8) &
-        uint(
-            ((1 << (GetPackedBitSize(params.size, params.dual_plane, params.max_weight) % 8)) - 1));
-    uint vec_index = (clear_byte_start - 1) >> 2;
-    color_endpoint_data[vec_index] =
-        bitfieldInsert(color_endpoint_data[vec_index], byte_insert, int((clear_byte_start - 1) % 4) * 8, 8);
+    const uint byte_insert = ExtractBits(color_endpoint_data, int(clear_byte_start - 1) * 8, 8) &
+                             uint(((1 << (weight_bits % 8)) - 1));
+    const uint vec_index = (clear_byte_start - 1) >> 2;
+    color_endpoint_data[vec_index] = bitfieldInsert(color_endpoint_data[vec_index], byte_insert,
+                                                    int((clear_byte_start - 1) % 4) * 8, 8);
     for (uint i = clear_byte_start; i < 16; ++i) {
-        uint idx = i >> 2;
+        const uint idx = i >> 2;
         color_endpoint_data[idx] = bitfieldInsert(color_endpoint_data[idx], 0, int(i % 4) * 8, 8);
     }
 
     // Re-init vector variables for next decode phase
     result_index = 0;
     color_bitsread = 0;
+    result_limit_reached = false;
 
+    // The limit for the Unquantize phase, avoids decoding more data than needed.
+    result_vector_max_index = params.size.x * params.size.y;
+    if (params.dual_plane) {
+        result_vector_max_index *= 2;
+    }
     DecodeIntegerSequence(params.max_weight, GetNumWeightValues(params.size, params.dual_plane));
 
-    uint unquantized_texel_weights[2 * 144];
-    UnquantizeTexelWeights(params.dual_plane, params.size, unquantized_texel_weights);
-
+    uvec4 unquantized_texel_weights[VECTOR_ARRAY_SIZE];
+    UnquantizeTexelWeights(params.size, params.dual_plane, unquantized_texel_weights);
     for (uint j = 0; j < block_dims.y; j++) {
         for (uint i = 0; i < block_dims.x; i++) {
             uint local_partition = 0;
@@ -1220,13 +1248,19 @@ void DecompressBlock(ivec3 coord) {
                 local_partition = Select2DPartition(partition_index, i, j, num_partitions,
                                                     (block_dims.y * block_dims.x) < 32);
             }
-            const uvec4 C0 = ReplicateByteTo16(endpoints[local_partition][0]);
-            const uvec4 C1 = ReplicateByteTo16(endpoints[local_partition][1]);
+            const uvec4 C0 = ReplicateByteTo16(endpoints0[local_partition]);
+            const uvec4 C1 = ReplicateByteTo16(endpoints1[local_partition]);
             const uint weight_offset = (j * block_dims.x + i);
-            const uint primary_weight = unquantized_texel_weights[weight_offset];
+            const uint array_index = weight_offset / 4;
+            const uint vector_index = weight_offset % 4;
+            const uint primary_weight = unquantized_texel_weights[array_index][vector_index];
             uvec4 weight_vec = uvec4(primary_weight);
             if (params.dual_plane) {
-                const uint secondary_weight = unquantized_texel_weights[weight_offset + 144];
+                const uint secondary_weight_offset = (j * block_dims.x + i) + ARRAY_NUM_ELEMENTS;
+                const uint secondary_array_index = secondary_weight_offset / 4;
+                const uint secondary_vector_index = secondary_weight_offset % 4;
+                const uint secondary_weight =
+                    unquantized_texel_weights[secondary_array_index][secondary_vector_index];
                 for (uint c = 0; c < 4; c++) {
                     const bool is_secondary = ((plane_index + 1u) & 3u) == c;
                     weight_vec[c] = is_secondary ? secondary_weight : primary_weight;
@@ -1240,12 +1274,11 @@ void DecompressBlock(ivec3 coord) {
     }
 }
 
-
 uint SwizzleOffset(uvec2 pos) {
-    uint x = pos.x;
-    uint y = pos.y;
-    return ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + ((x % 32) / 16) * 32 +
-                          (y % 2) * 16 + (x % 16);
+    const uint x = pos.x;
+    const uint y = pos.y;
+    return ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 +
+            ((x % 32) / 16) * 32 + (y % 2) * 16 + (x % 16);
 }
 
 void main() {

From 3494fce86467878cd62a2faf54da2958b7495a17 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Tue, 1 Aug 2023 19:43:22 -0400
Subject: [PATCH 11/33] gl

---
 src/video_core/renderer_opengl/util_shaders.cpp | 1 +
 1 file changed, 1 insertion(+)

diff --git a/src/video_core/renderer_opengl/util_shaders.cpp b/src/video_core/renderer_opengl/util_shaders.cpp
index 544982d18..c437013e6 100644
--- a/src/video_core/renderer_opengl/util_shaders.cpp
+++ b/src/video_core/renderer_opengl/util_shaders.cpp
@@ -68,6 +68,7 @@ void UtilShaders::ASTCDecode(Image& image, const StagingBufferMap& map,
                              std::span<const VideoCommon::SwizzleParameters> swizzles) {
     static constexpr GLuint BINDING_INPUT_BUFFER = 0;
     static constexpr GLuint BINDING_OUTPUT_IMAGE = 0;
+    program_manager.LocalMemoryWarmup();
 
     const Extent2D tile_size{
         .width = VideoCore::Surface::DefaultBlockWidth(image.info.format),

From 05ee37a1f0641a2e1a15e9052371a5ce0cd20058 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Tue, 1 Aug 2023 19:48:19 -0400
Subject: [PATCH 12/33] amd opts

---
 src/video_core/host_shaders/astc_decoder.comp | 29 +++++++++----------
 1 file changed, 13 insertions(+), 16 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index e8801b0ff..ca93dc2a2 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -154,7 +154,7 @@ uint FastReplicateTo8(uint value, uint num_bits) {
         return 0;
     }
     const uint array_index = value / 4;
-    const uint vector_index = value % 4;
+    const uint vector_index = bitfieldExtract(value, 0, 2);
     switch (num_bits) {
     case 1:
         return 255;
@@ -213,7 +213,7 @@ uint FastReplicateTo6(uint value, uint num_bits) {
         return 0;
     }
     const uint array_index = value / 4;
-    const uint vector_index = value % 4;
+    const uint vector_index = bitfieldExtract(value, 0, 2);
     switch (num_bits) {
     case 1:
         return 63;
@@ -536,8 +536,8 @@ void DecodeIntegerSequence(uint max_range, uint num_values) {
     }
 }
 
-void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits,
-                       out uvec4 color_values[8]) {
+uvec4 color_values[8];
+void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
     uint num_values = 0;
     for (uint i = 0; i < num_partitions; i++) {
         num_values += ((modes[i] >> 2) + 1) << 1;
@@ -664,10 +664,7 @@ ivec2 BitTransferSigned(int a, int b) {
 }
 
 uvec4 ClampByte(ivec4 color) {
-    for (uint i = 0; i < 4; ++i) {
-        color[i] = clamp(color[i], 0, 255);
-    }
-    return uvec4(color);
+    return uvec4(clamp(color, 0, 255));
 }
 
 ivec4 BlueContract(int a, int r, int g, int b) {
@@ -675,7 +672,7 @@ ivec4 BlueContract(int a, int r, int g, int b) {
 }
 
 void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode,
-                      in uvec4 color_values[8], inout uint colvals_index) {
+                      inout uint colvals_index) {
 #define READ_UINT_VALUES(N)                                                                        \
     uint v[N];                                                                                     \
     for (uint i = 0; i < N; i++) {                                                                 \
@@ -887,8 +884,9 @@ uint UnquantizeTexelWeight(EncodingData val) {
     return result;
 }
 
-void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane,
-                            out uvec4 unquantized_texel_weights[VECTOR_ARRAY_SIZE]) {
+uvec4 unquantized_texel_weights[VECTOR_ARRAY_SIZE];
+
+void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
     const uint Ds = uint((block_dims.x * 0.5f + 1024) / (block_dims.x - 1));
     const uint Dt = uint((block_dims.y * 0.5f + 1024) / (block_dims.y - 1));
     const uint num_planes = is_dual_plane ? 2 : 1;
@@ -1205,11 +1203,11 @@ void DecompressBlock(ivec3 coord) {
         // This decode phase should at most push 32 elements into the vector
         result_vector_max_index = 32;
 
-        uvec4 color_values[8];
+        // uvec4 color_values[8];
         uint colvals_index = 0;
-        DecodeColorValues(color_endpoint_mode, num_partitions, color_data_bits, color_values);
+        DecodeColorValues(color_endpoint_mode, num_partitions, color_data_bits);
         for (uint i = 0; i < num_partitions; i++) {
-            ComputeEndpoints(endpoints0[i], endpoints1[i], color_endpoint_mode[i], color_values,
+            ComputeEndpoints(endpoints0[i], endpoints1[i], color_endpoint_mode[i],
                              colvals_index);
         }
     }
@@ -1239,8 +1237,7 @@ void DecompressBlock(ivec3 coord) {
     }
     DecodeIntegerSequence(params.max_weight, GetNumWeightValues(params.size, params.dual_plane));
 
-    uvec4 unquantized_texel_weights[VECTOR_ARRAY_SIZE];
-    UnquantizeTexelWeights(params.size, params.dual_plane, unquantized_texel_weights);
+    UnquantizeTexelWeights(params.size, params.dual_plane);
     for (uint j = 0; j < block_dims.y; j++) {
         for (uint i = 0; i < block_dims.x; i++) {
             uint local_partition = 0;

From 6b0b584eba111dfd716990b62499c307c930f555 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Tue, 1 Aug 2023 20:11:01 -0400
Subject: [PATCH 13/33] uint result index

---
 src/video_core/host_shaders/astc_decoder.comp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index ca93dc2a2..fd38dcfe5 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -88,7 +88,7 @@ int color_bitsread = 0;
 #define VECTOR_ARRAY_SIZE DIVCEIL(ARRAY_NUM_ELEMENTS * 2, 4)
 uvec4 result_vector[VECTOR_ARRAY_SIZE];
 
-int result_index = 0;
+uint result_index = 0;
 uint result_vector_max_index;
 bool result_limit_reached = false;
 

From 5c1655969478ca0b47c9f52d333f798b614b71ea Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Tue, 1 Aug 2023 20:10:46 -0400
Subject: [PATCH 14/33] bfe instead of mod

---
 src/video_core/host_shaders/astc_decoder.comp | 28 ++++++++++---------
 1 file changed, 15 insertions(+), 13 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index fd38dcfe5..5346cba0c 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -21,6 +21,8 @@
 
 #endif
 
+#define bfe bitfieldExtract
+
 layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
 
 BEGIN_PUSH_CONSTANTS
@@ -132,7 +134,7 @@ void ResultEmplaceBack(EncodingData val) {
         return;
     }
     const uint array_index = result_index / 4;
-    const uint vector_index = result_index % 4;
+    const uint vector_index = bfe(result_index, 0, 2);
     result_vector[array_index][vector_index] = val.data;
     ++result_index;
 }
@@ -384,7 +386,7 @@ uint StreamColorBits(uint num_bits) {
 
 EncodingData GetEncodingFromVector(uint index) {
     const uint array_index = index / 4;
-    const uint vector_index = index % 4;
+    const uint vector_index = bfe(index, 0, 2);
 
     const uint data = result_vector[array_index][vector_index];
     return EncodingData(data);
@@ -393,7 +395,7 @@ EncodingData GetEncodingFromVector(uint index) {
 // Returns the number of bits required to encode n_vals values.
 uint GetBitLength(uint n_vals, uint encoding_index) {
     const EncodingData encoding_value =
-        EncodingData(encoding_values[encoding_index / 4][encoding_index % 4]);
+        EncodingData(encoding_values[encoding_index / 4][bfe(encoding_index, 0, 2)]);
     const uint encoding = Encoding(encoding_value);
     uint total_bits = NumBits(encoding_value) * n_vals;
     if (encoding == TRIT) {
@@ -513,7 +515,7 @@ void DecodeTritBlock(uint num_bits) {
 }
 
 void DecodeIntegerSequence(uint max_range, uint num_values) {
-    EncodingData val = EncodingData(encoding_values[max_range / 4][max_range % 4]);
+    EncodingData val = EncodingData(encoding_values[max_range / 4][bfe(max_range, 0, 2)]);
     const uint encoding = Encoding(val);
     const uint num_bits = NumBits(val);
     uint vals_decoded = 0;
@@ -565,7 +567,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
         A = ReplicateBitTo9((bitval & 1));
         switch (encoding) {
         case JUST_BITS:
-            color_values[out_index / 4][out_index % 4] = FastReplicateTo8(bitval, bitlen);
+            color_values[out_index / 4][bfe(out_index, 0, 2)] = FastReplicateTo8(bitval, bitlen);
             ++out_index;
             break;
         case TRIT: {
@@ -645,7 +647,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
             uint T = (D * C) + B;
             T ^= A;
             T = (A & 0x80) | (T >> 2);
-            color_values[out_index / 4][out_index % 4] = T;
+            color_values[out_index / 4][bfe(out_index, 0, 2)] = T;
             ++out_index;
         }
     }
@@ -676,14 +678,14 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode,
 #define READ_UINT_VALUES(N)                                                                        \
     uint v[N];                                                                                     \
     for (uint i = 0; i < N; i++) {                                                                 \
-        v[i] = color_values[colvals_index / 4][colvals_index % 4];                                 \
+        v[i] = color_values[colvals_index / 4][bfe(colvals_index, 0, 2)];                                 \
         ++colvals_index;                                                                           \
     }
 
 #define READ_INT_VALUES(N)                                                                         \
     int v[N];                                                                                      \
     for (uint i = 0; i < N; i++) {                                                                 \
-        v[i] = int(color_values[colvals_index / 4][colvals_index % 4]);                            \
+        v[i] = int(color_values[colvals_index / 4][bfe(colvals_index, 0, 2)]);                            \
         ++colvals_index;                                                                           \
     }
 
@@ -894,7 +896,7 @@ void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
     const uint loop_count = min(result_index, area * num_planes);
     for (uint itr = 0; itr < loop_count; ++itr) {
         const uint array_index = itr / 4;
-        const uint vector_index = itr % 4;
+        const uint vector_index = bfe(itr, 0, 2);
         result_vector[array_index][vector_index] =
             UnquantizeTexelWeight(GetEncodingFromVector(itr));
     }
@@ -921,7 +923,7 @@ void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
 #define VectorIndicesFromBase(offset_base)                                                         \
     const uint offset = is_dual_plane ? 2 * offset_base + plane : offset_base;                     \
     const uint array_index = offset / 4;                                                           \
-    const uint vector_index = offset % 4;
+    const uint vector_index = bfe(offset, 0, 2);
 
                 if (v0 < area) {
                     const uint offset_base = v0;
@@ -945,7 +947,7 @@ void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
                 }
                 const uint offset = (t * block_dims.x + s) + ARRAY_NUM_ELEMENTS * plane;
                 const uint array_index = offset / 4;
-                const uint vector_index = offset % 4;
+                const uint vector_index = bfe(offset, 0, 2);
                 unquantized_texel_weights[array_index][vector_index] = (uint(dot(p, w)) + 8) >> 4;
             }
         }
@@ -1249,13 +1251,13 @@ void DecompressBlock(ivec3 coord) {
             const uvec4 C1 = ReplicateByteTo16(endpoints1[local_partition]);
             const uint weight_offset = (j * block_dims.x + i);
             const uint array_index = weight_offset / 4;
-            const uint vector_index = weight_offset % 4;
+            const uint vector_index = bfe(weight_offset, 0, 2);
             const uint primary_weight = unquantized_texel_weights[array_index][vector_index];
             uvec4 weight_vec = uvec4(primary_weight);
             if (params.dual_plane) {
                 const uint secondary_weight_offset = (j * block_dims.x + i) + ARRAY_NUM_ELEMENTS;
                 const uint secondary_array_index = secondary_weight_offset / 4;
-                const uint secondary_vector_index = secondary_weight_offset % 4;
+                const uint secondary_vector_index = bfe(secondary_weight_offset, 0, 2);
                 const uint secondary_weight =
                     unquantized_texel_weights[secondary_array_index][secondary_vector_index];
                 for (uint c = 0; c < 4; c++) {

From c077e467c4fb10ef92c1369e9b9116ed653af1de Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Tue, 1 Aug 2023 20:19:45 -0400
Subject: [PATCH 15/33] global endpoints

---
 src/video_core/host_shaders/astc_decoder.comp | 76 +++++++++----------
 1 file changed, 36 insertions(+), 40 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index 5346cba0c..077bec576 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -94,6 +94,8 @@ uint result_index = 0;
 uint result_vector_max_index;
 bool result_limit_reached = false;
 
+uvec4 endpoints[2][4];
+
 // EncodingData helpers
 uint Encoding(EncodingData val) {
     return bitfieldExtract(val.data, 0, 8);
@@ -673,7 +675,7 @@ ivec4 BlueContract(int a, int r, int g, int b) {
     return ivec4(a, (r + b) >> 1, (g + b) >> 1, b);
 }
 
-void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode,
+void ComputeEndpoints(uint ep_index, uint color_endpoint_mode,
                       inout uint colvals_index) {
 #define READ_UINT_VALUES(N)                                                                        \
     uint v[N];                                                                                     \
@@ -692,22 +694,22 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode,
     switch (color_endpoint_mode) {
     case 0: {
         READ_UINT_VALUES(2)
-        ep1 = uvec4(0xFF, v[0], v[0], v[0]);
-        ep2 = uvec4(0xFF, v[1], v[1], v[1]);
+        endpoints[0][ep_index] = uvec4(0xFF, v[0], v[0], v[0]);
+        endpoints[1][ep_index] = uvec4(0xFF, v[1], v[1], v[1]);
         break;
     }
     case 1: {
         READ_UINT_VALUES(2)
         const uint L0 = (v[0] >> 2) | (v[1] & 0xC0);
         const uint L1 = min(L0 + (v[1] & 0x3F), 0xFFU);
-        ep1 = uvec4(0xFF, L0, L0, L0);
-        ep2 = uvec4(0xFF, L1, L1, L1);
+        endpoints[0][ep_index] = uvec4(0xFF, L0, L0, L0);
+        endpoints[1][ep_index] = uvec4(0xFF, L1, L1, L1);
         break;
     }
     case 4: {
         READ_UINT_VALUES(4)
-        ep1 = uvec4(v[2], v[0], v[0], v[0]);
-        ep2 = uvec4(v[3], v[1], v[1], v[1]);
+        endpoints[0][ep_index] = uvec4(v[2], v[0], v[0], v[0]);
+        endpoints[1][ep_index] = uvec4(v[3], v[1], v[1], v[1]);
         break;
     }
     case 5: {
@@ -718,24 +720,24 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode,
         transferred = BitTransferSigned(v[3], v[2]);
         v[3] = transferred.x;
         v[2] = transferred.y;
-        ep1 = ClampByte(ivec4(v[2], v[0], v[0], v[0]));
-        ep2 = ClampByte(ivec4(v[2] + v[3], v[0] + v[1], v[0] + v[1], v[0] + v[1]));
+        endpoints[0][ep_index] = ClampByte(ivec4(v[2], v[0], v[0], v[0]));
+        endpoints[1][ep_index] = ClampByte(ivec4(v[2] + v[3], v[0] + v[1], v[0] + v[1], v[0] + v[1]));
         break;
     }
     case 6: {
         READ_UINT_VALUES(4)
-        ep1 = uvec4(0xFF, (v[0] * v[3]) >> 8, (v[1] * v[3]) >> 8, (v[2] * v[3]) >> 8);
-        ep2 = uvec4(0xFF, v[0], v[1], v[2]);
+        endpoints[0][ep_index] = uvec4(0xFF, (v[0] * v[3]) >> 8, (v[1] * v[3]) >> 8, (v[2] * v[3]) >> 8);
+        endpoints[1][ep_index] = uvec4(0xFF, v[0], v[1], v[2]);
         break;
     }
     case 8: {
         READ_UINT_VALUES(6)
         if ((v[1] + v[3] + v[5]) >= (v[0] + v[2] + v[4])) {
-            ep1 = uvec4(0xFF, v[0], v[2], v[4]);
-            ep2 = uvec4(0xFF, v[1], v[3], v[5]);
+            endpoints[0][ep_index] = uvec4(0xFF, v[0], v[2], v[4]);
+            endpoints[1][ep_index] = uvec4(0xFF, v[1], v[3], v[5]);
         } else {
-            ep1 = uvec4(BlueContract(0xFF, int(v[1]), int(v[3]), int(v[5])));
-            ep2 = uvec4(BlueContract(0xFF, int(v[0]), int(v[2]), int(v[4])));
+            endpoints[0][ep_index] = uvec4(BlueContract(0xFF, int(v[1]), int(v[3]), int(v[5])));
+            endpoints[1][ep_index] = uvec4(BlueContract(0xFF, int(v[0]), int(v[2]), int(v[4])));
         }
         break;
     }
@@ -751,28 +753,28 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode,
         v[5] = transferred.x;
         v[4] = transferred.y;
         if ((v[1] + v[3] + v[5]) >= 0) {
-            ep1 = ClampByte(ivec4(0xFF, v[0], v[2], v[4]));
-            ep2 = ClampByte(ivec4(0xFF, v[0] + v[1], v[2] + v[3], v[4] + v[5]));
+            endpoints[0][ep_index] = ClampByte(ivec4(0xFF, v[0], v[2], v[4]));
+            endpoints[1][ep_index] = ClampByte(ivec4(0xFF, v[0] + v[1], v[2] + v[3], v[4] + v[5]));
         } else {
-            ep1 = ClampByte(BlueContract(0xFF, v[0] + v[1], v[2] + v[3], v[4] + v[5]));
-            ep2 = ClampByte(BlueContract(0xFF, v[0], v[2], v[4]));
+            endpoints[0][ep_index] = ClampByte(BlueContract(0xFF, v[0] + v[1], v[2] + v[3], v[4] + v[5]));
+            endpoints[1][ep_index] = ClampByte(BlueContract(0xFF, v[0], v[2], v[4]));
         }
         break;
     }
     case 10: {
         READ_UINT_VALUES(6)
-        ep1 = uvec4(v[4], (v[0] * v[3]) >> 8, (v[1] * v[3]) >> 8, (v[2] * v[3]) >> 8);
-        ep2 = uvec4(v[5], v[0], v[1], v[2]);
+        endpoints[0][ep_index] = uvec4(v[4], (v[0] * v[3]) >> 8, (v[1] * v[3]) >> 8, (v[2] * v[3]) >> 8);
+        endpoints[1][ep_index] = uvec4(v[5], v[0], v[1], v[2]);
         break;
     }
     case 12: {
         READ_UINT_VALUES(8)
         if ((v[1] + v[3] + v[5]) >= (v[0] + v[2] + v[4])) {
-            ep1 = uvec4(v[6], v[0], v[2], v[4]);
-            ep2 = uvec4(v[7], v[1], v[3], v[5]);
+            endpoints[0][ep_index] = uvec4(v[6], v[0], v[2], v[4]);
+            endpoints[1][ep_index] = uvec4(v[7], v[1], v[3], v[5]);
         } else {
-            ep1 = uvec4(BlueContract(int(v[7]), int(v[1]), int(v[3]), int(v[5])));
-            ep2 = uvec4(BlueContract(int(v[6]), int(v[0]), int(v[2]), int(v[4])));
+            endpoints[0][ep_index] = uvec4(BlueContract(int(v[7]), int(v[1]), int(v[3]), int(v[5])));
+            endpoints[1][ep_index] = uvec4(BlueContract(int(v[6]), int(v[0]), int(v[2]), int(v[4])));
         }
         break;
     }
@@ -794,18 +796,18 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode,
         v[6] = transferred.y;
 
         if ((v[1] + v[3] + v[5]) >= 0) {
-            ep1 = ClampByte(ivec4(v[6], v[0], v[2], v[4]));
-            ep2 = ClampByte(ivec4(v[7] + v[6], v[0] + v[1], v[2] + v[3], v[4] + v[5]));
+            endpoints[0][ep_index] = ClampByte(ivec4(v[6], v[0], v[2], v[4]));
+            endpoints[1][ep_index] = ClampByte(ivec4(v[7] + v[6], v[0] + v[1], v[2] + v[3], v[4] + v[5]));
         } else {
-            ep1 = ClampByte(BlueContract(v[6] + v[7], v[0] + v[1], v[2] + v[3], v[4] + v[5]));
-            ep2 = ClampByte(BlueContract(v[6], v[0], v[2], v[4]));
+            endpoints[0][ep_index] = ClampByte(BlueContract(v[6] + v[7], v[0] + v[1], v[2] + v[3], v[4] + v[5]));
+            endpoints[1][ep_index] = ClampByte(BlueContract(v[6], v[0], v[2], v[4]));
         }
         break;
     }
     default: {
         // HDR mode, or more likely a bug computing the color_endpoint_mode
-        ep1 = uvec4(0xFF, 0xFF, 0, 0);
-        ep2 = uvec4(0xFF, 0xFF, 0, 0);
+        endpoints[0][ep_index] = uvec4(0xFF, 0xFF, 0, 0);
+        endpoints[1][ep_index] = uvec4(0xFF, 0xFF, 0, 0);
         break;
     }
     }
@@ -1198,10 +1200,6 @@ void DecompressBlock(ivec3 coord) {
             color_endpoint_mode[i] = cem;
         }
     }
-
-    uvec4 endpoints0[4];
-    uvec4 endpoints1[4];
-    {
         // This decode phase should at most push 32 elements into the vector
         result_vector_max_index = 32;
 
@@ -1209,10 +1207,8 @@ void DecompressBlock(ivec3 coord) {
         uint colvals_index = 0;
         DecodeColorValues(color_endpoint_mode, num_partitions, color_data_bits);
         for (uint i = 0; i < num_partitions; i++) {
-            ComputeEndpoints(endpoints0[i], endpoints1[i], color_endpoint_mode[i],
-                             colvals_index);
+            ComputeEndpoints(i, color_endpoint_mode[i], colvals_index);
         }
-    }
     color_endpoint_data = local_buff;
     color_endpoint_data = bitfieldReverse(color_endpoint_data).wzyx;
     const uint clear_byte_start = (weight_bits >> 3) + 1;
@@ -1247,8 +1243,8 @@ void DecompressBlock(ivec3 coord) {
                 local_partition = Select2DPartition(partition_index, i, j, num_partitions,
                                                     (block_dims.y * block_dims.x) < 32);
             }
-            const uvec4 C0 = ReplicateByteTo16(endpoints0[local_partition]);
-            const uvec4 C1 = ReplicateByteTo16(endpoints1[local_partition]);
+            const uvec4 C0 = ReplicateByteTo16(endpoints[0][local_partition]);
+            const uvec4 C1 = ReplicateByteTo16(endpoints[1][local_partition]);
             const uint weight_offset = (j * block_dims.x + i);
             const uint array_index = weight_offset / 4;
             const uint vector_index = bfe(weight_offset, 0, 2);

From 553dd3e1202638fa8bad9fed56110ee447208ecf Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Tue, 1 Aug 2023 20:53:25 -0400
Subject: [PATCH 16/33] Revert "global endpoints"

This reverts commit d8f5bfd1df2b7469ef6abcee182aa110602d1751.
---
 src/video_core/host_shaders/astc_decoder.comp | 76 ++++++++++---------
 1 file changed, 40 insertions(+), 36 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index 077bec576..5346cba0c 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -94,8 +94,6 @@ uint result_index = 0;
 uint result_vector_max_index;
 bool result_limit_reached = false;
 
-uvec4 endpoints[2][4];
-
 // EncodingData helpers
 uint Encoding(EncodingData val) {
     return bitfieldExtract(val.data, 0, 8);
@@ -675,7 +673,7 @@ ivec4 BlueContract(int a, int r, int g, int b) {
     return ivec4(a, (r + b) >> 1, (g + b) >> 1, b);
 }
 
-void ComputeEndpoints(uint ep_index, uint color_endpoint_mode,
+void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode,
                       inout uint colvals_index) {
 #define READ_UINT_VALUES(N)                                                                        \
     uint v[N];                                                                                     \
@@ -694,22 +692,22 @@ void ComputeEndpoints(uint ep_index, uint color_endpoint_mode,
     switch (color_endpoint_mode) {
     case 0: {
         READ_UINT_VALUES(2)
-        endpoints[0][ep_index] = uvec4(0xFF, v[0], v[0], v[0]);
-        endpoints[1][ep_index] = uvec4(0xFF, v[1], v[1], v[1]);
+        ep1 = uvec4(0xFF, v[0], v[0], v[0]);
+        ep2 = uvec4(0xFF, v[1], v[1], v[1]);
         break;
     }
     case 1: {
         READ_UINT_VALUES(2)
         const uint L0 = (v[0] >> 2) | (v[1] & 0xC0);
         const uint L1 = min(L0 + (v[1] & 0x3F), 0xFFU);
-        endpoints[0][ep_index] = uvec4(0xFF, L0, L0, L0);
-        endpoints[1][ep_index] = uvec4(0xFF, L1, L1, L1);
+        ep1 = uvec4(0xFF, L0, L0, L0);
+        ep2 = uvec4(0xFF, L1, L1, L1);
         break;
     }
     case 4: {
         READ_UINT_VALUES(4)
-        endpoints[0][ep_index] = uvec4(v[2], v[0], v[0], v[0]);
-        endpoints[1][ep_index] = uvec4(v[3], v[1], v[1], v[1]);
+        ep1 = uvec4(v[2], v[0], v[0], v[0]);
+        ep2 = uvec4(v[3], v[1], v[1], v[1]);
         break;
     }
     case 5: {
@@ -720,24 +718,24 @@ void ComputeEndpoints(uint ep_index, uint color_endpoint_mode,
         transferred = BitTransferSigned(v[3], v[2]);
         v[3] = transferred.x;
         v[2] = transferred.y;
-        endpoints[0][ep_index] = ClampByte(ivec4(v[2], v[0], v[0], v[0]));
-        endpoints[1][ep_index] = ClampByte(ivec4(v[2] + v[3], v[0] + v[1], v[0] + v[1], v[0] + v[1]));
+        ep1 = ClampByte(ivec4(v[2], v[0], v[0], v[0]));
+        ep2 = ClampByte(ivec4(v[2] + v[3], v[0] + v[1], v[0] + v[1], v[0] + v[1]));
         break;
     }
     case 6: {
         READ_UINT_VALUES(4)
-        endpoints[0][ep_index] = uvec4(0xFF, (v[0] * v[3]) >> 8, (v[1] * v[3]) >> 8, (v[2] * v[3]) >> 8);
-        endpoints[1][ep_index] = uvec4(0xFF, v[0], v[1], v[2]);
+        ep1 = uvec4(0xFF, (v[0] * v[3]) >> 8, (v[1] * v[3]) >> 8, (v[2] * v[3]) >> 8);
+        ep2 = uvec4(0xFF, v[0], v[1], v[2]);
         break;
     }
     case 8: {
         READ_UINT_VALUES(6)
         if ((v[1] + v[3] + v[5]) >= (v[0] + v[2] + v[4])) {
-            endpoints[0][ep_index] = uvec4(0xFF, v[0], v[2], v[4]);
-            endpoints[1][ep_index] = uvec4(0xFF, v[1], v[3], v[5]);
+            ep1 = uvec4(0xFF, v[0], v[2], v[4]);
+            ep2 = uvec4(0xFF, v[1], v[3], v[5]);
         } else {
-            endpoints[0][ep_index] = uvec4(BlueContract(0xFF, int(v[1]), int(v[3]), int(v[5])));
-            endpoints[1][ep_index] = uvec4(BlueContract(0xFF, int(v[0]), int(v[2]), int(v[4])));
+            ep1 = uvec4(BlueContract(0xFF, int(v[1]), int(v[3]), int(v[5])));
+            ep2 = uvec4(BlueContract(0xFF, int(v[0]), int(v[2]), int(v[4])));
         }
         break;
     }
@@ -753,28 +751,28 @@ void ComputeEndpoints(uint ep_index, uint color_endpoint_mode,
         v[5] = transferred.x;
         v[4] = transferred.y;
         if ((v[1] + v[3] + v[5]) >= 0) {
-            endpoints[0][ep_index] = ClampByte(ivec4(0xFF, v[0], v[2], v[4]));
-            endpoints[1][ep_index] = ClampByte(ivec4(0xFF, v[0] + v[1], v[2] + v[3], v[4] + v[5]));
+            ep1 = ClampByte(ivec4(0xFF, v[0], v[2], v[4]));
+            ep2 = ClampByte(ivec4(0xFF, v[0] + v[1], v[2] + v[3], v[4] + v[5]));
         } else {
-            endpoints[0][ep_index] = ClampByte(BlueContract(0xFF, v[0] + v[1], v[2] + v[3], v[4] + v[5]));
-            endpoints[1][ep_index] = ClampByte(BlueContract(0xFF, v[0], v[2], v[4]));
+            ep1 = ClampByte(BlueContract(0xFF, v[0] + v[1], v[2] + v[3], v[4] + v[5]));
+            ep2 = ClampByte(BlueContract(0xFF, v[0], v[2], v[4]));
         }
         break;
     }
     case 10: {
         READ_UINT_VALUES(6)
-        endpoints[0][ep_index] = uvec4(v[4], (v[0] * v[3]) >> 8, (v[1] * v[3]) >> 8, (v[2] * v[3]) >> 8);
-        endpoints[1][ep_index] = uvec4(v[5], v[0], v[1], v[2]);
+        ep1 = uvec4(v[4], (v[0] * v[3]) >> 8, (v[1] * v[3]) >> 8, (v[2] * v[3]) >> 8);
+        ep2 = uvec4(v[5], v[0], v[1], v[2]);
         break;
     }
     case 12: {
         READ_UINT_VALUES(8)
         if ((v[1] + v[3] + v[5]) >= (v[0] + v[2] + v[4])) {
-            endpoints[0][ep_index] = uvec4(v[6], v[0], v[2], v[4]);
-            endpoints[1][ep_index] = uvec4(v[7], v[1], v[3], v[5]);
+            ep1 = uvec4(v[6], v[0], v[2], v[4]);
+            ep2 = uvec4(v[7], v[1], v[3], v[5]);
         } else {
-            endpoints[0][ep_index] = uvec4(BlueContract(int(v[7]), int(v[1]), int(v[3]), int(v[5])));
-            endpoints[1][ep_index] = uvec4(BlueContract(int(v[6]), int(v[0]), int(v[2]), int(v[4])));
+            ep1 = uvec4(BlueContract(int(v[7]), int(v[1]), int(v[3]), int(v[5])));
+            ep2 = uvec4(BlueContract(int(v[6]), int(v[0]), int(v[2]), int(v[4])));
         }
         break;
     }
@@ -796,18 +794,18 @@ void ComputeEndpoints(uint ep_index, uint color_endpoint_mode,
         v[6] = transferred.y;
 
         if ((v[1] + v[3] + v[5]) >= 0) {
-            endpoints[0][ep_index] = ClampByte(ivec4(v[6], v[0], v[2], v[4]));
-            endpoints[1][ep_index] = ClampByte(ivec4(v[7] + v[6], v[0] + v[1], v[2] + v[3], v[4] + v[5]));
+            ep1 = ClampByte(ivec4(v[6], v[0], v[2], v[4]));
+            ep2 = ClampByte(ivec4(v[7] + v[6], v[0] + v[1], v[2] + v[3], v[4] + v[5]));
         } else {
-            endpoints[0][ep_index] = ClampByte(BlueContract(v[6] + v[7], v[0] + v[1], v[2] + v[3], v[4] + v[5]));
-            endpoints[1][ep_index] = ClampByte(BlueContract(v[6], v[0], v[2], v[4]));
+            ep1 = ClampByte(BlueContract(v[6] + v[7], v[0] + v[1], v[2] + v[3], v[4] + v[5]));
+            ep2 = ClampByte(BlueContract(v[6], v[0], v[2], v[4]));
         }
         break;
     }
     default: {
         // HDR mode, or more likely a bug computing the color_endpoint_mode
-        endpoints[0][ep_index] = uvec4(0xFF, 0xFF, 0, 0);
-        endpoints[1][ep_index] = uvec4(0xFF, 0xFF, 0, 0);
+        ep1 = uvec4(0xFF, 0xFF, 0, 0);
+        ep2 = uvec4(0xFF, 0xFF, 0, 0);
         break;
     }
     }
@@ -1200,6 +1198,10 @@ void DecompressBlock(ivec3 coord) {
             color_endpoint_mode[i] = cem;
         }
     }
+
+    uvec4 endpoints0[4];
+    uvec4 endpoints1[4];
+    {
         // This decode phase should at most push 32 elements into the vector
         result_vector_max_index = 32;
 
@@ -1207,8 +1209,10 @@ void DecompressBlock(ivec3 coord) {
         uint colvals_index = 0;
         DecodeColorValues(color_endpoint_mode, num_partitions, color_data_bits);
         for (uint i = 0; i < num_partitions; i++) {
-            ComputeEndpoints(i, color_endpoint_mode[i], colvals_index);
+            ComputeEndpoints(endpoints0[i], endpoints1[i], color_endpoint_mode[i],
+                             colvals_index);
         }
+    }
     color_endpoint_data = local_buff;
     color_endpoint_data = bitfieldReverse(color_endpoint_data).wzyx;
     const uint clear_byte_start = (weight_bits >> 3) + 1;
@@ -1243,8 +1247,8 @@ void DecompressBlock(ivec3 coord) {
                 local_partition = Select2DPartition(partition_index, i, j, num_partitions,
                                                     (block_dims.y * block_dims.x) < 32);
             }
-            const uvec4 C0 = ReplicateByteTo16(endpoints[0][local_partition]);
-            const uvec4 C1 = ReplicateByteTo16(endpoints[1][local_partition]);
+            const uvec4 C0 = ReplicateByteTo16(endpoints0[local_partition]);
+            const uvec4 C1 = ReplicateByteTo16(endpoints1[local_partition]);
             const uint weight_offset = (j * block_dims.x + i);
             const uint array_index = weight_offset / 4;
             const uint vector_index = bfe(weight_offset, 0, 2);

From f41fb3ec0b12761c723f26e5f44d1c7f15468153 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Tue, 1 Aug 2023 20:52:14 -0400
Subject: [PATCH 17/33] Revert "bfe instead of mod"

This reverts commit 86006a3b09e8a8c17d2ade61be76736a79e3f58a.
---
 src/video_core/host_shaders/astc_decoder.comp | 28 +++++++++----------
 1 file changed, 13 insertions(+), 15 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index 5346cba0c..fd38dcfe5 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -21,8 +21,6 @@
 
 #endif
 
-#define bfe bitfieldExtract
-
 layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
 
 BEGIN_PUSH_CONSTANTS
@@ -134,7 +132,7 @@ void ResultEmplaceBack(EncodingData val) {
         return;
     }
     const uint array_index = result_index / 4;
-    const uint vector_index = bfe(result_index, 0, 2);
+    const uint vector_index = result_index % 4;
     result_vector[array_index][vector_index] = val.data;
     ++result_index;
 }
@@ -386,7 +384,7 @@ uint StreamColorBits(uint num_bits) {
 
 EncodingData GetEncodingFromVector(uint index) {
     const uint array_index = index / 4;
-    const uint vector_index = bfe(index, 0, 2);
+    const uint vector_index = index % 4;
 
     const uint data = result_vector[array_index][vector_index];
     return EncodingData(data);
@@ -395,7 +393,7 @@ EncodingData GetEncodingFromVector(uint index) {
 // Returns the number of bits required to encode n_vals values.
 uint GetBitLength(uint n_vals, uint encoding_index) {
     const EncodingData encoding_value =
-        EncodingData(encoding_values[encoding_index / 4][bfe(encoding_index, 0, 2)]);
+        EncodingData(encoding_values[encoding_index / 4][encoding_index % 4]);
     const uint encoding = Encoding(encoding_value);
     uint total_bits = NumBits(encoding_value) * n_vals;
     if (encoding == TRIT) {
@@ -515,7 +513,7 @@ void DecodeTritBlock(uint num_bits) {
 }
 
 void DecodeIntegerSequence(uint max_range, uint num_values) {
-    EncodingData val = EncodingData(encoding_values[max_range / 4][bfe(max_range, 0, 2)]);
+    EncodingData val = EncodingData(encoding_values[max_range / 4][max_range % 4]);
     const uint encoding = Encoding(val);
     const uint num_bits = NumBits(val);
     uint vals_decoded = 0;
@@ -567,7 +565,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
         A = ReplicateBitTo9((bitval & 1));
         switch (encoding) {
         case JUST_BITS:
-            color_values[out_index / 4][bfe(out_index, 0, 2)] = FastReplicateTo8(bitval, bitlen);
+            color_values[out_index / 4][out_index % 4] = FastReplicateTo8(bitval, bitlen);
             ++out_index;
             break;
         case TRIT: {
@@ -647,7 +645,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
             uint T = (D * C) + B;
             T ^= A;
             T = (A & 0x80) | (T >> 2);
-            color_values[out_index / 4][bfe(out_index, 0, 2)] = T;
+            color_values[out_index / 4][out_index % 4] = T;
             ++out_index;
         }
     }
@@ -678,14 +676,14 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode,
 #define READ_UINT_VALUES(N)                                                                        \
     uint v[N];                                                                                     \
     for (uint i = 0; i < N; i++) {                                                                 \
-        v[i] = color_values[colvals_index / 4][bfe(colvals_index, 0, 2)];                                 \
+        v[i] = color_values[colvals_index / 4][colvals_index % 4];                                 \
         ++colvals_index;                                                                           \
     }
 
 #define READ_INT_VALUES(N)                                                                         \
     int v[N];                                                                                      \
     for (uint i = 0; i < N; i++) {                                                                 \
-        v[i] = int(color_values[colvals_index / 4][bfe(colvals_index, 0, 2)]);                            \
+        v[i] = int(color_values[colvals_index / 4][colvals_index % 4]);                            \
         ++colvals_index;                                                                           \
     }
 
@@ -896,7 +894,7 @@ void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
     const uint loop_count = min(result_index, area * num_planes);
     for (uint itr = 0; itr < loop_count; ++itr) {
         const uint array_index = itr / 4;
-        const uint vector_index = bfe(itr, 0, 2);
+        const uint vector_index = itr % 4;
         result_vector[array_index][vector_index] =
             UnquantizeTexelWeight(GetEncodingFromVector(itr));
     }
@@ -923,7 +921,7 @@ void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
 #define VectorIndicesFromBase(offset_base)                                                         \
     const uint offset = is_dual_plane ? 2 * offset_base + plane : offset_base;                     \
     const uint array_index = offset / 4;                                                           \
-    const uint vector_index = bfe(offset, 0, 2);
+    const uint vector_index = offset % 4;
 
                 if (v0 < area) {
                     const uint offset_base = v0;
@@ -947,7 +945,7 @@ void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
                 }
                 const uint offset = (t * block_dims.x + s) + ARRAY_NUM_ELEMENTS * plane;
                 const uint array_index = offset / 4;
-                const uint vector_index = bfe(offset, 0, 2);
+                const uint vector_index = offset % 4;
                 unquantized_texel_weights[array_index][vector_index] = (uint(dot(p, w)) + 8) >> 4;
             }
         }
@@ -1251,13 +1249,13 @@ void DecompressBlock(ivec3 coord) {
             const uvec4 C1 = ReplicateByteTo16(endpoints1[local_partition]);
             const uint weight_offset = (j * block_dims.x + i);
             const uint array_index = weight_offset / 4;
-            const uint vector_index = bfe(weight_offset, 0, 2);
+            const uint vector_index = weight_offset % 4;
             const uint primary_weight = unquantized_texel_weights[array_index][vector_index];
             uvec4 weight_vec = uvec4(primary_weight);
             if (params.dual_plane) {
                 const uint secondary_weight_offset = (j * block_dims.x + i) + ARRAY_NUM_ELEMENTS;
                 const uint secondary_array_index = secondary_weight_offset / 4;
-                const uint secondary_vector_index = bfe(secondary_weight_offset, 0, 2);
+                const uint secondary_vector_index = secondary_weight_offset % 4;
                 const uint secondary_weight =
                     unquantized_texel_weights[secondary_array_index][secondary_vector_index];
                 for (uint c = 0; c < 4; c++) {

From f2cf81e0b69652bec91857fd6d447e0e4ea137b7 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Tue, 1 Aug 2023 20:52:17 -0400
Subject: [PATCH 18/33] Revert "uint result index"

This reverts commit 0e978786b5a8e7382005d8b1e16cfa12f3eeb775.
---
 src/video_core/host_shaders/astc_decoder.comp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index fd38dcfe5..ca93dc2a2 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -88,7 +88,7 @@ int color_bitsread = 0;
 #define VECTOR_ARRAY_SIZE DIVCEIL(ARRAY_NUM_ELEMENTS * 2, 4)
 uvec4 result_vector[VECTOR_ARRAY_SIZE];
 
-uint result_index = 0;
+int result_index = 0;
 uint result_vector_max_index;
 bool result_limit_reached = false;
 

From 441b8471070dfbf8724d6ec195c0083ad4d9bd2b Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Tue, 1 Aug 2023 23:22:21 -0400
Subject: [PATCH 19/33] more packing

---
 src/video_core/host_shaders/astc_decoder.comp | 212 +++++++++---------
 1 file changed, 106 insertions(+), 106 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index ca93dc2a2..0e611ede5 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -88,7 +88,7 @@ int color_bitsread = 0;
 #define VECTOR_ARRAY_SIZE DIVCEIL(ARRAY_NUM_ELEMENTS * 2, 4)
 uvec4 result_vector[VECTOR_ARRAY_SIZE];
 
-int result_index = 0;
+uint result_index = 0;
 uint result_vector_max_index;
 bool result_limit_reached = false;
 
@@ -131,8 +131,8 @@ void ResultEmplaceBack(EncodingData val) {
         result_limit_reached = true;
         return;
     }
-    const uint array_index = result_index / 4;
-    const uint vector_index = result_index % 4;
+    const uint array_index = result_index / 4u;
+    const uint vector_index = result_index % 4u;
     result_vector[array_index][vector_index] = val.data;
     ++result_index;
 }
@@ -428,69 +428,68 @@ uint BitsOp(uint bits, uint start, uint end) {
 
 void DecodeQuintBlock(uint num_bits) {
     uvec3 m;
-    uvec3 q;
-    uint Q;
+    uvec4 qQ;
     m[0] = StreamColorBits(num_bits);
-    Q = StreamColorBits(3);
+    qQ.w = StreamColorBits(3);
     m[1] = StreamColorBits(num_bits);
-    Q |= StreamColorBits(2) << 3;
+    qQ.w |= StreamColorBits(2) << 3;
     m[2] = StreamColorBits(num_bits);
-    Q |= StreamColorBits(2) << 5;
-    if (BitsOp(Q, 1, 2) == 3 && BitsOp(Q, 5, 6) == 0) {
-        q.x = 4;
-        q.y = 4;
-        q.z = (BitsBracket(Q, 0) << 2) | ((BitsBracket(Q, 4) & ~BitsBracket(Q, 0)) << 1) |
-              (BitsBracket(Q, 3) & ~BitsBracket(Q, 0));
+    qQ.w |= StreamColorBits(2) << 5;
+    if (BitsOp(qQ.w, 1, 2) == 3 && BitsOp(qQ.w, 5, 6) == 0) {
+        qQ.x = 4;
+        qQ.y = 4;
+        qQ.z = (BitsBracket(qQ.w, 0) << 2) | ((BitsBracket(qQ.w, 4) & ~BitsBracket(qQ.w, 0)) << 1) |
+              (BitsBracket(qQ.w, 3) & ~BitsBracket(qQ.w, 0));
     } else {
         uint C = 0;
-        if (BitsOp(Q, 1, 2) == 3) {
-            q.z = 4;
-            C = (BitsOp(Q, 3, 4) << 3) | ((~BitsOp(Q, 5, 6) & 3) << 1) | BitsBracket(Q, 0);
+        if (BitsOp(qQ.w, 1, 2) == 3) {
+            qQ.z = 4;
+            C = (BitsOp(qQ.w, 3, 4) << 3) | ((~BitsOp(qQ.w, 5, 6) & 3) << 1) | BitsBracket(qQ.w, 0);
         } else {
-            q.z = BitsOp(Q, 5, 6);
-            C = BitsOp(Q, 0, 4);
+            qQ.z = BitsOp(qQ.w, 5, 6);
+            C = BitsOp(qQ.w, 0, 4);
         }
         if (BitsOp(C, 0, 2) == 5) {
-            q.y = 4;
-            q.x = BitsOp(C, 3, 4);
+            qQ.y = 4;
+            qQ.x = BitsOp(C, 3, 4);
         } else {
-            q.y = BitsOp(C, 3, 4);
-            q.x = BitsOp(C, 0, 2);
+            qQ.y = BitsOp(C, 3, 4);
+            qQ.x = BitsOp(C, 0, 2);
         }
     }
     for (uint i = 0; i < 3; i++) {
-        const EncodingData val = CreateEncodingData(QUINT, num_bits, m[i], q[i]);
+        const EncodingData val = CreateEncodingData(QUINT, num_bits, m[i], qQ[i]);
         ResultEmplaceBack(val);
     }
 }
 
 void DecodeTritBlock(uint num_bits) {
-    uint m[5];
-    uint t[5];
-    uint T;
+    uvec4 m;
+    uvec4 t;
+    uvec3 Tm5t5;
     m[0] = StreamColorBits(num_bits);
-    T = StreamColorBits(2);
+    Tm5t5.x = StreamColorBits(2);
     m[1] = StreamColorBits(num_bits);
-    T |= StreamColorBits(2) << 2;
+    Tm5t5.x |= StreamColorBits(2) << 2;
     m[2] = StreamColorBits(num_bits);
-    T |= StreamColorBits(1) << 4;
+    Tm5t5.x |= StreamColorBits(1) << 4;
     m[3] = StreamColorBits(num_bits);
-    T |= StreamColorBits(2) << 5;
-    m[4] = StreamColorBits(num_bits);
-    T |= StreamColorBits(1) << 7;
+    Tm5t5.x |= StreamColorBits(2) << 5;
+    Tm5t5.y = StreamColorBits(num_bits);
+    Tm5t5.x |= StreamColorBits(1) << 7;
     uint C = 0;
-    if (BitsOp(T, 2, 4) == 7) {
-        C = (BitsOp(T, 5, 7) << 2) | BitsOp(T, 0, 1);
-        t[4] = 2;
+    if (BitsOp(Tm5t5.x, 2, 4) == 7) {
+        C = (BitsOp(Tm5t5.x, 5, 7) << 2) | BitsOp(Tm5t5.x, 0, 1);
+        Tm5t5.z = 2;
         t[3] = 2;
     } else {
-        C = BitsOp(T, 0, 4);
-        if (BitsOp(T, 5, 6) == 3) {
-            t[4] = 2;
-            t[3] = BitsBracket(T, 7);
+        C = BitsOp(Tm5t5.x, 0, 4);
+        if (BitsOp(Tm5t5.x, 5, 6) == 3) {
+            Tm5t5.z = 2;
+            t[3] = BitsBracket(Tm5t5.x, 7);
         } else {
-            t[4] = BitsBracket(T, 7);
-            t[3] = BitsOp(T, 5, 6);
+            Tm5t5.z = BitsBracket(Tm5t5.x, 7);
+            t[3] = BitsOp(Tm5t5.x, 5, 6);
         }
     }
     if (BitsOp(C, 0, 1) == 3) {
@@ -506,10 +505,12 @@ void DecodeTritBlock(uint num_bits) {
         t[1] = BitsOp(C, 2, 3);
         t[0] = (BitsBracket(C, 1) << 1) | (BitsBracket(C, 0) & ~BitsBracket(C, 1));
     }
-    for (uint i = 0; i < 5; i++) {
+    for (uint i = 0; i < 4; i++) {
         const EncodingData val = CreateEncodingData(TRIT, num_bits, m[i], t[i]);
         ResultEmplaceBack(val);
     }
+    const EncodingData val = CreateEncodingData(TRIT, num_bits, Tm5t5.y, Tm5t5.z);
+    ResultEmplaceBack(val);
 }
 
 void DecodeIntegerSequence(uint max_range, uint num_values) {
@@ -674,129 +675,128 @@ ivec4 BlueContract(int a, int r, int g, int b) {
 void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode,
                       inout uint colvals_index) {
 #define READ_UINT_VALUES(N)                                                                        \
-    uint v[N];                                                                                     \
+    uvec4 V[2];                                                                                    \
     for (uint i = 0; i < N; i++) {                                                                 \
-        v[i] = color_values[colvals_index / 4][colvals_index % 4];                                 \
+        V[i / 4][i % 4] = color_values[colvals_index / 4][colvals_index % 4];                      \
         ++colvals_index;                                                                           \
     }
-
 #define READ_INT_VALUES(N)                                                                         \
-    int v[N];                                                                                      \
+    ivec4 V[2];                                                                                    \
     for (uint i = 0; i < N; i++) {                                                                 \
-        v[i] = int(color_values[colvals_index / 4][colvals_index % 4]);                            \
+        V[i / 4][i % 4] = int(color_values[colvals_index / 4][colvals_index % 4]);                 \
         ++colvals_index;                                                                           \
     }
 
     switch (color_endpoint_mode) {
     case 0: {
         READ_UINT_VALUES(2)
-        ep1 = uvec4(0xFF, v[0], v[0], v[0]);
-        ep2 = uvec4(0xFF, v[1], v[1], v[1]);
+        ep1 = uvec4(0xFF, V[0].x, V[0].x, V[0].x);
+        ep2 = uvec4(0xFF, V[0].y, V[0].y, V[0].y);
         break;
     }
     case 1: {
         READ_UINT_VALUES(2)
-        const uint L0 = (v[0] >> 2) | (v[1] & 0xC0);
-        const uint L1 = min(L0 + (v[1] & 0x3F), 0xFFU);
+        const uint L0 = (V[0].x >> 2) | (V[0].y & 0xC0);
+        const uint L1 = min(L0 + (V[0].y & 0x3F), 0xFFU);
         ep1 = uvec4(0xFF, L0, L0, L0);
         ep2 = uvec4(0xFF, L1, L1, L1);
         break;
     }
     case 4: {
         READ_UINT_VALUES(4)
-        ep1 = uvec4(v[2], v[0], v[0], v[0]);
-        ep2 = uvec4(v[3], v[1], v[1], v[1]);
+        ep1 = uvec4(V[0].z, V[0].x, V[0].x, V[0].x);
+        ep2 = uvec4(V[0].w, V[0].y, V[0].y, V[0].y);
         break;
     }
     case 5: {
         READ_INT_VALUES(4)
-        ivec2 transferred = BitTransferSigned(v[1], v[0]);
-        v[1] = transferred.x;
-        v[0] = transferred.y;
-        transferred = BitTransferSigned(v[3], v[2]);
-        v[3] = transferred.x;
-        v[2] = transferred.y;
-        ep1 = ClampByte(ivec4(v[2], v[0], v[0], v[0]));
-        ep2 = ClampByte(ivec4(v[2] + v[3], v[0] + v[1], v[0] + v[1], v[0] + v[1]));
+        ivec2 transferred = BitTransferSigned(V[0].y, V[0].x);
+        V[0].y = transferred.x;
+        V[0].x = transferred.y;
+        transferred = BitTransferSigned(V[0].w, V[0].z);
+        V[0].w = transferred.x;
+        V[0].z = transferred.y;
+        ep1 = ClampByte(ivec4(V[0].z, V[0].x, V[0].x, V[0].x));
+        ep2 = ClampByte(ivec4(V[0].z + V[0].w, V[0].x + V[0].y, V[0].x + V[0].y, V[0].x + V[0].y));
         break;
     }
     case 6: {
         READ_UINT_VALUES(4)
-        ep1 = uvec4(0xFF, (v[0] * v[3]) >> 8, (v[1] * v[3]) >> 8, (v[2] * v[3]) >> 8);
-        ep2 = uvec4(0xFF, v[0], v[1], v[2]);
+        ep1 = uvec4(0xFF, (V[0].x * V[0].w) >> 8, (V[0].y * V[0].w) >> 8, (V[0].z * V[0].w) >> 8);
+        ep2 = uvec4(0xFF, V[0].x, V[0].y, V[0].z);
         break;
     }
     case 8: {
         READ_UINT_VALUES(6)
-        if ((v[1] + v[3] + v[5]) >= (v[0] + v[2] + v[4])) {
-            ep1 = uvec4(0xFF, v[0], v[2], v[4]);
-            ep2 = uvec4(0xFF, v[1], v[3], v[5]);
+        if ((V[0].y + V[0].w + V[1].y) >= (V[0].x + V[0].z + V[1].x)) {
+            ep1 = uvec4(0xFF, V[0].x, V[0].z, V[1].x);
+            ep2 = uvec4(0xFF, V[0].y, V[0].w, V[1].y);
         } else {
-            ep1 = uvec4(BlueContract(0xFF, int(v[1]), int(v[3]), int(v[5])));
-            ep2 = uvec4(BlueContract(0xFF, int(v[0]), int(v[2]), int(v[4])));
+            ep1 = uvec4(BlueContract(0xFF, int(V[0].y), int(V[0].w), int(V[1].y)));
+            ep2 = uvec4(BlueContract(0xFF, int(V[0].x), int(V[0].z), int(V[1].x)));
         }
         break;
     }
     case 9: {
         READ_INT_VALUES(6)
-        ivec2 transferred = BitTransferSigned(v[1], v[0]);
-        v[1] = transferred.x;
-        v[0] = transferred.y;
-        transferred = BitTransferSigned(v[3], v[2]);
-        v[3] = transferred.x;
-        v[2] = transferred.y;
-        transferred = BitTransferSigned(v[5], v[4]);
-        v[5] = transferred.x;
-        v[4] = transferred.y;
-        if ((v[1] + v[3] + v[5]) >= 0) {
-            ep1 = ClampByte(ivec4(0xFF, v[0], v[2], v[4]));
-            ep2 = ClampByte(ivec4(0xFF, v[0] + v[1], v[2] + v[3], v[4] + v[5]));
+        ivec2 transferred = BitTransferSigned(V[0].y, V[0].x);
+        V[0].y = transferred.x;
+        V[0].x = transferred.y;
+        transferred = BitTransferSigned(V[0].w, V[0].z);
+        V[0].w = transferred.x;
+        V[0].z = transferred.y;
+        transferred = BitTransferSigned(V[1].y, V[1].x);
+        V[1].y = transferred.x;
+        V[1].x = transferred.y;
+        if ((V[0].y + V[0].w + V[1].y) >= 0) {
+            ep1 = ClampByte(ivec4(0xFF, V[0].x, V[0].z, V[1].x));
+            ep2 = ClampByte(ivec4(0xFF, V[0].x + V[0].y, V[0].z + V[0].w, V[1].x + V[1].y));
         } else {
-            ep1 = ClampByte(BlueContract(0xFF, v[0] + v[1], v[2] + v[3], v[4] + v[5]));
-            ep2 = ClampByte(BlueContract(0xFF, v[0], v[2], v[4]));
+            ep1 = ClampByte(BlueContract(0xFF, V[0].x + V[0].y, V[0].z + V[0].w, V[1].x + V[1].y));
+            ep2 = ClampByte(BlueContract(0xFF, V[0].x, V[0].z, V[1].x));
         }
         break;
     }
     case 10: {
         READ_UINT_VALUES(6)
-        ep1 = uvec4(v[4], (v[0] * v[3]) >> 8, (v[1] * v[3]) >> 8, (v[2] * v[3]) >> 8);
-        ep2 = uvec4(v[5], v[0], v[1], v[2]);
+        ep1 = uvec4(V[1].x, (V[0].x * V[0].w) >> 8, (V[0].y * V[0].w) >> 8, (V[0].z * V[0].w) >> 8);
+        ep2 = uvec4(V[1].y, V[0].x, V[0].y, V[0].z);
         break;
     }
     case 12: {
         READ_UINT_VALUES(8)
-        if ((v[1] + v[3] + v[5]) >= (v[0] + v[2] + v[4])) {
-            ep1 = uvec4(v[6], v[0], v[2], v[4]);
-            ep2 = uvec4(v[7], v[1], v[3], v[5]);
+        if ((V[0].y + V[0].w + V[1].y) >= (V[0].x + V[0].z + V[1].x)) {
+            ep1 = uvec4(V[1].z, V[0].x, V[0].z, V[1].x);
+            ep2 = uvec4(V[1].w, V[0].y, V[0].w, V[1].y);
         } else {
-            ep1 = uvec4(BlueContract(int(v[7]), int(v[1]), int(v[3]), int(v[5])));
-            ep2 = uvec4(BlueContract(int(v[6]), int(v[0]), int(v[2]), int(v[4])));
+            ep1 = uvec4(BlueContract(int(V[1].w), int(V[0].y), int(V[0].w), int(V[1].y)));
+            ep2 = uvec4(BlueContract(int(V[1].z), int(V[0].x), int(V[0].z), int(V[1].x)));
         }
         break;
     }
     case 13: {
         READ_INT_VALUES(8)
-        ivec2 transferred = BitTransferSigned(v[1], v[0]);
-        v[1] = transferred.x;
-        v[0] = transferred.y;
-        transferred = BitTransferSigned(v[3], v[2]);
-        v[3] = transferred.x;
-        v[2] = transferred.y;
+        ivec2 transferred = BitTransferSigned(V[0].y, V[0].x);
+        V[0].y = transferred.x;
+        V[0].x = transferred.y;
+        transferred = BitTransferSigned(V[0].w, V[0].z);
+        V[0].w = transferred.x;
+        V[0].z = transferred.y;
 
-        transferred = BitTransferSigned(v[5], v[4]);
-        v[5] = transferred.x;
-        v[4] = transferred.y;
+        transferred = BitTransferSigned(V[1].y, V[1].x);
+        V[1].y = transferred.x;
+        V[1].x = transferred.y;
 
-        transferred = BitTransferSigned(v[7], v[6]);
-        v[7] = transferred.x;
-        v[6] = transferred.y;
+        transferred = BitTransferSigned(V[1].w, V[1].z);
+        V[1].w = transferred.x;
+        V[1].z = transferred.y;
 
-        if ((v[1] + v[3] + v[5]) >= 0) {
-            ep1 = ClampByte(ivec4(v[6], v[0], v[2], v[4]));
-            ep2 = ClampByte(ivec4(v[7] + v[6], v[0] + v[1], v[2] + v[3], v[4] + v[5]));
+        if ((V[0].y + V[0].w + V[1].y) >= 0) {
+            ep1 = ClampByte(ivec4(V[1].z, V[0].x, V[0].z, V[1].x));
+            ep2 = ClampByte(ivec4(V[1].w + V[1].z, V[0].x + V[0].y, V[0].z + V[0].w, V[1].x + V[1].y));
         } else {
-            ep1 = ClampByte(BlueContract(v[6] + v[7], v[0] + v[1], v[2] + v[3], v[4] + v[5]));
-            ep2 = ClampByte(BlueContract(v[6], v[0], v[2], v[4]));
+            ep1 = ClampByte(BlueContract(V[1].z + V[1].w, V[0].x + V[0].y, V[0].z + V[0].w, V[1].x + V[1].y));
+            ep2 = ClampByte(BlueContract(V[1].z, V[0].x, V[0].z, V[1].x));
         }
         break;
     }

From 75ac7845ce2f9fe46602eeac03207040659e6bc9 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Tue, 1 Aug 2023 23:33:47 -0400
Subject: [PATCH 20/33] error/void extent funcs

---
 src/video_core/host_shaders/astc_decoder.comp | 91 +++++++++----------
 1 file changed, 43 insertions(+), 48 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index 0e611ede5..63b3d9f44 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -40,9 +40,6 @@ struct TexelWeightParams {
     uvec2 size;
     uint max_weight;
     bool dual_plane;
-    bool error_state;
-    bool void_extent_ldr;
-    bool void_extent_hdr;
 };
 
 layout(binding = BINDING_INPUT_BUFFER, std430) readonly restrict buffer InputBufferU32 {
@@ -983,28 +980,54 @@ int FindLayout(uint mode) {
     return 5;
 }
 
-TexelWeightParams DecodeBlockInfo() {
-    TexelWeightParams params = TexelWeightParams(uvec2(0), 0, false, false, false, false);
-    uint mode = StreamBits(11);
+
+void FillError(ivec3 coord) {
+    for (uint j = 0; j < block_dims.y; j++) {
+        for (uint i = 0; i < block_dims.x; i++) {
+            imageStore(dest_image, coord + ivec3(i, j, 0), vec4(0.0, 0.0, 0.0, 0.0));
+        }
+    }
+}
+
+void FillVoidExtentLDR(ivec3 coord) {
+    SkipBits(52);
+    const uint r_u = StreamBits(16);
+    const uint g_u = StreamBits(16);
+    const uint b_u = StreamBits(16);
+    const uint a_u = StreamBits(16);
+    const float a = float(a_u) / 65535.0f;
+    const float r = float(r_u) / 65535.0f;
+    const float g = float(g_u) / 65535.0f;
+    const float b = float(b_u) / 65535.0f;
+    for (uint j = 0; j < block_dims.y; j++) {
+        for (uint i = 0; i < block_dims.x; i++) {
+            imageStore(dest_image, coord + ivec3(i, j, 0), vec4(r, g, b, a));
+        }
+    }
+}
+
+bool IsError(uint mode) {
     if ((mode & 0x1ff) == 0x1fc) {
         if ((mode & 0x200) != 0) {
-            params.void_extent_hdr = true;
-        } else {
-            params.void_extent_ldr = true;
+            // params.void_extent_hdr = true;
+            return true;
         }
         if ((mode & 0x400) == 0 || StreamBits(1) == 0) {
-            params.error_state = true;
+            return true;
         }
-        return params;
+        return false;
     }
     if ((mode & 0xf) == 0) {
-        params.error_state = true;
-        return params;
+        return true;
     }
     if ((mode & 3) == 0 && (mode & 0x1c0) == 0x1c0) {
-        params.error_state = true;
-        return params;
+        return true;
     }
+    return false;
+}
+
+TexelWeightParams DecodeBlockInfo(uint mode) {
+    TexelWeightParams params = TexelWeightParams(uvec2(0), 0, false);
     uint A, B;
     uint mode_layout = FindLayout(mode);
     switch (mode_layout) {
@@ -1053,7 +1076,6 @@ TexelWeightParams DecodeBlockInfo() {
         params.size = uvec2(A + 6, B + 6);
         break;
     default:
-        params.error_state = true;
         break;
     }
     params.dual_plane = (mode_layout != 9) && ((mode & 0x400) != 0);
@@ -1072,42 +1094,15 @@ TexelWeightParams DecodeBlockInfo() {
     return params;
 }
 
-void FillError(ivec3 coord) {
-    for (uint j = 0; j < block_dims.y; j++) {
-        for (uint i = 0; i < block_dims.x; i++) {
-            imageStore(dest_image, coord + ivec3(i, j, 0), vec4(0.0, 0.0, 0.0, 0.0));
-        }
-    }
-}
-
-void FillVoidExtentLDR(ivec3 coord) {
-    SkipBits(52);
-    const uint r_u = StreamBits(16);
-    const uint g_u = StreamBits(16);
-    const uint b_u = StreamBits(16);
-    const uint a_u = StreamBits(16);
-    const float a = float(a_u) / 65535.0f;
-    const float r = float(r_u) / 65535.0f;
-    const float g = float(g_u) / 65535.0f;
-    const float b = float(b_u) / 65535.0f;
-    for (uint j = 0; j < block_dims.y; j++) {
-        for (uint i = 0; i < block_dims.x; i++) {
-            imageStore(dest_image, coord + ivec3(i, j, 0), vec4(r, g, b, a));
-        }
-    }
-}
-
 void DecompressBlock(ivec3 coord) {
-    const TexelWeightParams params = DecodeBlockInfo();
-    if (params.error_state) {
+    uint mode = StreamBits(11);
+    const TexelWeightParams params = DecodeBlockInfo(mode);
+    if (IsError(mode)) {
         FillError(coord);
         return;
     }
-    if (params.void_extent_hdr) {
-        FillError(coord);
-        return;
-    }
-    if (params.void_extent_ldr) {
+    if ((mode & 0x1ff) == 0x1fc) {
+        // params.void_extent_ldr = true;
         FillVoidExtentLDR(coord);
         return;
     }

From fd2051b401bf0469bf199ce972289abad8087620 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Tue, 1 Aug 2023 23:41:46 -0400
Subject: [PATCH 21/33] remove TexelWeightParams

---
 src/video_core/host_shaders/astc_decoder.comp | 77 ++++++++-----------
 1 file changed, 31 insertions(+), 46 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index 63b3d9f44..f497c4d17 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -36,12 +36,6 @@ struct EncodingData {
     uint data;
 };
 
-struct TexelWeightParams {
-    uvec2 size;
-    uint max_weight;
-    bool dual_plane;
-};
-
 layout(binding = BINDING_INPUT_BUFFER, std430) readonly restrict buffer InputBufferU32 {
     uvec4 astc_data[];
 };
@@ -1026,59 +1020,50 @@ bool IsError(uint mode) {
     return false;
 }
 
-TexelWeightParams DecodeBlockInfo(uint mode) {
-    TexelWeightParams params = TexelWeightParams(uvec2(0), 0, false);
+uvec2 DecodeBlockSize(uint mode) {
     uint A, B;
-    uint mode_layout = FindLayout(mode);
-    switch (mode_layout) {
+    switch (FindLayout(mode)) {
     case 0:
         A = (mode >> 5) & 0x3;
         B = (mode >> 7) & 0x3;
-        params.size = uvec2(B + 4, A + 2);
-        break;
+        return uvec2(B + 4, A + 2);
     case 1:
         A = (mode >> 5) & 0x3;
         B = (mode >> 7) & 0x3;
-        params.size = uvec2(B + 8, A + 2);
-        break;
+        return uvec2(B + 8, A + 2);
     case 2:
         A = (mode >> 5) & 0x3;
         B = (mode >> 7) & 0x3;
-        params.size = uvec2(A + 2, B + 8);
-        break;
+        return uvec2(A + 2, B + 8);
     case 3:
         A = (mode >> 5) & 0x3;
         B = (mode >> 7) & 0x1;
-        params.size = uvec2(A + 2, B + 6);
-        break;
+        return uvec2(A + 2, B + 6);
     case 4:
         A = (mode >> 5) & 0x3;
         B = (mode >> 7) & 0x1;
-        params.size = uvec2(B + 2, A + 2);
-        break;
+        return uvec2(B + 2, A + 2);
     case 5:
         A = (mode >> 5) & 0x3;
-        params.size = uvec2(12, A + 2);
-        break;
+        return uvec2(12, A + 2);
     case 6:
         A = (mode >> 5) & 0x3;
-        params.size = uvec2(A + 2, 12);
-        break;
+        return uvec2(A + 2, 12);
     case 7:
-        params.size = uvec2(6, 10);
-        break;
+        return uvec2(6, 10);
     case 8:
-        params.size = uvec2(10, 6);
-        break;
+        return uvec2(10, 6);
     case 9:
         A = (mode >> 5) & 0x3;
         B = (mode >> 9) & 0x3;
-        params.size = uvec2(A + 6, B + 6);
-        break;
+        return uvec2(A + 6, B + 6);
     default:
-        break;
+        return uvec2(0);
     }
-    params.dual_plane = (mode_layout != 9) && ((mode & 0x400) != 0);
+}
+
+uint DecodeMaxWeight(uint mode) {
+    const uint mode_layout = FindLayout(mode);
     uint weight_index = (mode & 0x10) != 0 ? 1 : 0;
     if (mode_layout < 5) {
         weight_index |= (mode & 0x3) << 1;
@@ -1089,14 +1074,11 @@ TexelWeightParams DecodeBlockInfo(uint mode) {
     if ((mode_layout != 9) && ((mode & 0x200) != 0)) {
         weight_index += 6;
     }
-    params.max_weight = weight_index + 1;
-
-    return params;
+    return weight_index + 1;
 }
 
 void DecompressBlock(ivec3 coord) {
     uint mode = StreamBits(11);
-    const TexelWeightParams params = DecodeBlockInfo(mode);
     if (IsError(mode)) {
         FillError(coord);
         return;
@@ -1106,12 +1088,15 @@ void DecompressBlock(ivec3 coord) {
         FillVoidExtentLDR(coord);
         return;
     }
-    if ((params.size.x > block_dims.x) || (params.size.y > block_dims.y)) {
+    const uvec2 size_params = DecodeBlockSize(mode);
+    if ((size_params.x > block_dims.x) || (size_params.y > block_dims.y)) {
         FillError(coord);
         return;
     }
     const uint num_partitions = StreamBits(2) + 1;
-    if (num_partitions > 4 || (num_partitions == 4 && params.dual_plane)) {
+    const uint mode_layout = FindLayout(mode);
+    const bool dual_plane = (mode_layout != 9) && ((mode & 0x400) != 0);
+    if (num_partitions > 4 || (num_partitions == 4 && dual_plane)) {
         FillError(coord);
         return;
     }
@@ -1127,7 +1112,8 @@ void DecompressBlock(ivec3 coord) {
         base_cem = StreamBits(6);
     }
     const uint base_mode = base_cem & 3;
-    const uint weight_bits = GetPackedBitSize(params.size, params.dual_plane, params.max_weight);
+    const uint max_weight = DecodeMaxWeight(mode);
+    const uint weight_bits = GetPackedBitSize(size_params, dual_plane, max_weight);
     uint remaining_bits = 128 - weight_bits - total_bitsread;
     uint extra_cem_bits = 0;
     if (base_mode > 0) {
@@ -1146,7 +1132,7 @@ void DecompressBlock(ivec3 coord) {
         }
     }
     remaining_bits -= extra_cem_bits;
-    const uint plane_selector_bits = params.dual_plane ? 2 : 0;
+    const uint plane_selector_bits = dual_plane ? 2 : 0;
     remaining_bits -= plane_selector_bits;
     if (remaining_bits > 128) {
         // Bad data, more remaining bits than 4 bytes
@@ -1198,7 +1184,6 @@ void DecompressBlock(ivec3 coord) {
         // This decode phase should at most push 32 elements into the vector
         result_vector_max_index = 32;
 
-        // uvec4 color_values[8];
         uint colvals_index = 0;
         DecodeColorValues(color_endpoint_mode, num_partitions, color_data_bits);
         for (uint i = 0; i < num_partitions; i++) {
@@ -1226,13 +1211,13 @@ void DecompressBlock(ivec3 coord) {
     result_limit_reached = false;
 
     // The limit for the Unquantize phase, avoids decoding more data than needed.
-    result_vector_max_index = params.size.x * params.size.y;
-    if (params.dual_plane) {
+    result_vector_max_index = size_params.x * size_params.y;
+    if (dual_plane) {
         result_vector_max_index *= 2;
     }
-    DecodeIntegerSequence(params.max_weight, GetNumWeightValues(params.size, params.dual_plane));
+    DecodeIntegerSequence(max_weight, GetNumWeightValues(size_params, dual_plane));
 
-    UnquantizeTexelWeights(params.size, params.dual_plane);
+    UnquantizeTexelWeights(size_params, dual_plane);
     for (uint j = 0; j < block_dims.y; j++) {
         for (uint i = 0; i < block_dims.x; i++) {
             uint local_partition = 0;
@@ -1247,7 +1232,7 @@ void DecompressBlock(ivec3 coord) {
             const uint vector_index = weight_offset % 4;
             const uint primary_weight = unquantized_texel_weights[array_index][vector_index];
             uvec4 weight_vec = uvec4(primary_weight);
-            if (params.dual_plane) {
+            if (dual_plane) {
                 const uint secondary_weight_offset = (j * block_dims.x + i) + ARRAY_NUM_ELEMENTS;
                 const uint secondary_array_index = secondary_weight_offset / 4;
                 const uint secondary_vector_index = secondary_weight_offset % 4;

From 7a0d7e7668b13ecf67760843f5a72cf681d8ea34 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Tue, 1 Aug 2023 23:47:10 -0400
Subject: [PATCH 22/33] small_block opt

---
 src/video_core/host_shaders/astc_decoder.comp | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index f497c4d17..a33c916ac 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -264,8 +264,8 @@ uint Hash52(uint p) {
     return p;
 }
 
-uint Select2DPartition(uint seed, uint x, uint y, uint partition_count, bool small_block) {
-    if (small_block) {
+uint Select2DPartition(uint seed, uint x, uint y, uint partition_count) {
+    if ((block_dims.y * block_dims.x) < 32) {
         x <<= 1;
         y <<= 1;
     }
@@ -1222,8 +1222,7 @@ void DecompressBlock(ivec3 coord) {
         for (uint i = 0; i < block_dims.x; i++) {
             uint local_partition = 0;
             if (num_partitions > 1) {
-                local_partition = Select2DPartition(partition_index, i, j, num_partitions,
-                                                    (block_dims.y * block_dims.x) < 32);
+                local_partition = Select2DPartition(partition_index, i, j, num_partitions);
             }
             const uvec4 C0 = ReplicateByteTo16(endpoints0[local_partition]);
             const uvec4 C1 = ReplicateByteTo16(endpoints1[local_partition]);

From 5a78b35b1abf071bd62b1ff8d0cb939bd30a549f Mon Sep 17 00:00:00 2001
From: ameerj <52414509+ameerj@users.noreply.github.com>
Date: Wed, 28 Jun 2023 01:24:52 -0400
Subject: [PATCH 23/33] vulkan dims specialization

---
 src/video_core/host_shaders/CMakeLists.txt    |  57 ++++++++++
 src/video_core/host_shaders/astc_decoder.comp |  42 ++++---
 .../host_shaders/astc_decoder_spv_includes.h  |  20 ++++
 .../renderer_vulkan/vk_compute_pass.cpp       | 104 ++++++++++++++++--
 .../renderer_vulkan/vk_compute_pass.h         |   2 +
 5 files changed, 198 insertions(+), 27 deletions(-)
 create mode 100644 src/video_core/host_shaders/astc_decoder_spv_includes.h

diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt
index e61d9af80..20e8388ee 100644
--- a/src/video_core/host_shaders/CMakeLists.txt
+++ b/src/video_core/host_shaders/CMakeLists.txt
@@ -13,6 +13,11 @@ set(GLSL_INCLUDES
     ${FIDELITYFX_FILES}
 )
 
+set(ASTC_INCLUDES
+    # astc_decoder_glsl_includes.h
+    astc_decoder_spv_includes.h
+)
+
 set(SHADER_FILES
     astc_decoder.comp
     blit_color_float.frag
@@ -95,9 +100,60 @@ if (NOT GLSLANG_ERROR STREQUAL "")
     set(QUIET_FLAG "")
 endif()
 
+macro(ASTC_GEN)
+    # paired list of valid astc block dimensions
+    set(ASTC_WIDTHS  4 5 5 6 6 6 8 8 8 10 10 10 10 12 12)
+    set(ASTC_HEIGHTS 4 4 5 4 5 6 5 6 8  5  6  8 10 10 12)
+    list(LENGTH ASTC_WIDTHS NUM_ASTC_FORMATS)
+    math(EXPR NUM_ASTC_FORMATS "${NUM_ASTC_FORMATS}-1")
+    foreach(i RANGE ${NUM_ASTC_FORMATS})
+        list(GET ASTC_WIDTHS ${i} ASTC_WIDTH)
+        list(GET ASTC_HEIGHTS ${i} ASTC_HEIGHT)
+
+        # Vulkan SPIR-V Specialization
+
+        string(TOUPPER ${SHADER_NAME}_${ASTC_WIDTH}x${ASTC_HEIGHT}_SPV SPIRV_VARIABLE_NAME)
+        set(SPIRV_HEADER_FILE ${SHADER_DIR}/${SHADER_NAME}_${ASTC_WIDTH}x${ASTC_HEIGHT}_spv.h)
+        add_custom_command(
+            OUTPUT
+                ${SPIRV_HEADER_FILE}
+            COMMAND
+                ${GLSLANGVALIDATOR} -V -DBLOCK_WIDTH=${ASTC_WIDTH} -DBLOCK_HEIGHT=${ASTC_HEIGHT} ${QUIET_FLAG} ${GLSL_FLAGS} --variable-name ${SPIRV_VARIABLE_NAME} -o ${SPIRV_HEADER_FILE} ${SOURCE_FILE}
+            MAIN_DEPENDENCY
+                ${SOURCE_FILE}
+        )
+        set(SHADER_HEADERS ${SHADER_HEADERS} ${SPIRV_HEADER_FILE})
+
+        # GLSL Specialization
+        # Disabled as there was no noticeable performance uplift specializing the shaders for OGL
+
+        # set(SOURCE_HEADER_FILE ${SHADER_DIR}/${SHADER_NAME}_${ASTC_WIDTH}x${ASTC_HEIGHT}.h)
+        # set(SHADER_DEFINES "#define BLOCK_WIDTH ${ASTC_WIDTH}" "#define BLOCK_HEIGHT ${ASTC_HEIGHT}")
+        # set(DEFINES_LINE_NUMBER 14)
+        # string(TOUPPER ${SHADER_NAME}_${ASTC_WIDTH}x${ASTC_HEIGHT} GLSL_VARIABLE_NAME)
+        # add_custom_command(
+        #     OUTPUT
+        #         ${SOURCE_HEADER_FILE}
+        #     COMMAND
+        #         ${CMAKE_COMMAND} -P ${HEADER_GENERATOR} ${SOURCE_FILE} ${SOURCE_HEADER_FILE} ${INPUT_FILE} "${SHADER_DEFINES}" ${DEFINES_LINE_NUMBER} ${GLSL_VARIABLE_NAME}
+        #     MAIN_DEPENDENCY
+        #         ${SOURCE_FILE}
+        #     DEPENDS
+        #         ${INPUT_FILE}
+        #         ${SOURCE_FILE}
+        # )
+        # set(SHADER_HEADERS ${SHADER_HEADERS} ${SOURCE_HEADER_FILE})
+    endforeach()
+endmacro()
+
 foreach(FILENAME IN ITEMS ${SHADER_FILES})
     string(REPLACE "." "_" SHADER_NAME ${FILENAME})
     set(SOURCE_FILE ${CMAKE_CURRENT_SOURCE_DIR}/${FILENAME})
+
+    if (${FILENAME} MATCHES "astc_decoder.comp")
+        ASTC_GEN()
+    endif()
+
     # Skip generating source headers on Vulkan exclusive files
     if (NOT ${FILENAME} MATCHES "vulkan.*")
         set(SOURCE_HEADER_FILE ${SHADER_DIR}/${SHADER_NAME}.h)
@@ -151,6 +207,7 @@ endforeach()
 
 set(SHADER_SOURCES ${SHADER_FILES})
 list(APPEND SHADER_SOURCES ${GLSL_INCLUDES})
+list(APPEND SHADER_SOURCES ${ASTC_INCLUDES})
 
 add_custom_target(host_shaders
     DEPENDS
diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index a33c916ac..b4bb8299f 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -24,7 +24,9 @@
 layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
 
 BEGIN_PUSH_CONSTANTS
+#ifndef BLOCK_WIDTH
 UNIFORM(1) uvec2 block_dims;
+#endif
 UNIFORM(2) uint layer_stride;
 UNIFORM(3) uint block_size;
 UNIFORM(4) uint x_shift;
@@ -75,7 +77,15 @@ int color_bitsread = 0;
 // At most will require BLOCK_WIDTH x BLOCK_HEIGHT x 2 in dual plane mode
 // So the maximum would be 144 (12 x 12) elements, x 2 for two planes
 #define DIVCEIL(number, divisor) (number + divisor - 1) / divisor
+
+#ifndef BLOCK_WIDTH
+#define BLOCK_WIDTH block_dims.x
+#define BLOCK_HEIGHT block_dims.y
 #define ARRAY_NUM_ELEMENTS 144
+#else
+#define ARRAY_NUM_ELEMENTS BLOCK_WIDTH * BLOCK_HEIGHT
+#endif
+
 #define VECTOR_ARRAY_SIZE DIVCEIL(ARRAY_NUM_ELEMENTS * 2, 4)
 uvec4 result_vector[VECTOR_ARRAY_SIZE];
 
@@ -265,7 +275,7 @@ uint Hash52(uint p) {
 }
 
 uint Select2DPartition(uint seed, uint x, uint y, uint partition_count) {
-    if ((block_dims.y * block_dims.x) < 32) {
+    if ((BLOCK_WIDTH * BLOCK_HEIGHT) < 32) {
         x <<= 1;
         y <<= 1;
     }
@@ -878,8 +888,8 @@ uint UnquantizeTexelWeight(EncodingData val) {
 uvec4 unquantized_texel_weights[VECTOR_ARRAY_SIZE];
 
 void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
-    const uint Ds = uint((block_dims.x * 0.5f + 1024) / (block_dims.x - 1));
-    const uint Dt = uint((block_dims.y * 0.5f + 1024) / (block_dims.y - 1));
+    const uint Ds = uint((BLOCK_WIDTH * 0.5f + 1024) / (BLOCK_WIDTH - 1));
+    const uint Dt = uint((BLOCK_HEIGHT * 0.5f + 1024) / (BLOCK_HEIGHT - 1));
     const uint num_planes = is_dual_plane ? 2 : 1;
     const uint area = size.x * size.y;
     const uint loop_count = min(result_index, area * num_planes);
@@ -890,8 +900,8 @@ void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
             UnquantizeTexelWeight(GetEncodingFromVector(itr));
     }
     for (uint plane = 0; plane < num_planes; ++plane) {
-        for (uint t = 0; t < block_dims.y; t++) {
-            for (uint s = 0; s < block_dims.x; s++) {
+        for (uint t = 0; t < BLOCK_HEIGHT; t++) {
+            for (uint s = 0; s < BLOCK_WIDTH; s++) {
                 const uint cs = Ds * s;
                 const uint ct = Dt * t;
                 const uint gs = (cs * (size.x - 1) + 32) >> 6;
@@ -934,7 +944,7 @@ void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
                     VectorIndicesFromBase(offset_base);
                     p.w = result_vector[array_index][vector_index];
                 }
-                const uint offset = (t * block_dims.x + s) + ARRAY_NUM_ELEMENTS * plane;
+                const uint offset = (t * BLOCK_WIDTH + s) + ARRAY_NUM_ELEMENTS * plane;
                 const uint array_index = offset / 4;
                 const uint vector_index = offset % 4;
                 unquantized_texel_weights[array_index][vector_index] = (uint(dot(p, w)) + 8) >> 4;
@@ -976,8 +986,8 @@ int FindLayout(uint mode) {
 
 
 void FillError(ivec3 coord) {
-    for (uint j = 0; j < block_dims.y; j++) {
-        for (uint i = 0; i < block_dims.x; i++) {
+    for (uint j = 0; j < BLOCK_HEIGHT; j++) {
+        for (uint i = 0; i < BLOCK_WIDTH; i++) {
             imageStore(dest_image, coord + ivec3(i, j, 0), vec4(0.0, 0.0, 0.0, 0.0));
         }
     }
@@ -993,8 +1003,8 @@ void FillVoidExtentLDR(ivec3 coord) {
     const float r = float(r_u) / 65535.0f;
     const float g = float(g_u) / 65535.0f;
     const float b = float(b_u) / 65535.0f;
-    for (uint j = 0; j < block_dims.y; j++) {
-        for (uint i = 0; i < block_dims.x; i++) {
+    for (uint j = 0; j < BLOCK_HEIGHT; j++) {
+        for (uint i = 0; i < BLOCK_WIDTH; i++) {
             imageStore(dest_image, coord + ivec3(i, j, 0), vec4(r, g, b, a));
         }
     }
@@ -1089,7 +1099,7 @@ void DecompressBlock(ivec3 coord) {
         return;
     }
     const uvec2 size_params = DecodeBlockSize(mode);
-    if ((size_params.x > block_dims.x) || (size_params.y > block_dims.y)) {
+    if ((size_params.x > BLOCK_WIDTH) || (size_params.y > BLOCK_HEIGHT)) {
         FillError(coord);
         return;
     }
@@ -1218,21 +1228,21 @@ void DecompressBlock(ivec3 coord) {
     DecodeIntegerSequence(max_weight, GetNumWeightValues(size_params, dual_plane));
 
     UnquantizeTexelWeights(size_params, dual_plane);
-    for (uint j = 0; j < block_dims.y; j++) {
-        for (uint i = 0; i < block_dims.x; i++) {
+    for (uint j = 0; j < BLOCK_HEIGHT; j++) {
+        for (uint i = 0; i < BLOCK_WIDTH; i++) {
             uint local_partition = 0;
             if (num_partitions > 1) {
                 local_partition = Select2DPartition(partition_index, i, j, num_partitions);
             }
             const uvec4 C0 = ReplicateByteTo16(endpoints0[local_partition]);
             const uvec4 C1 = ReplicateByteTo16(endpoints1[local_partition]);
-            const uint weight_offset = (j * block_dims.x + i);
+            const uint weight_offset = (j * BLOCK_WIDTH + i);
             const uint array_index = weight_offset / 4;
             const uint vector_index = weight_offset % 4;
             const uint primary_weight = unquantized_texel_weights[array_index][vector_index];
             uvec4 weight_vec = uvec4(primary_weight);
             if (dual_plane) {
-                const uint secondary_weight_offset = (j * block_dims.x + i) + ARRAY_NUM_ELEMENTS;
+                const uint secondary_weight_offset = (j * BLOCK_WIDTH + i) + ARRAY_NUM_ELEMENTS;
                 const uint secondary_array_index = secondary_weight_offset / 4;
                 const uint secondary_vector_index = secondary_weight_offset % 4;
                 const uint secondary_weight =
@@ -1270,7 +1280,7 @@ void main() {
     offset += (pos.x >> GOB_SIZE_X_SHIFT) << x_shift;
     offset += swizzle;
 
-    const ivec3 coord = ivec3(gl_GlobalInvocationID * uvec3(block_dims, 1));
+    const ivec3 coord = ivec3(gl_GlobalInvocationID * uvec3(BLOCK_WIDTH, BLOCK_HEIGHT, 1));
     if (any(greaterThanEqual(coord, imageSize(dest_image)))) {
         return;
     }
diff --git a/src/video_core/host_shaders/astc_decoder_spv_includes.h b/src/video_core/host_shaders/astc_decoder_spv_includes.h
new file mode 100644
index 000000000..44ee50c5f
--- /dev/null
+++ b/src/video_core/host_shaders/astc_decoder_spv_includes.h
@@ -0,0 +1,20 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include "video_core/host_shaders/astc_decoder_comp_10x10_spv.h"
+#include "video_core/host_shaders/astc_decoder_comp_10x5_spv.h"
+#include "video_core/host_shaders/astc_decoder_comp_10x6_spv.h"
+#include "video_core/host_shaders/astc_decoder_comp_10x8_spv.h"
+#include "video_core/host_shaders/astc_decoder_comp_12x10_spv.h"
+#include "video_core/host_shaders/astc_decoder_comp_12x12_spv.h"
+#include "video_core/host_shaders/astc_decoder_comp_4x4_spv.h"
+#include "video_core/host_shaders/astc_decoder_comp_5x4_spv.h"
+#include "video_core/host_shaders/astc_decoder_comp_5x5_spv.h"
+#include "video_core/host_shaders/astc_decoder_comp_6x5_spv.h"
+#include "video_core/host_shaders/astc_decoder_comp_6x6_spv.h"
+#include "video_core/host_shaders/astc_decoder_comp_8x5_spv.h"
+#include "video_core/host_shaders/astc_decoder_comp_8x6_spv.h"
+#include "video_core/host_shaders/astc_decoder_comp_8x8_spv.h"
+#include "video_core/host_shaders/astc_decoder_comp_spv.h"
diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
index 54ee030ce..f22342252 100644
--- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
@@ -11,7 +11,7 @@
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "common/div_ceil.h"
-#include "video_core/host_shaders/astc_decoder_comp_spv.h"
+#include "video_core/host_shaders/astc_decoder_spv_includes.h"
 #include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h"
 #include "video_core/host_shaders/vulkan_uint8_comp_spv.h"
 #include "video_core/renderer_vulkan/vk_compute_pass.h"
@@ -124,13 +124,62 @@ constexpr std::array<VkDescriptorUpdateTemplateEntry, ASTC_NUM_BINDINGS>
     }};
 
 struct AstcPushConstants {
-    std::array<u32, 2> blocks_dims;
     u32 layer_stride;
     u32 block_size;
     u32 x_shift;
     u32 block_height;
     u32 block_height_mask;
 };
+
+size_t AstcFormatIndex(VideoCore::Surface::PixelFormat format) {
+    switch (format) {
+    case VideoCore::Surface::PixelFormat::ASTC_2D_4X4_SRGB:
+    case VideoCore::Surface::PixelFormat::ASTC_2D_4X4_UNORM:
+        return 0;
+    case VideoCore::Surface::PixelFormat::ASTC_2D_5X4_SRGB:
+    case VideoCore::Surface::PixelFormat::ASTC_2D_5X4_UNORM:
+        return 1;
+    case VideoCore::Surface::PixelFormat::ASTC_2D_5X5_SRGB:
+    case VideoCore::Surface::PixelFormat::ASTC_2D_5X5_UNORM:
+        return 2;
+    case VideoCore::Surface::PixelFormat::ASTC_2D_6X5_SRGB:
+    case VideoCore::Surface::PixelFormat::ASTC_2D_6X5_UNORM:
+        return 3;
+    case VideoCore::Surface::PixelFormat::ASTC_2D_6X6_SRGB:
+    case VideoCore::Surface::PixelFormat::ASTC_2D_6X6_UNORM:
+        return 4;
+    case VideoCore::Surface::PixelFormat::ASTC_2D_8X5_SRGB:
+    case VideoCore::Surface::PixelFormat::ASTC_2D_8X5_UNORM:
+        return 5;
+    case VideoCore::Surface::PixelFormat::ASTC_2D_8X6_SRGB:
+    case VideoCore::Surface::PixelFormat::ASTC_2D_8X6_UNORM:
+        return 6;
+    case VideoCore::Surface::PixelFormat::ASTC_2D_8X8_SRGB:
+    case VideoCore::Surface::PixelFormat::ASTC_2D_8X8_UNORM:
+        return 7;
+    case VideoCore::Surface::PixelFormat::ASTC_2D_10X5_SRGB:
+    case VideoCore::Surface::PixelFormat::ASTC_2D_10X5_UNORM:
+        return 8;
+    case VideoCore::Surface::PixelFormat::ASTC_2D_10X6_SRGB:
+    case VideoCore::Surface::PixelFormat::ASTC_2D_10X6_UNORM:
+        return 9;
+    case VideoCore::Surface::PixelFormat::ASTC_2D_10X8_SRGB:
+    case VideoCore::Surface::PixelFormat::ASTC_2D_10X8_UNORM:
+        return 10;
+    case VideoCore::Surface::PixelFormat::ASTC_2D_10X10_SRGB:
+    case VideoCore::Surface::PixelFormat::ASTC_2D_10X10_UNORM:
+        return 11;
+    case VideoCore::Surface::PixelFormat::ASTC_2D_12X10_SRGB:
+    case VideoCore::Surface::PixelFormat::ASTC_2D_12X10_UNORM:
+        return 12;
+    case VideoCore::Surface::PixelFormat::ASTC_2D_12X12_SRGB:
+    case VideoCore::Surface::PixelFormat::ASTC_2D_12X12_UNORM:
+        return 13;
+    default:
+        UNREACHABLE();
+        return 0;
+    }
+}
 } // Anonymous namespace
 
 ComputePass::ComputePass(const Device& device_, DescriptorPool& descriptor_pool,
@@ -312,19 +361,53 @@ ASTCDecoderPass::ASTCDecoderPass(const Device& device_, Scheduler& scheduler_,
                   COMPUTE_PUSH_CONSTANT_RANGE<sizeof(AstcPushConstants)>, ASTC_DECODER_COMP_SPV),
       scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
       compute_pass_descriptor_queue{compute_pass_descriptor_queue_}, memory_allocator{
-                                                                         memory_allocator_} {}
+                                                                         memory_allocator_} {
+    // These must match the order found in AstcFormatIndex
+    static constexpr std::array<std::span<const u32>, 14> ASTC_SHADERS{
+        ASTC_DECODER_COMP_4X4_SPV,   ASTC_DECODER_COMP_5X4_SPV,   ASTC_DECODER_COMP_5X5_SPV,
+        ASTC_DECODER_COMP_6X5_SPV,   ASTC_DECODER_COMP_6X6_SPV,   ASTC_DECODER_COMP_8X5_SPV,
+        ASTC_DECODER_COMP_8X6_SPV,   ASTC_DECODER_COMP_8X8_SPV,   ASTC_DECODER_COMP_10X5_SPV,
+        ASTC_DECODER_COMP_10X6_SPV,  ASTC_DECODER_COMP_10X8_SPV,  ASTC_DECODER_COMP_10X10_SPV,
+        ASTC_DECODER_COMP_12X10_SPV, ASTC_DECODER_COMP_12X12_SPV,
+    };
+    for (size_t index = 0; index < ASTC_SHADERS.size(); ++index) {
+        const auto& code = ASTC_SHADERS[index];
+        const auto module_ = device.GetLogical().CreateShaderModule({
+            .sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
+            .pNext = nullptr,
+            .flags = 0,
+            .codeSize = static_cast<u32>(code.size_bytes()),
+            .pCode = code.data(),
+        });
+        device.SaveShader(code);
+        astc_pipelines[index] = device.GetLogical().CreateComputePipeline({
+            .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
+            .pNext = nullptr,
+            .flags = 0,
+            .stage{
+                .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
+                .pNext = nullptr,
+                .flags = 0,
+                .stage = VK_SHADER_STAGE_COMPUTE_BIT,
+                .module = *module_,
+                .pName = "main",
+                .pSpecializationInfo = nullptr,
+            },
+            .layout = *layout,
+            .basePipelineHandle = nullptr,
+            .basePipelineIndex = 0,
+        });
+    }
+}
 
 ASTCDecoderPass::~ASTCDecoderPass() = default;
 
 void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
                                std::span<const VideoCommon::SwizzleParameters> swizzles) {
     using namespace VideoCommon::Accelerated;
-    const std::array<u32, 2> block_dims{
-        VideoCore::Surface::DefaultBlockWidth(image.info.format),
-        VideoCore::Surface::DefaultBlockHeight(image.info.format),
-    };
+
     scheduler.RequestOutsideRenderPassOperationContext();
-    const VkPipeline vk_pipeline = *pipeline;
+    const VkPipeline vk_pipeline = *astc_pipelines[AstcFormatIndex(image.info.format)];
     const VkImageAspectFlags aspect_mask = image.AspectMask();
     const VkImage vk_image = image.Handle();
     const bool is_initialized = image.ExchangeInitialization();
@@ -371,10 +454,9 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
         ASSERT(params.origin == (std::array<u32, 3>{0, 0, 0}));
         ASSERT(params.destination == (std::array<s32, 3>{0, 0, 0}));
         ASSERT(params.bytes_per_block_log2 == 4);
-        scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, block_dims,
-                          params, descriptor_data](vk::CommandBuffer cmdbuf) {
+        scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, params,
+                          descriptor_data](vk::CommandBuffer cmdbuf) {
             const AstcPushConstants uniforms{
-                .blocks_dims = block_dims,
                 .layer_stride = params.layer_stride,
                 .block_size = params.block_size,
                 .x_shift = params.x_shift,
diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h
index dd3927376..1f264bea6 100644
--- a/src/video_core/renderer_vulkan/vk_compute_pass.h
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.h
@@ -95,6 +95,8 @@ public:
                   std::span<const VideoCommon::SwizzleParameters> swizzles);
 
 private:
+    std::array<vk::Pipeline, 14> astc_pipelines;
+
     Scheduler& scheduler;
     StagingBufferPool& staging_buffer_pool;
     ComputePassDescriptorQueue& compute_pass_descriptor_queue;

From 8ce158bce6912b2263f1724e6c09d8b517ef18c3 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Wed, 2 Aug 2023 18:15:59 -0400
Subject: [PATCH 24/33] Revert "vulkan dims specialization"

This reverts commit e6243058f2269bd79ac8479d58e55feec2611e9d.
---
 src/video_core/host_shaders/CMakeLists.txt    |  57 ----------
 src/video_core/host_shaders/astc_decoder.comp |  42 +++----
 .../host_shaders/astc_decoder_spv_includes.h  |  20 ----
 .../renderer_vulkan/vk_compute_pass.cpp       | 104 ++----------------
 .../renderer_vulkan/vk_compute_pass.h         |   2 -
 5 files changed, 27 insertions(+), 198 deletions(-)
 delete mode 100644 src/video_core/host_shaders/astc_decoder_spv_includes.h

diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt
index 20e8388ee..e61d9af80 100644
--- a/src/video_core/host_shaders/CMakeLists.txt
+++ b/src/video_core/host_shaders/CMakeLists.txt
@@ -13,11 +13,6 @@ set(GLSL_INCLUDES
     ${FIDELITYFX_FILES}
 )
 
-set(ASTC_INCLUDES
-    # astc_decoder_glsl_includes.h
-    astc_decoder_spv_includes.h
-)
-
 set(SHADER_FILES
     astc_decoder.comp
     blit_color_float.frag
@@ -100,60 +95,9 @@ if (NOT GLSLANG_ERROR STREQUAL "")
     set(QUIET_FLAG "")
 endif()
 
-macro(ASTC_GEN)
-    # paired list of valid astc block dimensions
-    set(ASTC_WIDTHS  4 5 5 6 6 6 8 8 8 10 10 10 10 12 12)
-    set(ASTC_HEIGHTS 4 4 5 4 5 6 5 6 8  5  6  8 10 10 12)
-    list(LENGTH ASTC_WIDTHS NUM_ASTC_FORMATS)
-    math(EXPR NUM_ASTC_FORMATS "${NUM_ASTC_FORMATS}-1")
-    foreach(i RANGE ${NUM_ASTC_FORMATS})
-        list(GET ASTC_WIDTHS ${i} ASTC_WIDTH)
-        list(GET ASTC_HEIGHTS ${i} ASTC_HEIGHT)
-
-        # Vulkan SPIR-V Specialization
-
-        string(TOUPPER ${SHADER_NAME}_${ASTC_WIDTH}x${ASTC_HEIGHT}_SPV SPIRV_VARIABLE_NAME)
-        set(SPIRV_HEADER_FILE ${SHADER_DIR}/${SHADER_NAME}_${ASTC_WIDTH}x${ASTC_HEIGHT}_spv.h)
-        add_custom_command(
-            OUTPUT
-                ${SPIRV_HEADER_FILE}
-            COMMAND
-                ${GLSLANGVALIDATOR} -V -DBLOCK_WIDTH=${ASTC_WIDTH} -DBLOCK_HEIGHT=${ASTC_HEIGHT} ${QUIET_FLAG} ${GLSL_FLAGS} --variable-name ${SPIRV_VARIABLE_NAME} -o ${SPIRV_HEADER_FILE} ${SOURCE_FILE}
-            MAIN_DEPENDENCY
-                ${SOURCE_FILE}
-        )
-        set(SHADER_HEADERS ${SHADER_HEADERS} ${SPIRV_HEADER_FILE})
-
-        # GLSL Specialization
-        # Disabled as there was no noticeable performance uplift specializing the shaders for OGL
-
-        # set(SOURCE_HEADER_FILE ${SHADER_DIR}/${SHADER_NAME}_${ASTC_WIDTH}x${ASTC_HEIGHT}.h)
-        # set(SHADER_DEFINES "#define BLOCK_WIDTH ${ASTC_WIDTH}" "#define BLOCK_HEIGHT ${ASTC_HEIGHT}")
-        # set(DEFINES_LINE_NUMBER 14)
-        # string(TOUPPER ${SHADER_NAME}_${ASTC_WIDTH}x${ASTC_HEIGHT} GLSL_VARIABLE_NAME)
-        # add_custom_command(
-        #     OUTPUT
-        #         ${SOURCE_HEADER_FILE}
-        #     COMMAND
-        #         ${CMAKE_COMMAND} -P ${HEADER_GENERATOR} ${SOURCE_FILE} ${SOURCE_HEADER_FILE} ${INPUT_FILE} "${SHADER_DEFINES}" ${DEFINES_LINE_NUMBER} ${GLSL_VARIABLE_NAME}
-        #     MAIN_DEPENDENCY
-        #         ${SOURCE_FILE}
-        #     DEPENDS
-        #         ${INPUT_FILE}
-        #         ${SOURCE_FILE}
-        # )
-        # set(SHADER_HEADERS ${SHADER_HEADERS} ${SOURCE_HEADER_FILE})
-    endforeach()
-endmacro()
-
 foreach(FILENAME IN ITEMS ${SHADER_FILES})
     string(REPLACE "." "_" SHADER_NAME ${FILENAME})
     set(SOURCE_FILE ${CMAKE_CURRENT_SOURCE_DIR}/${FILENAME})
-
-    if (${FILENAME} MATCHES "astc_decoder.comp")
-        ASTC_GEN()
-    endif()
-
     # Skip generating source headers on Vulkan exclusive files
     if (NOT ${FILENAME} MATCHES "vulkan.*")
         set(SOURCE_HEADER_FILE ${SHADER_DIR}/${SHADER_NAME}.h)
@@ -207,7 +151,6 @@ endforeach()
 
 set(SHADER_SOURCES ${SHADER_FILES})
 list(APPEND SHADER_SOURCES ${GLSL_INCLUDES})
-list(APPEND SHADER_SOURCES ${ASTC_INCLUDES})
 
 add_custom_target(host_shaders
     DEPENDS
diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index b4bb8299f..a33c916ac 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -24,9 +24,7 @@
 layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
 
 BEGIN_PUSH_CONSTANTS
-#ifndef BLOCK_WIDTH
 UNIFORM(1) uvec2 block_dims;
-#endif
 UNIFORM(2) uint layer_stride;
 UNIFORM(3) uint block_size;
 UNIFORM(4) uint x_shift;
@@ -77,15 +75,7 @@ int color_bitsread = 0;
 // At most will require BLOCK_WIDTH x BLOCK_HEIGHT x 2 in dual plane mode
 // So the maximum would be 144 (12 x 12) elements, x 2 for two planes
 #define DIVCEIL(number, divisor) (number + divisor - 1) / divisor
-
-#ifndef BLOCK_WIDTH
-#define BLOCK_WIDTH block_dims.x
-#define BLOCK_HEIGHT block_dims.y
 #define ARRAY_NUM_ELEMENTS 144
-#else
-#define ARRAY_NUM_ELEMENTS BLOCK_WIDTH * BLOCK_HEIGHT
-#endif
-
 #define VECTOR_ARRAY_SIZE DIVCEIL(ARRAY_NUM_ELEMENTS * 2, 4)
 uvec4 result_vector[VECTOR_ARRAY_SIZE];
 
@@ -275,7 +265,7 @@ uint Hash52(uint p) {
 }
 
 uint Select2DPartition(uint seed, uint x, uint y, uint partition_count) {
-    if ((BLOCK_WIDTH * BLOCK_HEIGHT) < 32) {
+    if ((block_dims.y * block_dims.x) < 32) {
         x <<= 1;
         y <<= 1;
     }
@@ -888,8 +878,8 @@ uint UnquantizeTexelWeight(EncodingData val) {
 uvec4 unquantized_texel_weights[VECTOR_ARRAY_SIZE];
 
 void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
-    const uint Ds = uint((BLOCK_WIDTH * 0.5f + 1024) / (BLOCK_WIDTH - 1));
-    const uint Dt = uint((BLOCK_HEIGHT * 0.5f + 1024) / (BLOCK_HEIGHT - 1));
+    const uint Ds = uint((block_dims.x * 0.5f + 1024) / (block_dims.x - 1));
+    const uint Dt = uint((block_dims.y * 0.5f + 1024) / (block_dims.y - 1));
     const uint num_planes = is_dual_plane ? 2 : 1;
     const uint area = size.x * size.y;
     const uint loop_count = min(result_index, area * num_planes);
@@ -900,8 +890,8 @@ void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
             UnquantizeTexelWeight(GetEncodingFromVector(itr));
     }
     for (uint plane = 0; plane < num_planes; ++plane) {
-        for (uint t = 0; t < BLOCK_HEIGHT; t++) {
-            for (uint s = 0; s < BLOCK_WIDTH; s++) {
+        for (uint t = 0; t < block_dims.y; t++) {
+            for (uint s = 0; s < block_dims.x; s++) {
                 const uint cs = Ds * s;
                 const uint ct = Dt * t;
                 const uint gs = (cs * (size.x - 1) + 32) >> 6;
@@ -944,7 +934,7 @@ void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
                     VectorIndicesFromBase(offset_base);
                     p.w = result_vector[array_index][vector_index];
                 }
-                const uint offset = (t * BLOCK_WIDTH + s) + ARRAY_NUM_ELEMENTS * plane;
+                const uint offset = (t * block_dims.x + s) + ARRAY_NUM_ELEMENTS * plane;
                 const uint array_index = offset / 4;
                 const uint vector_index = offset % 4;
                 unquantized_texel_weights[array_index][vector_index] = (uint(dot(p, w)) + 8) >> 4;
@@ -986,8 +976,8 @@ int FindLayout(uint mode) {
 
 
 void FillError(ivec3 coord) {
-    for (uint j = 0; j < BLOCK_HEIGHT; j++) {
-        for (uint i = 0; i < BLOCK_WIDTH; i++) {
+    for (uint j = 0; j < block_dims.y; j++) {
+        for (uint i = 0; i < block_dims.x; i++) {
             imageStore(dest_image, coord + ivec3(i, j, 0), vec4(0.0, 0.0, 0.0, 0.0));
         }
     }
@@ -1003,8 +993,8 @@ void FillVoidExtentLDR(ivec3 coord) {
     const float r = float(r_u) / 65535.0f;
     const float g = float(g_u) / 65535.0f;
     const float b = float(b_u) / 65535.0f;
-    for (uint j = 0; j < BLOCK_HEIGHT; j++) {
-        for (uint i = 0; i < BLOCK_WIDTH; i++) {
+    for (uint j = 0; j < block_dims.y; j++) {
+        for (uint i = 0; i < block_dims.x; i++) {
             imageStore(dest_image, coord + ivec3(i, j, 0), vec4(r, g, b, a));
         }
     }
@@ -1099,7 +1089,7 @@ void DecompressBlock(ivec3 coord) {
         return;
     }
     const uvec2 size_params = DecodeBlockSize(mode);
-    if ((size_params.x > BLOCK_WIDTH) || (size_params.y > BLOCK_HEIGHT)) {
+    if ((size_params.x > block_dims.x) || (size_params.y > block_dims.y)) {
         FillError(coord);
         return;
     }
@@ -1228,21 +1218,21 @@ void DecompressBlock(ivec3 coord) {
     DecodeIntegerSequence(max_weight, GetNumWeightValues(size_params, dual_plane));
 
     UnquantizeTexelWeights(size_params, dual_plane);
-    for (uint j = 0; j < BLOCK_HEIGHT; j++) {
-        for (uint i = 0; i < BLOCK_WIDTH; i++) {
+    for (uint j = 0; j < block_dims.y; j++) {
+        for (uint i = 0; i < block_dims.x; i++) {
             uint local_partition = 0;
             if (num_partitions > 1) {
                 local_partition = Select2DPartition(partition_index, i, j, num_partitions);
             }
             const uvec4 C0 = ReplicateByteTo16(endpoints0[local_partition]);
             const uvec4 C1 = ReplicateByteTo16(endpoints1[local_partition]);
-            const uint weight_offset = (j * BLOCK_WIDTH + i);
+            const uint weight_offset = (j * block_dims.x + i);
             const uint array_index = weight_offset / 4;
             const uint vector_index = weight_offset % 4;
             const uint primary_weight = unquantized_texel_weights[array_index][vector_index];
             uvec4 weight_vec = uvec4(primary_weight);
             if (dual_plane) {
-                const uint secondary_weight_offset = (j * BLOCK_WIDTH + i) + ARRAY_NUM_ELEMENTS;
+                const uint secondary_weight_offset = (j * block_dims.x + i) + ARRAY_NUM_ELEMENTS;
                 const uint secondary_array_index = secondary_weight_offset / 4;
                 const uint secondary_vector_index = secondary_weight_offset % 4;
                 const uint secondary_weight =
@@ -1280,7 +1270,7 @@ void main() {
     offset += (pos.x >> GOB_SIZE_X_SHIFT) << x_shift;
     offset += swizzle;
 
-    const ivec3 coord = ivec3(gl_GlobalInvocationID * uvec3(BLOCK_WIDTH, BLOCK_HEIGHT, 1));
+    const ivec3 coord = ivec3(gl_GlobalInvocationID * uvec3(block_dims, 1));
     if (any(greaterThanEqual(coord, imageSize(dest_image)))) {
         return;
     }
diff --git a/src/video_core/host_shaders/astc_decoder_spv_includes.h b/src/video_core/host_shaders/astc_decoder_spv_includes.h
deleted file mode 100644
index 44ee50c5f..000000000
--- a/src/video_core/host_shaders/astc_decoder_spv_includes.h
+++ /dev/null
@@ -1,20 +0,0 @@
-// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
-
-#pragma once
-
-#include "video_core/host_shaders/astc_decoder_comp_10x10_spv.h"
-#include "video_core/host_shaders/astc_decoder_comp_10x5_spv.h"
-#include "video_core/host_shaders/astc_decoder_comp_10x6_spv.h"
-#include "video_core/host_shaders/astc_decoder_comp_10x8_spv.h"
-#include "video_core/host_shaders/astc_decoder_comp_12x10_spv.h"
-#include "video_core/host_shaders/astc_decoder_comp_12x12_spv.h"
-#include "video_core/host_shaders/astc_decoder_comp_4x4_spv.h"
-#include "video_core/host_shaders/astc_decoder_comp_5x4_spv.h"
-#include "video_core/host_shaders/astc_decoder_comp_5x5_spv.h"
-#include "video_core/host_shaders/astc_decoder_comp_6x5_spv.h"
-#include "video_core/host_shaders/astc_decoder_comp_6x6_spv.h"
-#include "video_core/host_shaders/astc_decoder_comp_8x5_spv.h"
-#include "video_core/host_shaders/astc_decoder_comp_8x6_spv.h"
-#include "video_core/host_shaders/astc_decoder_comp_8x8_spv.h"
-#include "video_core/host_shaders/astc_decoder_comp_spv.h"
diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
index f22342252..54ee030ce 100644
--- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
@@ -11,7 +11,7 @@
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "common/div_ceil.h"
-#include "video_core/host_shaders/astc_decoder_spv_includes.h"
+#include "video_core/host_shaders/astc_decoder_comp_spv.h"
 #include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h"
 #include "video_core/host_shaders/vulkan_uint8_comp_spv.h"
 #include "video_core/renderer_vulkan/vk_compute_pass.h"
@@ -124,62 +124,13 @@ constexpr std::array<VkDescriptorUpdateTemplateEntry, ASTC_NUM_BINDINGS>
     }};
 
 struct AstcPushConstants {
+    std::array<u32, 2> blocks_dims;
     u32 layer_stride;
     u32 block_size;
     u32 x_shift;
     u32 block_height;
     u32 block_height_mask;
 };
-
-size_t AstcFormatIndex(VideoCore::Surface::PixelFormat format) {
-    switch (format) {
-    case VideoCore::Surface::PixelFormat::ASTC_2D_4X4_SRGB:
-    case VideoCore::Surface::PixelFormat::ASTC_2D_4X4_UNORM:
-        return 0;
-    case VideoCore::Surface::PixelFormat::ASTC_2D_5X4_SRGB:
-    case VideoCore::Surface::PixelFormat::ASTC_2D_5X4_UNORM:
-        return 1;
-    case VideoCore::Surface::PixelFormat::ASTC_2D_5X5_SRGB:
-    case VideoCore::Surface::PixelFormat::ASTC_2D_5X5_UNORM:
-        return 2;
-    case VideoCore::Surface::PixelFormat::ASTC_2D_6X5_SRGB:
-    case VideoCore::Surface::PixelFormat::ASTC_2D_6X5_UNORM:
-        return 3;
-    case VideoCore::Surface::PixelFormat::ASTC_2D_6X6_SRGB:
-    case VideoCore::Surface::PixelFormat::ASTC_2D_6X6_UNORM:
-        return 4;
-    case VideoCore::Surface::PixelFormat::ASTC_2D_8X5_SRGB:
-    case VideoCore::Surface::PixelFormat::ASTC_2D_8X5_UNORM:
-        return 5;
-    case VideoCore::Surface::PixelFormat::ASTC_2D_8X6_SRGB:
-    case VideoCore::Surface::PixelFormat::ASTC_2D_8X6_UNORM:
-        return 6;
-    case VideoCore::Surface::PixelFormat::ASTC_2D_8X8_SRGB:
-    case VideoCore::Surface::PixelFormat::ASTC_2D_8X8_UNORM:
-        return 7;
-    case VideoCore::Surface::PixelFormat::ASTC_2D_10X5_SRGB:
-    case VideoCore::Surface::PixelFormat::ASTC_2D_10X5_UNORM:
-        return 8;
-    case VideoCore::Surface::PixelFormat::ASTC_2D_10X6_SRGB:
-    case VideoCore::Surface::PixelFormat::ASTC_2D_10X6_UNORM:
-        return 9;
-    case VideoCore::Surface::PixelFormat::ASTC_2D_10X8_SRGB:
-    case VideoCore::Surface::PixelFormat::ASTC_2D_10X8_UNORM:
-        return 10;
-    case VideoCore::Surface::PixelFormat::ASTC_2D_10X10_SRGB:
-    case VideoCore::Surface::PixelFormat::ASTC_2D_10X10_UNORM:
-        return 11;
-    case VideoCore::Surface::PixelFormat::ASTC_2D_12X10_SRGB:
-    case VideoCore::Surface::PixelFormat::ASTC_2D_12X10_UNORM:
-        return 12;
-    case VideoCore::Surface::PixelFormat::ASTC_2D_12X12_SRGB:
-    case VideoCore::Surface::PixelFormat::ASTC_2D_12X12_UNORM:
-        return 13;
-    default:
-        UNREACHABLE();
-        return 0;
-    }
-}
 } // Anonymous namespace
 
 ComputePass::ComputePass(const Device& device_, DescriptorPool& descriptor_pool,
@@ -361,53 +312,19 @@ ASTCDecoderPass::ASTCDecoderPass(const Device& device_, Scheduler& scheduler_,
                   COMPUTE_PUSH_CONSTANT_RANGE<sizeof(AstcPushConstants)>, ASTC_DECODER_COMP_SPV),
       scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
       compute_pass_descriptor_queue{compute_pass_descriptor_queue_}, memory_allocator{
-                                                                         memory_allocator_} {
-    // These must match the order found in AstcFormatIndex
-    static constexpr std::array<std::span<const u32>, 14> ASTC_SHADERS{
-        ASTC_DECODER_COMP_4X4_SPV,   ASTC_DECODER_COMP_5X4_SPV,   ASTC_DECODER_COMP_5X5_SPV,
-        ASTC_DECODER_COMP_6X5_SPV,   ASTC_DECODER_COMP_6X6_SPV,   ASTC_DECODER_COMP_8X5_SPV,
-        ASTC_DECODER_COMP_8X6_SPV,   ASTC_DECODER_COMP_8X8_SPV,   ASTC_DECODER_COMP_10X5_SPV,
-        ASTC_DECODER_COMP_10X6_SPV,  ASTC_DECODER_COMP_10X8_SPV,  ASTC_DECODER_COMP_10X10_SPV,
-        ASTC_DECODER_COMP_12X10_SPV, ASTC_DECODER_COMP_12X12_SPV,
-    };
-    for (size_t index = 0; index < ASTC_SHADERS.size(); ++index) {
-        const auto& code = ASTC_SHADERS[index];
-        const auto module_ = device.GetLogical().CreateShaderModule({
-            .sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
-            .pNext = nullptr,
-            .flags = 0,
-            .codeSize = static_cast<u32>(code.size_bytes()),
-            .pCode = code.data(),
-        });
-        device.SaveShader(code);
-        astc_pipelines[index] = device.GetLogical().CreateComputePipeline({
-            .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
-            .pNext = nullptr,
-            .flags = 0,
-            .stage{
-                .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
-                .pNext = nullptr,
-                .flags = 0,
-                .stage = VK_SHADER_STAGE_COMPUTE_BIT,
-                .module = *module_,
-                .pName = "main",
-                .pSpecializationInfo = nullptr,
-            },
-            .layout = *layout,
-            .basePipelineHandle = nullptr,
-            .basePipelineIndex = 0,
-        });
-    }
-}
+                                                                         memory_allocator_} {}
 
 ASTCDecoderPass::~ASTCDecoderPass() = default;
 
 void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
                                std::span<const VideoCommon::SwizzleParameters> swizzles) {
     using namespace VideoCommon::Accelerated;
-
+    const std::array<u32, 2> block_dims{
+        VideoCore::Surface::DefaultBlockWidth(image.info.format),
+        VideoCore::Surface::DefaultBlockHeight(image.info.format),
+    };
     scheduler.RequestOutsideRenderPassOperationContext();
-    const VkPipeline vk_pipeline = *astc_pipelines[AstcFormatIndex(image.info.format)];
+    const VkPipeline vk_pipeline = *pipeline;
     const VkImageAspectFlags aspect_mask = image.AspectMask();
     const VkImage vk_image = image.Handle();
     const bool is_initialized = image.ExchangeInitialization();
@@ -454,9 +371,10 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
         ASSERT(params.origin == (std::array<u32, 3>{0, 0, 0}));
         ASSERT(params.destination == (std::array<s32, 3>{0, 0, 0}));
         ASSERT(params.bytes_per_block_log2 == 4);
-        scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, params,
-                          descriptor_data](vk::CommandBuffer cmdbuf) {
+        scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, block_dims,
+                          params, descriptor_data](vk::CommandBuffer cmdbuf) {
             const AstcPushConstants uniforms{
+                .blocks_dims = block_dims,
                 .layer_stride = params.layer_stride,
                 .block_size = params.block_size,
                 .x_shift = params.x_shift,
diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h
index 1f264bea6..dd3927376 100644
--- a/src/video_core/renderer_vulkan/vk_compute_pass.h
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.h
@@ -95,8 +95,6 @@ public:
                   std::span<const VideoCommon::SwizzleParameters> swizzles);
 
 private:
-    std::array<vk::Pipeline, 14> astc_pipelines;
-
     Scheduler& scheduler;
     StagingBufferPool& staging_buffer_pool;
     ComputePassDescriptorQueue& compute_pass_descriptor_queue;

From b36e645fee2104501be7e426132159dfc97eea05 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Wed, 2 Aug 2023 18:33:43 -0400
Subject: [PATCH 25/33] undo uint

---
 src/video_core/host_shaders/astc_decoder.comp | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index a33c916ac..547cd881f 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -79,7 +79,7 @@ int color_bitsread = 0;
 #define VECTOR_ARRAY_SIZE DIVCEIL(ARRAY_NUM_ELEMENTS * 2, 4)
 uvec4 result_vector[VECTOR_ARRAY_SIZE];
 
-uint result_index = 0;
+int result_index = 0;
 uint result_vector_max_index;
 bool result_limit_reached = false;
 
@@ -122,8 +122,8 @@ void ResultEmplaceBack(EncodingData val) {
         result_limit_reached = true;
         return;
     }
-    const uint array_index = result_index / 4u;
-    const uint vector_index = result_index % 4u;
+    const uint array_index = result_index / 4;
+    const uint vector_index = result_index % 4;
     result_vector[array_index][vector_index] = val.data;
     ++result_index;
 }

From 31a0cff0367bddb51785145fba1089b625e06b62 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Sun, 6 Aug 2023 12:18:55 -0400
Subject: [PATCH 26/33] minor

---
 src/video_core/host_shaders/astc_decoder.comp | 18 ++++++------------
 1 file changed, 6 insertions(+), 12 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index 547cd881f..9d9532a98 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -809,18 +809,15 @@ uint UnquantizeTexelWeight(EncodingData val) {
     const uint A = ReplicateBitTo7((bitval & 1));
     uint B = 0, C = 0, D = 0;
     uint result = 0;
+    const uint bitlen_0_results[5] = {0, 16, 32, 48, 64};
     switch (encoding) {
     case JUST_BITS:
-        result = FastReplicateTo6(bitval, bitlen);
-        break;
+        return FastReplicateTo6(bitval, bitlen);
     case TRIT: {
         D = QuintTritValue(val);
         switch (bitlen) {
-        case 0: {
-            const uint results[3] = {0, 32, 63};
-            result = results[D];
-            break;
-        }
+        case 0:
+            return bitlen_0_results[D * 2];
         case 1: {
             C = 50;
             break;
@@ -845,11 +842,8 @@ uint UnquantizeTexelWeight(EncodingData val) {
     case QUINT: {
         D = QuintTritValue(val);
         switch (bitlen) {
-        case 0: {
-            const uint results[5] = {0, 16, 32, 47, 63};
-            result = results[D];
-            break;
-        }
+        case 0:
+            return bitlen_0_results[D];
         case 1: {
             C = 28;
             break;

From 913803bf653989d2d08ddd39caf2216b55275dfc Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Sun, 6 Aug 2023 13:32:35 -0400
Subject: [PATCH 27/33] Compute Replicate

---
 src/video_core/host_shaders/astc_decoder.comp | 105 ++++--------------
 1 file changed, 20 insertions(+), 85 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index 9d9532a98..5e922d1fe 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -140,98 +140,33 @@ uint ReplicateBitTo9(uint value) {
     return value * 511;
 }
 
-uint FastReplicateTo8(uint value, uint num_bits) {
-    if (value == 0) {
+uint ReplicateBits(uint value, uint num_bits, uint to_bit) {
+    if (value == 0 || num_bits == 0) {
         return 0;
     }
-    const uint array_index = value / 4;
-    const uint vector_index = bitfieldExtract(value, 0, 2);
-    switch (num_bits) {
-    case 1:
-        return 255;
-    case 2: {
-        const uvec4 REPLICATE_2_BIT_TO_8_TABLE = (uvec4(0, 85, 170, 255));
-        return REPLICATE_2_BIT_TO_8_TABLE[vector_index];
+    if (num_bits >= to_bit) {
+        return value;
     }
-    case 3: {
-        const uvec4 REPLICATE_3_BIT_TO_8_TABLE[2] =
-            uvec4[](uvec4(0, 36, 73, 109), uvec4(146, 182, 219, 255));
-        return REPLICATE_3_BIT_TO_8_TABLE[array_index][vector_index];
+    const uint v = value & uint((1 << num_bits) - 1);
+    uint res = v;
+    uint reslen = num_bits;
+    while (reslen < to_bit) {
+        const uint num_dst_bits_to_shift_up = min(num_bits, to_bit - reslen);
+        const uint num_src_bits_to_shift_down = num_bits - num_dst_bits_to_shift_up;
+
+        res <<= num_dst_bits_to_shift_up;
+        res |= (v >> num_src_bits_to_shift_down);
+        reslen += num_bits;
     }
-    case 4: {
-        const uvec4 REPLICATE_4_BIT_TO_8_TABLE[4] =
-            uvec4[](uvec4(0, 17, 34, 51), uvec4(68, 85, 102, 119), uvec4(136, 153, 170, 187),
-                    uvec4(204, 221, 238, 255));
-        return REPLICATE_4_BIT_TO_8_TABLE[array_index][vector_index];
-    }
-    case 5: {
-        const uvec4 REPLICATE_5_BIT_TO_8_TABLE[8] =
-            uvec4[](uvec4(0, 8, 16, 24), uvec4(33, 41, 49, 57), uvec4(66, 74, 82, 90),
-                    uvec4(99, 107, 115, 123), uvec4(132, 140, 148, 156), uvec4(165, 173, 181, 189),
-                    uvec4(198, 206, 214, 222), uvec4(231, 239, 247, 255));
-        return REPLICATE_5_BIT_TO_8_TABLE[array_index][vector_index];
-    }
-    case 6: {
-        const uvec4 REPLICATE_6_BIT_TO_8_TABLE[16] = uvec4[](
-            uvec4(0, 4, 8, 12), uvec4(16, 20, 24, 28), uvec4(32, 36, 40, 44), uvec4(48, 52, 56, 60),
-            uvec4(65, 69, 73, 77), uvec4(81, 85, 89, 93), uvec4(97, 101, 105, 109),
-            uvec4(113, 117, 121, 125), uvec4(130, 134, 138, 142), uvec4(146, 150, 154, 158),
-            uvec4(162, 166, 170, 174), uvec4(178, 182, 186, 190), uvec4(195, 199, 203, 207),
-            uvec4(211, 215, 219, 223), uvec4(227, 231, 235, 239), uvec4(243, 247, 251, 255));
-        return REPLICATE_6_BIT_TO_8_TABLE[array_index][vector_index];
-    }
-    case 7: {
-        const uvec4 REPLICATE_7_BIT_TO_8_TABLE[32] =
-            uvec4[](uvec4(0, 2, 4, 6), uvec4(8, 10, 12, 14), uvec4(16, 18, 20, 22),
-                    uvec4(24, 26, 28, 30), uvec4(32, 34, 36, 38), uvec4(40, 42, 44, 46),
-                    uvec4(48, 50, 52, 54), uvec4(56, 58, 60, 62), uvec4(64, 66, 68, 70),
-                    uvec4(72, 74, 76, 78), uvec4(80, 82, 84, 86), uvec4(88, 90, 92, 94),
-                    uvec4(96, 98, 100, 102), uvec4(104, 106, 108, 110), uvec4(112, 114, 116, 118),
-                    uvec4(120, 122, 124, 126), uvec4(129, 131, 133, 135), uvec4(137, 139, 141, 143),
-                    uvec4(145, 147, 149, 151), uvec4(153, 155, 157, 159), uvec4(161, 163, 165, 167),
-                    uvec4(169, 171, 173, 175), uvec4(177, 179, 181, 183), uvec4(185, 187, 189, 191),
-                    uvec4(193, 195, 197, 199), uvec4(201, 203, 205, 207), uvec4(209, 211, 213, 215),
-                    uvec4(217, 219, 221, 223), uvec4(225, 227, 229, 231), uvec4(233, 235, 237, 239),
-                    uvec4(241, 243, 245, 247), uvec4(249, 251, 253, 255));
-        return REPLICATE_7_BIT_TO_8_TABLE[array_index][vector_index];
-    }
-    }
-    return value;
+    return res;
+}
+
+uint FastReplicateTo8(uint value, uint num_bits) {
+    return ReplicateBits(value, num_bits, 8);
 }
 
 uint FastReplicateTo6(uint value, uint num_bits) {
-    if (value == 0) {
-        return 0;
-    }
-    const uint array_index = value / 4;
-    const uint vector_index = bitfieldExtract(value, 0, 2);
-    switch (num_bits) {
-    case 1:
-        return 63;
-    case 2: {
-        const uvec4 REPLICATE_2_BIT_TO_6_TABLE = uvec4(0, 21, 42, 63);
-        return REPLICATE_2_BIT_TO_6_TABLE[vector_index];
-    }
-    case 3: {
-        const uvec4 REPLICATE_3_BIT_TO_6_TABLE[2] =
-            uvec4[](uvec4(0, 9, 18, 27), uvec4(36, 45, 54, 63));
-        return REPLICATE_3_BIT_TO_6_TABLE[array_index][vector_index];
-    }
-    case 4: {
-        const uvec4 REPLICATE_4_BIT_TO_6_TABLE[4] =
-            uvec4[](uvec4(0, 4, 8, 12), uvec4(17, 21, 25, 29), uvec4(34, 38, 42, 46),
-                    uvec4(51, 55, 59, 63));
-        return REPLICATE_4_BIT_TO_6_TABLE[array_index][vector_index];
-    }
-    case 5: {
-        const uvec4 REPLICATE_5_BIT_TO_6_TABLE[8] =
-            uvec4[](uvec4(0, 2, 4, 6), uvec4(8, 10, 12, 14), uvec4(16, 18, 20, 22),
-                    uvec4(24, 26, 28, 30), uvec4(33, 35, 37, 39), uvec4(41, 43, 45, 47),
-                    uvec4(49, 51, 53, 55), uvec4(57, 59, 61, 63));
-        return REPLICATE_5_BIT_TO_6_TABLE[array_index][vector_index];
-    }
-    }
-    return value;
+    return ReplicateBits(value, num_bits, 6);
 }
 
 uint Div3Floor(uint v) {

From b18c1fb1bbbcc235e9a1cbbb2704fed2e5895e61 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Sun, 6 Aug 2023 14:49:34 -0400
Subject: [PATCH 28/33] HACK: Avoid swizzling and reuploading ASTC image every
 frame

---
 .../renderer_opengl/gl_texture_cache.cpp           |  9 ++++++++-
 .../renderer_vulkan/vk_staging_buffer_pool.cpp     |  1 +
 .../renderer_vulkan/vk_texture_cache.cpp           | 10 +++++++++-
 src/video_core/texture_cache/texture_cache.h       | 14 ++++++++++++--
 src/video_core/textures/decoders.cpp               |  9 +++++++++
 5 files changed, 39 insertions(+), 4 deletions(-)

diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp
index 9cafd2983..9ca544ec9 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp
@@ -554,7 +554,14 @@ void TextureCacheRuntime::Finish() {
 }
 
 StagingBufferMap TextureCacheRuntime::UploadStagingBuffer(size_t size) {
-    return staging_buffer_pool.RequestUploadBuffer(size);
+    static StagingBufferMap result;
+    static size_t last_size = 0;
+    if (size == last_size) {
+        return result;
+    }
+    last_size = size;
+    result = staging_buffer_pool.RequestUploadBuffer(size);
+    return result;
 }
 
 StagingBufferMap TextureCacheRuntime::DownloadStagingBuffer(size_t size) {
diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
index ce92f66ab..27328dab2 100644
--- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
+++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
@@ -214,6 +214,7 @@ StagingBufferPool::StagingBuffersCache& StagingBufferPool::GetCache(MemoryUsage
 }
 
 void StagingBufferPool::ReleaseCache(MemoryUsage usage) {
+    return;
     ReleaseLevel(GetCache(usage), current_delete_level);
 }
 
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
index b3e17c332..51be57b19 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@@ -846,7 +846,15 @@ void TextureCacheRuntime::Finish() {
 }
 
 StagingBufferRef TextureCacheRuntime::UploadStagingBuffer(size_t size) {
-    return staging_buffer_pool.Request(size, MemoryUsage::Upload);
+    static StagingBufferRef result;
+    static size_t last_size = 0;
+    if (size == last_size) {
+        return result;
+    }
+    LOG_ERROR(Debug, "Called");
+    last_size = size;
+    result = staging_buffer_pool.Request(size, MemoryUsage::Upload);
+    return result;
 }
 
 StagingBufferRef TextureCacheRuntime::DownloadStagingBuffer(size_t size, bool deferred) {
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index 4457b366f..6b87987de 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -1021,8 +1021,18 @@ void TextureCache<P>::UploadImageContents(Image& image, StagingBuffer& staging)
     const GPUVAddr gpu_addr = image.gpu_addr;
 
     if (True(image.flags & ImageFlagBits::AcceleratedUpload)) {
-        gpu_memory->ReadBlock(gpu_addr, mapped_span.data(), mapped_span.size_bytes(),
-                              VideoCommon::CacheType::NoTextureCache);
+        static u64 last_size = 0;
+        bool has_run = false;
+        if (last_size == image.unswizzled_size_bytes) {
+            has_run = true;
+        }
+        last_size = image.unswizzled_size_bytes;
+
+        if (!has_run) {
+            LOG_ERROR(Debug, "Called");
+            gpu_memory->ReadBlock(gpu_addr, mapped_span.data(), mapped_span.size_bytes(),
+                                  VideoCommon::CacheType::NoTextureCache);
+        }
         const auto uploads = FullUploadSwizzles(image.info);
         runtime.AccelerateImageUpload(image, staging, uploads);
         return;
diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp
index 95bcdd37b..dde4b0904 100644
--- a/src/video_core/textures/decoders.cpp
+++ b/src/video_core/textures/decoders.cpp
@@ -88,6 +88,7 @@ void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32
     }
 }
 
+
 template <bool TO_LINEAR, u32 BYTES_PER_PIXEL>
 void SwizzleSubrectImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 height,
                         u32 depth, u32 origin_x, u32 origin_y, u32 extent_x, u32 num_lines,
@@ -95,6 +96,14 @@ void SwizzleSubrectImpl(std::span<u8> output, std::span<const u8> input, u32 wid
     // 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_z = 0;
+    static u32 last_width = 0;
+    static u32 last_height = 0;
+    if (last_width == width && last_height == height) {
+        return;
+    }
+    LOG_ERROR(Debug, "Called");
+    last_width = width;
+    last_height = height;
 
     // We can configure here a custom pitch
     // As it's not exposed 'width * BYTES_PER_PIXEL' will be the expected pitch.

From 9058486b9b9833d0554d3e3f11a582859657b21f Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Sun, 6 Aug 2023 14:55:05 -0400
Subject: [PATCH 29/33] Revert "HACK: Avoid swizzling and reuploading ASTC
 image every frame"

This reverts commit b18c1fb1bbbcc235e9a1cbbb2704fed2e5895e61.
---
 .../renderer_opengl/gl_texture_cache.cpp           |  9 +--------
 .../renderer_vulkan/vk_staging_buffer_pool.cpp     |  1 -
 .../renderer_vulkan/vk_texture_cache.cpp           | 10 +---------
 src/video_core/texture_cache/texture_cache.h       | 14 ++------------
 src/video_core/textures/decoders.cpp               |  9 ---------
 5 files changed, 4 insertions(+), 39 deletions(-)

diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp
index 9ca544ec9..9cafd2983 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp
@@ -554,14 +554,7 @@ void TextureCacheRuntime::Finish() {
 }
 
 StagingBufferMap TextureCacheRuntime::UploadStagingBuffer(size_t size) {
-    static StagingBufferMap result;
-    static size_t last_size = 0;
-    if (size == last_size) {
-        return result;
-    }
-    last_size = size;
-    result = staging_buffer_pool.RequestUploadBuffer(size);
-    return result;
+    return staging_buffer_pool.RequestUploadBuffer(size);
 }
 
 StagingBufferMap TextureCacheRuntime::DownloadStagingBuffer(size_t size) {
diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
index 27328dab2..ce92f66ab 100644
--- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
+++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
@@ -214,7 +214,6 @@ StagingBufferPool::StagingBuffersCache& StagingBufferPool::GetCache(MemoryUsage
 }
 
 void StagingBufferPool::ReleaseCache(MemoryUsage usage) {
-    return;
     ReleaseLevel(GetCache(usage), current_delete_level);
 }
 
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
index 51be57b19..b3e17c332 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@@ -846,15 +846,7 @@ void TextureCacheRuntime::Finish() {
 }
 
 StagingBufferRef TextureCacheRuntime::UploadStagingBuffer(size_t size) {
-    static StagingBufferRef result;
-    static size_t last_size = 0;
-    if (size == last_size) {
-        return result;
-    }
-    LOG_ERROR(Debug, "Called");
-    last_size = size;
-    result = staging_buffer_pool.Request(size, MemoryUsage::Upload);
-    return result;
+    return staging_buffer_pool.Request(size, MemoryUsage::Upload);
 }
 
 StagingBufferRef TextureCacheRuntime::DownloadStagingBuffer(size_t size, bool deferred) {
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index 6b87987de..4457b366f 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -1021,18 +1021,8 @@ void TextureCache<P>::UploadImageContents(Image& image, StagingBuffer& staging)
     const GPUVAddr gpu_addr = image.gpu_addr;
 
     if (True(image.flags & ImageFlagBits::AcceleratedUpload)) {
-        static u64 last_size = 0;
-        bool has_run = false;
-        if (last_size == image.unswizzled_size_bytes) {
-            has_run = true;
-        }
-        last_size = image.unswizzled_size_bytes;
-
-        if (!has_run) {
-            LOG_ERROR(Debug, "Called");
-            gpu_memory->ReadBlock(gpu_addr, mapped_span.data(), mapped_span.size_bytes(),
-                                  VideoCommon::CacheType::NoTextureCache);
-        }
+        gpu_memory->ReadBlock(gpu_addr, mapped_span.data(), mapped_span.size_bytes(),
+                              VideoCommon::CacheType::NoTextureCache);
         const auto uploads = FullUploadSwizzles(image.info);
         runtime.AccelerateImageUpload(image, staging, uploads);
         return;
diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp
index dde4b0904..95bcdd37b 100644
--- a/src/video_core/textures/decoders.cpp
+++ b/src/video_core/textures/decoders.cpp
@@ -88,7 +88,6 @@ void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32
     }
 }
 
-
 template <bool TO_LINEAR, u32 BYTES_PER_PIXEL>
 void SwizzleSubrectImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 height,
                         u32 depth, u32 origin_x, u32 origin_y, u32 extent_x, u32 num_lines,
@@ -96,14 +95,6 @@ void SwizzleSubrectImpl(std::span<u8> output, std::span<const u8> input, u32 wid
     // 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_z = 0;
-    static u32 last_width = 0;
-    static u32 last_height = 0;
-    if (last_width == width && last_height == height) {
-        return;
-    }
-    LOG_ERROR(Debug, "Called");
-    last_width = width;
-    last_height = height;
 
     // We can configure here a custom pitch
     // As it's not exposed 'width * BYTES_PER_PIXEL' will be the expected pitch.

From 70f8ffb787a8bf394dfdc33420da3e6632472e5f Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Wed, 9 Aug 2023 17:21:33 -0400
Subject: [PATCH 30/33] GetUnquantizedWeightVector

---
 src/video_core/host_shaders/astc_decoder.comp | 130 +++++++++---------
 1 file changed, 62 insertions(+), 68 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index 5e922d1fe..4014d4bfe 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -804,11 +804,7 @@ uint UnquantizeTexelWeight(EncodingData val) {
     return result;
 }
 
-uvec4 unquantized_texel_weights[VECTOR_ARRAY_SIZE];
-
 void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
-    const uint Ds = uint((block_dims.x * 0.5f + 1024) / (block_dims.x - 1));
-    const uint Dt = uint((block_dims.y * 0.5f + 1024) / (block_dims.y - 1));
     const uint num_planes = is_dual_plane ? 2 : 1;
     const uint area = size.x * size.y;
     const uint loop_count = min(result_index, area * num_planes);
@@ -818,58 +814,71 @@ void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
         result_vector[array_index][vector_index] =
             UnquantizeTexelWeight(GetEncodingFromVector(itr));
     }
-    for (uint plane = 0; plane < num_planes; ++plane) {
-        for (uint t = 0; t < block_dims.y; t++) {
-            for (uint s = 0; s < block_dims.x; s++) {
-                const uint cs = Ds * s;
-                const uint ct = Dt * t;
-                const uint gs = (cs * (size.x - 1) + 32) >> 6;
-                const uint gt = (ct * (size.y - 1) + 32) >> 6;
-                const uint js = gs >> 4;
-                const uint fs = gs & 0xF;
-                const uint jt = gt >> 4;
-                const uint ft = gt & 0x0F;
-                const uint w11 = (fs * ft + 8) >> 4;
-                const uint w10 = ft - w11;
-                const uint w01 = fs - w11;
-                const uint w00 = 16 - fs - ft + w11;
-                const uvec4 w = uvec4(w00, w01, w10, w11);
-                const uint v0 = jt * size.x + js;
+}
 
-                uvec4 p = uvec4(0);
-
-#define VectorIndicesFromBase(offset_base)                                                         \
-    const uint offset = is_dual_plane ? 2 * offset_base + plane : offset_base;                     \
-    const uint array_index = offset / 4;                                                           \
+uint GetUnquantizedTexelWieght(uint offset_base, uint plane, bool is_dual_plane) {
+    const uint offset = is_dual_plane ? 2 * offset_base + plane : offset_base;
+    const uint array_index = offset / 4;
     const uint vector_index = offset % 4;
+    return result_vector[array_index][vector_index];
+}
 
-                if (v0 < area) {
-                    const uint offset_base = v0;
-                    VectorIndicesFromBase(offset_base);
-                    p.x = result_vector[array_index][vector_index];
-                }
-                if ((v0 + 1) < (area)) {
-                    const uint offset_base = v0 + 1;
-                    VectorIndicesFromBase(offset_base);
-                    p.y = result_vector[array_index][vector_index];
-                }
-                if ((v0 + size.x) < (area)) {
-                    const uint offset_base = v0 + size.x;
-                    VectorIndicesFromBase(offset_base);
-                    p.z = result_vector[array_index][vector_index];
-                }
-                if ((v0 + size.x + 1) < (area)) {
-                    const uint offset_base = v0 + size.x + 1;
-                    VectorIndicesFromBase(offset_base);
-                    p.w = result_vector[array_index][vector_index];
-                }
-                const uint offset = (t * block_dims.x + s) + ARRAY_NUM_ELEMENTS * plane;
-                const uint array_index = offset / 4;
-                const uint vector_index = offset % 4;
-                unquantized_texel_weights[array_index][vector_index] = (uint(dot(p, w)) + 8) >> 4;
-            }
+uvec4 GetUnquantizedWeightVector(uint t, uint s, uvec2 size, uint plane_index, bool is_dual_plane) {
+    const uint Ds = uint((block_dims.x * 0.5f + 1024) / (block_dims.x - 1));
+    const uint Dt = uint((block_dims.y * 0.5f + 1024) / (block_dims.y - 1));
+    const uint area = size.x * size.y;
+
+    const uint cs = Ds * s;
+    const uint ct = Dt * t;
+    const uint gs = (cs * (size.x - 1) + 32) >> 6;
+    const uint gt = (ct * (size.y - 1) + 32) >> 6;
+    const uint js = gs >> 4;
+    const uint fs = gs & 0xF;
+    const uint jt = gt >> 4;
+    const uint ft = gt & 0x0F;
+    const uint w11 = (fs * ft + 8) >> 4;
+    const uint w10 = ft - w11;
+    const uint w01 = fs - w11;
+    const uint w00 = 16 - fs - ft + w11;
+    const uvec4 w = uvec4(w00, w01, w10, w11);
+    const uint v0 = jt * size.x + js;
+
+    uvec4 p0 = uvec4(0);
+    uvec4 p1 = uvec4(0);
+
+    if (v0 < area) {
+        const uint offset_base = v0;
+        p0.x = GetUnquantizedTexelWieght(offset_base, 0, is_dual_plane);
+        p1.x = GetUnquantizedTexelWieght(offset_base, 1, is_dual_plane);
+    }
+    if ((v0 + 1) < (area)) {
+        const uint offset_base = v0 + 1;
+        p0.y = GetUnquantizedTexelWieght(offset_base, 0, is_dual_plane);
+        p1.y = GetUnquantizedTexelWieght(offset_base, 1, is_dual_plane);
+    }
+    if ((v0 + size.x) < (area)) {
+        const uint offset_base = v0 + size.x;
+        p0.z = GetUnquantizedTexelWieght(offset_base, 0, is_dual_plane);
+        p1.z = GetUnquantizedTexelWieght(offset_base, 1, is_dual_plane);
+    }
+    if ((v0 + size.x + 1) < (area)) {
+        const uint offset_base = v0 + size.x + 1;
+        p0.w = GetUnquantizedTexelWieght(offset_base, 0, is_dual_plane);
+        p1.w = GetUnquantizedTexelWieght(offset_base, 1, is_dual_plane);
+    }
+
+    const uint primary_weight = (uint(dot(p0, w)) + 8) >> 4;
+
+    uvec4 weight_vec = uvec4(primary_weight);
+
+    if (is_dual_plane) {
+        const uint secondary_weight = (uint(dot(p1, w)) + 8) >> 4;
+        for (uint c = 0; c < 4; c++) {
+            const bool is_secondary = ((plane_index + 1u) & 3u) == c;
+            weight_vec[c] = is_secondary ? secondary_weight : primary_weight;
         }
     }
+    return weight_vec;
 }
 
 int FindLayout(uint mode) {
@@ -1155,25 +1164,10 @@ void DecompressBlock(ivec3 coord) {
             }
             const uvec4 C0 = ReplicateByteTo16(endpoints0[local_partition]);
             const uvec4 C1 = ReplicateByteTo16(endpoints1[local_partition]);
-            const uint weight_offset = (j * block_dims.x + i);
-            const uint array_index = weight_offset / 4;
-            const uint vector_index = weight_offset % 4;
-            const uint primary_weight = unquantized_texel_weights[array_index][vector_index];
-            uvec4 weight_vec = uvec4(primary_weight);
-            if (dual_plane) {
-                const uint secondary_weight_offset = (j * block_dims.x + i) + ARRAY_NUM_ELEMENTS;
-                const uint secondary_array_index = secondary_weight_offset / 4;
-                const uint secondary_vector_index = secondary_weight_offset % 4;
-                const uint secondary_weight =
-                    unquantized_texel_weights[secondary_array_index][secondary_vector_index];
-                for (uint c = 0; c < 4; c++) {
-                    const bool is_secondary = ((plane_index + 1u) & 3u) == c;
-                    weight_vec[c] = is_secondary ? secondary_weight : primary_weight;
-                }
-            }
+            const uvec4 weight_vec = GetUnquantizedWeightVector(j, i, size_params, plane_index, dual_plane);
             const vec4 Cf =
                 vec4((C0 * (uvec4(64) - weight_vec) + C1 * weight_vec + uvec4(32)) / 64);
-            const vec4 p = (Cf / 65535.0);
+            const vec4 p = (Cf / 65535.0f);
             imageStore(dest_image, coord + ivec3(i, j, 0), p.gbar);
         }
     }

From 71857e889e3403621d78a2fceaf8265b1113f89f Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Wed, 9 Aug 2023 18:34:57 -0400
Subject: [PATCH 31/33] flatten result vector

---
 src/video_core/host_shaders/astc_decoder.comp | 19 +++++--------------
 1 file changed, 5 insertions(+), 14 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index 4014d4bfe..70f147d80 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -77,7 +77,7 @@ int color_bitsread = 0;
 #define DIVCEIL(number, divisor) (number + divisor - 1) / divisor
 #define ARRAY_NUM_ELEMENTS 144
 #define VECTOR_ARRAY_SIZE DIVCEIL(ARRAY_NUM_ELEMENTS * 2, 4)
-uvec4 result_vector[VECTOR_ARRAY_SIZE];
+uint result_vector[ARRAY_NUM_ELEMENTS * 2];
 
 int result_index = 0;
 uint result_vector_max_index;
@@ -122,9 +122,7 @@ void ResultEmplaceBack(EncodingData val) {
         result_limit_reached = true;
         return;
     }
-    const uint array_index = result_index / 4;
-    const uint vector_index = result_index % 4;
-    result_vector[array_index][vector_index] = val.data;
+    result_vector[result_index] = val.data;
     ++result_index;
 }
 
@@ -309,10 +307,7 @@ uint StreamColorBits(uint num_bits) {
 }
 
 EncodingData GetEncodingFromVector(uint index) {
-    const uint array_index = index / 4;
-    const uint vector_index = index % 4;
-
-    const uint data = result_vector[array_index][vector_index];
+    const uint data = result_vector[index];
     return EncodingData(data);
 }
 
@@ -809,18 +804,14 @@ void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) {
     const uint area = size.x * size.y;
     const uint loop_count = min(result_index, area * num_planes);
     for (uint itr = 0; itr < loop_count; ++itr) {
-        const uint array_index = itr / 4;
-        const uint vector_index = itr % 4;
-        result_vector[array_index][vector_index] =
+        result_vector[itr] =
             UnquantizeTexelWeight(GetEncodingFromVector(itr));
     }
 }
 
 uint GetUnquantizedTexelWieght(uint offset_base, uint plane, bool is_dual_plane) {
     const uint offset = is_dual_plane ? 2 * offset_base + plane : offset_base;
-    const uint array_index = offset / 4;
-    const uint vector_index = offset % 4;
-    return result_vector[array_index][vector_index];
+    return result_vector[offset];
 }
 
 uvec4 GetUnquantizedWeightVector(uint t, uint s, uvec2 size, uint plane_index, bool is_dual_plane) {

From 0f7220c9c8073ad1d2ab5dbc713b8c7d7a62bd0c Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Wed, 9 Aug 2023 18:38:37 -0400
Subject: [PATCH 32/33] flatten encoding_values

---
 src/video_core/host_shaders/astc_decoder.comp | 21 +++++++++----------
 1 file changed, 10 insertions(+), 11 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index 70f147d80..a5d9c97b4 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -54,13 +54,13 @@ const uint TRIT = 2u;
 
 // ASTC Encodings data, sorted in ascending order based on their BitLength value
 // (see GetBitLength() function)
-const uvec4 encoding_values[6] = uvec4[](
-    uvec4((JUST_BITS), (JUST_BITS | (1u << 8u)), (TRIT), (JUST_BITS | (2u << 8u))),
-    uvec4((QUINT), (TRIT | (1u << 8u)), (JUST_BITS | (3u << 8u)), (QUINT | (1u << 8u))),
-    uvec4((TRIT | (2u << 8u)), (JUST_BITS | (4u << 8u)), (QUINT | (2u << 8u)), (TRIT | (3u << 8u))),
-    uvec4((JUST_BITS | (5u << 8u)), (QUINT | (3u << 8u)), (TRIT | (4u << 8u)), (JUST_BITS | (6u << 8u))),
-    uvec4((QUINT | (4u << 8u)), (TRIT | (5u << 8u)), (JUST_BITS | (7u << 8u)), (QUINT | (5u << 8u))),
-    uvec4((TRIT | (6u << 8u)), (JUST_BITS | (8u << 8u)), 0u, 0u));
+const uint encoding_values[22] = uint[](
+    (JUST_BITS), (JUST_BITS | (1u << 8u)), (TRIT), (JUST_BITS | (2u << 8u)),
+    (QUINT), (TRIT | (1u << 8u)), (JUST_BITS | (3u << 8u)), (QUINT | (1u << 8u)),
+    (TRIT | (2u << 8u)), (JUST_BITS | (4u << 8u)), (QUINT | (2u << 8u)), (TRIT | (3u << 8u)),
+    (JUST_BITS | (5u << 8u)), (QUINT | (3u << 8u)), (TRIT | (4u << 8u)), (JUST_BITS | (6u << 8u)),
+    (QUINT | (4u << 8u)), (TRIT | (5u << 8u)), (JUST_BITS | (7u << 8u)), (QUINT | (5u << 8u)),
+    (TRIT | (6u << 8u)), (JUST_BITS | (8u << 8u)));
 
 // Input ASTC texture globals
 int total_bitsread = 0;
@@ -313,8 +313,7 @@ EncodingData GetEncodingFromVector(uint index) {
 
 // Returns the number of bits required to encode n_vals values.
 uint GetBitLength(uint n_vals, uint encoding_index) {
-    const EncodingData encoding_value =
-        EncodingData(encoding_values[encoding_index / 4][encoding_index % 4]);
+    const EncodingData encoding_value = EncodingData(encoding_values[encoding_index]);
     const uint encoding = Encoding(encoding_value);
     uint total_bits = NumBits(encoding_value) * n_vals;
     if (encoding == TRIT) {
@@ -435,7 +434,7 @@ void DecodeTritBlock(uint num_bits) {
 }
 
 void DecodeIntegerSequence(uint max_range, uint num_values) {
-    EncodingData val = EncodingData(encoding_values[max_range / 4][max_range % 4]);
+    EncodingData val = EncodingData(encoding_values[max_range]);
     const uint encoding = Encoding(val);
     const uint num_bits = NumBits(val);
     uint vals_decoded = 0;
@@ -467,7 +466,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
     // Find the largest encoding that's within color_data_bits
     // TODO(ameerj): profile with binary search
     int range = 0;
-    while (++range < ((encoding_values.length() * 4) - 2)) {
+    while (++range < encoding_values.length()) {
         const uint bit_length = GetBitLength(num_values, range);
         if (bit_length > color_data_bits) {
             break;

From 5c25712af9530dfd27960036141989a6c2f0c3bc Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Wed, 9 Aug 2023 18:45:52 -0400
Subject: [PATCH 33/33] flatten color_values

---
 src/video_core/host_shaders/astc_decoder.comp | 23 ++++++++-----------
 1 file changed, 9 insertions(+), 14 deletions(-)

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index a5d9c97b4..5ff17cd0c 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -457,8 +457,7 @@ void DecodeIntegerSequence(uint max_range, uint num_values) {
     }
 }
 
-uvec4 color_values[8];
-void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
+void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits, out uint color_values[32]) {
     uint num_values = 0;
     for (uint i = 0; i < num_partitions; i++) {
         num_values += ((modes[i] >> 2) + 1) << 1;
@@ -486,8 +485,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
         A = ReplicateBitTo9((bitval & 1));
         switch (encoding) {
         case JUST_BITS:
-            color_values[out_index / 4][out_index % 4] = FastReplicateTo8(bitval, bitlen);
-            ++out_index;
+            color_values[++out_index] = FastReplicateTo8(bitval, bitlen);
             break;
         case TRIT: {
             D = QuintTritValue(val);
@@ -566,8 +564,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
             uint T = (D * C) + B;
             T ^= A;
             T = (A & 0x80) | (T >> 2);
-            color_values[out_index / 4][out_index % 4] = T;
-            ++out_index;
+            color_values[++out_index] = T;
         }
     }
 }
@@ -592,19 +589,17 @@ ivec4 BlueContract(int a, int r, int g, int b) {
     return ivec4(a, (r + b) >> 1, (g + b) >> 1, b);
 }
 
-void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode,
+void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode, uint color_values[32],
                       inout uint colvals_index) {
 #define READ_UINT_VALUES(N)                                                                        \
     uvec4 V[2];                                                                                    \
     for (uint i = 0; i < N; i++) {                                                                 \
-        V[i / 4][i % 4] = color_values[colvals_index / 4][colvals_index % 4];                      \
-        ++colvals_index;                                                                           \
+        V[i / 4][i % 4] = color_values[++colvals_index];                      \
     }
 #define READ_INT_VALUES(N)                                                                         \
     ivec4 V[2];                                                                                    \
     for (uint i = 0; i < N; i++) {                                                                 \
-        V[i / 4][i % 4] = int(color_values[colvals_index / 4][colvals_index % 4]);                 \
-        ++colvals_index;                                                                           \
+        V[i / 4][i % 4] = int(color_values[++colvals_index]);                      \
     }
 
     switch (color_endpoint_mode) {
@@ -1111,11 +1106,11 @@ void DecompressBlock(ivec3 coord) {
     {
         // This decode phase should at most push 32 elements into the vector
         result_vector_max_index = 32;
-
+        uint color_values[32];
         uint colvals_index = 0;
-        DecodeColorValues(color_endpoint_mode, num_partitions, color_data_bits);
+        DecodeColorValues(color_endpoint_mode, num_partitions, color_data_bits, color_values);
         for (uint i = 0; i < num_partitions; i++) {
-            ComputeEndpoints(endpoints0[i], endpoints1[i], color_endpoint_mode[i],
+            ComputeEndpoints(endpoints0[i], endpoints1[i], color_endpoint_mode[i], color_values,
                              colvals_index);
         }
     }