diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.glsl b/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.glsl
index d902ff3a67..adcc0cbb52 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.glsl
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.glsl
@@ -244,7 +244,7 @@ vec2 dequantize(uint ib, uint iqs, uint a_offset) {
     // Add parity bit
     const uint sign8 = sign7 | (bitCount(sign7) << 7);
     const uint sign = sign8 >> (iqs % 8);
-    const u8vec4 grid = unpack8(iq2xxs_grid[qs][(iqs % 8) / 4] >> (8 * (iqs % 4)));
+    const u8vec4 grid = unpack8(iq2_grid[qs][(iqs % 8) / 4] >> (8 * (iqs % 4)));
     bool sign0 = (sign & 1) != 0;
     bool sign1 = (sign & 2) != 0;
     return db * vec2(
@@ -264,7 +264,7 @@ vec4 dequantize4(uint ib, uint iqs, uint a_offset) {
     // Add parity bit
     const uint sign8 = sign7 | (bitCount(sign7) << 7);
     const uint sign = sign8 >> (iqs % 8);
-    const u8vec4 grid = unpack8(iq2xxs_grid[qs][(iqs % 8) / 4] >> (8 * (iqs % 4)));
+    const u8vec4 grid = unpack8(iq2_grid[qs][(iqs % 8) / 4] >> (8 * (iqs % 4)));
     bool sign0 = (sign & 1) != 0;
     bool sign1 = (sign & 2) != 0;
     bool sign2 = (sign & 4) != 0;
@@ -287,7 +287,7 @@ vec2 dequantize(uint ib, uint iqs, uint a_offset) {
     // Add parity bit
     const uint sign8 = sign7 | (bitCount(sign7) << 7);
     const uint sign = sign8 >> (iqs % 8);
-    const u8vec4 grid = unpack8(iq2xs_grid[qs & 511][(iqs % 8) / 4] >> (8 * (iqs % 4)));
+    const u8vec4 grid = unpack8(iq2_grid[qs & 511][(iqs % 8) / 4] >> (8 * (iqs % 4)));
     bool sign0 = (sign & 1) != 0;
     bool sign1 = (sign & 2) != 0;
     return db * vec2(
@@ -303,7 +303,7 @@ vec4 dequantize4(uint ib, uint iqs, uint a_offset) {
     // Add parity bit
     const uint sign8 = sign7 | (bitCount(sign7) << 7);
     const uint sign = sign8 >> (iqs % 8);
-    const u8vec4 grid = unpack8(iq2xs_grid[qs & 511][(iqs % 8) / 4] >> (8 * (iqs % 4)));
+    const u8vec4 grid = unpack8(iq2_grid[qs & 511][(iqs % 8) / 4] >> (8 * (iqs % 4)));
     bool sign0 = (sign & 1) != 0;
     bool sign1 = (sign & 2) != 0;
     bool sign2 = (sign & 4) != 0;
@@ -329,7 +329,7 @@ vec2 dequantize(uint ib, uint iqs, uint a_offset) {
     const uint sign = data_a[a_offset + ib].qs[QUANT_K / 8 + ib8] >> (iqs % 8);
 
     const float db = 0.25 * (0.5 + scale);
-    const u8vec4 grid = unpack8(iq2s_grid[qs | ((qh << (8 - qhshift)) & 0x300)][(iqs % 8) / 4]);
+    const u8vec4 grid = unpack8(iq2_grid[qs | ((qh << (8 - qhshift)) & 0x300)][(iqs % 8) / 4]);
     bool sign0 = (sign & 1) != 0;
     bool sign1 = (sign & 2) != 0;
     return db * vec2(
@@ -348,7 +348,7 @@ vec4 dequantize4(uint ib, uint iqs, uint a_offset) {
     const uint sign = data_a[a_offset + ib].qs[QUANT_K / 8 + ib8] >> (iqs % 8);
 
     const float db = 0.25 * (0.5 + scale);
-    const u8vec4 grid = unpack8(iq2s_grid[qs | ((qh << (8 - qhshift)) & 0x300)][(iqs % 8) / 4]);
+    const u8vec4 grid = unpack8(iq2_grid[qs | ((qh << (8 - qhshift)) & 0x300)][(iqs % 8) / 4]);
     bool sign0 = (sign & 1) != 0;
     bool sign1 = (sign & 2) != 0;
     bool sign2 = (sign & 4) != 0;
@@ -376,7 +376,7 @@ vec2 dequantize(uint ib, uint iqs, uint a_offset) {
     // Add parity bit
     const uint sign8 = sign7 | (bitCount(sign7) << 7);
     const uint sign = sign8 >> (iqs % 8);
-    const u8vec4 grid = unpack8(iq3xxs_grid[qs] >> (8 * (iqs % 4)));
+    const u8vec4 grid = unpack8(iq3_grid[qs] >> (8 * (iqs % 4)));
     bool sign0 = (sign & 1) != 0;
     bool sign1 = (sign & 2) != 0;
     return db * vec2(
@@ -396,7 +396,7 @@ vec4 dequantize4(uint ib, uint iqs, uint a_offset) {
     // Add parity bit
     const uint sign8 = sign7 | (bitCount(sign7) << 7);
     const uint sign = sign8 >> (iqs % 8);
-    const u8vec4 grid = unpack8(iq3xxs_grid[qs]);
+    const u8vec4 grid = unpack8(iq3_grid[qs]);
     bool sign0 = (sign & 1) != 0;
     bool sign1 = (sign & 2) != 0;
     bool sign2 = (sign & 4) != 0;
@@ -419,7 +419,7 @@ vec2 dequantize(uint ib, uint iqs, uint a_offset) {
     bool sign0 = (sign & 1) != 0;
     bool sign1 = (sign & 2) != 0;
     const float db = 1 + 2 * ((scale >> (4 * ((iqs / 32) & 1))) & 0xf);
-    const uint32_t grid = iq3s_grid[qs | ((qh << (8 - ((iqs / 4) % 8))) & 256)] >> (8 * (iqs % 4));
+    const uint32_t grid = iq3_grid[qs | ((qh << (8 - ((iqs / 4) % 8))) & 256)] >> (8 * (iqs % 4));
     return db * vec2(
         int(grid & 0xFF) * (sign0 ? -1.0 : 1.0),
         int((grid >> 8) & 0xFF) * (sign1 ? -1.0 : 1.0)
@@ -437,7 +437,7 @@ vec4 dequantize4(uint ib, uint iqs, uint a_offset) {
     bool sign2 = (sign & 4) != 0;
     bool sign3 = (sign & 8) != 0;
     const float db = 1 + 2 * ((scale >> (4 * (ib32 & 1))) & 0xf);
-    const uint32_t grid = iq3s_grid[qs | ((qh << (8 - ib4 % 8)) & 256)] >> (8 * (iqs % 4));
+    const uint32_t grid = iq3_grid[qs | ((qh << (8 - ib4 % 8)) & 256)] >> (8 * (iqs % 4));
     return db * vec4(
         int(grid & 0xFF) * (sign0 ? -1.0 : 1.0),
         int((grid >> 8) & 0xFF) * (sign1 ? -1.0 : 1.0),
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.glsl b/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.glsl
index 1c45c6a78c..140c1a1c75 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.glsl
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.glsl
@@ -898,7 +898,7 @@ float16_t dequantFuncIQ2_XXS(const in decodeBufIQ2_XXS bl, const in uint blockCo
     uint sign = bitfieldExtract(signscale, 7 * int(ib8), 7);
     sign |= bitCount(sign) << 7;
 
-    uint g2 = iq2xxs_grid[qs][(idx & 4) >> 2];
+    uint g2 = iq2_grid[qs][(idx & 4) >> 2];
     g2 >>= (idx & 2) * 8;
     const vec2 g = vec2(unpack8(g2));
 
@@ -923,7 +923,7 @@ f16vec4 dequantFuncIQ2_XXS_v(const in decodeBufIQ2_XXS bl, const in uint blockCo
     sign |= bitCount(sign) << 7;
     const uint sb = sign >> (idx & 7u);
 
-    const uint   g2 = iq2xxs_grid[qs][(idx & 4) >> 2];
+    const uint   g2 = iq2_grid[qs][(idx & 4) >> 2];
     const u8vec4 g  = unpack8(g2);
 
     return f16vec4(
@@ -953,7 +953,7 @@ float16_t dequantFuncIQ2_XS(const in decodeBufIQ2_XS bl, const in uint blockCoor
 
     uint sign = uint(qs >> 9);
     sign |= bitCount(sign) << 7;
-    uint g2 = iq2xs_grid[qs & 0x1FF][(idx & 4) >> 2];
+    uint g2 = iq2_grid[qs & 0x1FF][(idx & 4) >> 2];
     g2 >>= (idx & 2) * 8;
     const vec2 g = vec2(unpack8(g2));
 
@@ -976,7 +976,7 @@ f16vec4 dequantFuncIQ2_XS_v(const in decodeBufIQ2_XS bl, const in uint blockCoor
     sign |= bitCount(sign) << 7;
     const uint sb = sign >> (idx & 7u);
 
-    const uint   g2 = iq2xs_grid[qs & 0x1FF][(idx & 4) >> 2];
+    const uint   g2 = iq2_grid[qs & 0x1FF][(idx & 4) >> 2];
     const u8vec4 g  = unpack8(g2);
 
     return f16vec4(
@@ -1008,7 +1008,7 @@ float16_t dequantFuncIQ2_S(const in decodeBufIQ2_S bl, const in uint blockCoords
     const float d = float(bl.block.d);
     const float db = d * 0.25 * (0.5 + scale);
     const ivec2 sign01 = 1 - (2 & ivec2(sign << 1, sign));
-    uint g2 = iq2s_grid[qs | ((qh << (8 - qhshift)) & 0x300)][(idx & 4) >> 2];
+    uint g2 = iq2_grid[qs | ((qh << (8 - qhshift)) & 0x300)][(idx & 4) >> 2];
     g2 >>= (idx & 2) * 8;
     const vec2 v = db * vec2(sign01) * vec2(unpack8(g2));
     return float16_t(v[idx & 1]);
@@ -1030,7 +1030,7 @@ f16vec4 dequantFuncIQ2_S_v(const in decodeBufIQ2_S bl, const in uint blockCoords
     const float d  = float(bl.block.d);
     const float db = d * 0.25 * (0.5 + scale);
 
-    const uint   g2 = iq2s_grid[qs | ((qh << (8 - qhshift)) & 0x300)][(idx & 4) >> 2];
+    const uint   g2 = iq2_grid[qs | ((qh << (8 - qhshift)) & 0x300)][(idx & 4) >> 2];
     const u8vec4 g  = unpack8(g2);
 
     return f16vec4(
@@ -1068,7 +1068,7 @@ float16_t dequantFuncIQ3_XXS(const in decodeBufIQ3_XXS bl, const in uint blockCo
     const uint32_t sign7 = bitfieldExtract(signs, 7 * (int(iqs / 2) % 4), 7);
     const uint sign = (sign7 | (bitCount(sign7) << 7)) >> (idx & 0x6);
     const ivec2 sign01 = ivec2(1 - (2 & ivec2(sign << 1, sign)));
-    const uint grid = iq3xxs_grid[qs] >> (16 * ((idx & 2) >> 1));
+    const uint grid = iq3_grid[qs] >> (16 * ((idx & 2) >> 1));
     const vec2 v = db * vec2(sign01) * vec2(unpack8(grid).xy);
     return float16_t(v[idx & 1]);
 }
@@ -1089,7 +1089,7 @@ f16vec4 dequantFuncIQ3_XXS_v(const in decodeBufIQ3_XXS bl, const in uint blockCo
     const uint sign7 = bitfieldExtract(signs, 7 * (int(iqs / 2) % 4), 7);
     const uint sb    = (sign7 | (bitCount(sign7) << 7)) >> (idx & 0x6u);
 
-    const uint   grid = iq3xxs_grid[qs];
+    const uint   grid = iq3_grid[qs];
     const u8vec4 g    = unpack8(grid);
 
     return f16vec4(
@@ -1119,7 +1119,7 @@ float16_t dequantFuncIQ3_S(const in decodeBufIQ3_S bl, const in uint blockCoords
     const uint scale = bl.block.scales[iqs / 16];
     const ivec2 sign01 = ivec2(1 - (2 & ivec2(sign << 1, sign)));
     const float db = d * (1 + 2 * ((scale >> (4 * (iqh & 1))) & 0xf));
-    const uint32_t grid = iq3s_grid[qs | ((qh << (8 - (iqs % 8))) & 256)] >> ((idx & 2) << 3);
+    const uint32_t grid = iq3_grid[qs | ((qh << (8 - (iqs % 8))) & 256)] >> ((idx & 2) << 3);
     const vec2 v = db * vec2(sign01) * vec2(unpack8(grid).xy);
 
     return float16_t(v[idx & 1]);
@@ -1139,7 +1139,7 @@ f16vec4 dequantFuncIQ3_S_v(const in decodeBufIQ3_S bl, const in uint blockCoords
     const uint  scale = bl.block.scales[iqs / 16];
     const float db    = d * (1 + 2 * ((scale >> (4 * (iqh & 1))) & 0xf));
 
-    const uint   grid = iq3s_grid[qs | ((qh << (8 - (iqs % 8))) & 256)];
+    const uint   grid = iq3_grid[qs | ((qh << (8 - (iqs % 8))) & 256)];
     const u8vec4 g    = unpack8(grid);
 
     return f16vec4(
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_s.comp b/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_s.comp
index 78490162cd..f834b78562 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_s.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_s.comp
@@ -29,7 +29,7 @@ void main() {
         uint qs = data_a[ib].qs[4 * ib32 + l];
         const uint8_t sign = data_a[ib].qs[QUANT_K / 8 + 4 * ib32 + l];
         qs |= (qh << (8 - 2 * l)) & 0x300;
-        const uvec2 grid = iq2s_grid[qs];
+        const uvec2 grid = iq2_grid[qs];
         const u8vec4 grid0 = unpack8(grid.x);
         const u8vec4 grid1 = unpack8(grid.y);
         data_b[b_idx + 8 * l + 0] = D_TYPE(db[l/2] * grid0.x * ((sign & 1) != 0 ? -1.0 : 1.0));
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xs.comp b/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xs.comp
index 9b8ce0a7f8..f5bba538f3 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xs.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xs.comp
@@ -28,7 +28,7 @@ void main() {
         uint16_t qs = data_a[ib].qs[4 * ib32 + l];
         const uint sign7 = qs >> 9;
         const uint sign8 = sign7 | (bitCount(sign7) << 7); // parity bit
-        const uvec2 grid = iq2xs_grid[qs & 511];
+        const uvec2 grid = iq2_grid[qs & 511];
         const u8vec4 grid0 = unpack8(grid.x);
         const u8vec4 grid1 = unpack8(grid.y);
         data_b[b_idx + 8 * l + 0] = D_TYPE(db[l/2] * grid0.x * ((sign8 & 1) != 0 ? -1.0 : 1.0));
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xxs.comp b/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xxs.comp
index aacf07d0f8..dca5b84c6a 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xxs.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xxs.comp
@@ -34,7 +34,7 @@ void main() {
         const uint sign7 = bitfieldExtract(signscale, 7 * int(l), 7);
         const uint sign8 = sign7 | (bitCount(sign7) << 7); // parity bit
         const uint qs = data_a[ib].qs[8 * is + l];
-        const uvec2 grid = iq2xxs_grid[qs];
+        const uvec2 grid = iq2_grid[qs];
         const u8vec4 grid0 = unpack8(grid.x);
         const u8vec4 grid1 = unpack8(grid.y);
         data_b[b_idx + 8 * l + 0] = D_TYPE(db * grid0.x * ((sign8 & 1) != 0 ? -1.0 : 1.0));
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_s.comp b/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_s.comp
index f2c20b1d2c..b18a9d0cd1 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_s.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_s.comp
@@ -31,7 +31,7 @@ void main() {
         const uint qs = data_a[ib].qs[iqs];
         const uint gidx = qs | ((qh << (8 - l)) & 256);
         const uint8_t signs = data_a[ib].signs[iqs / 2] >> (4 * (l & 1));
-        const u8vec4 grid = unpack8(iq3s_grid[gidx]);
+        const u8vec4 grid = unpack8(iq3_grid[gidx]);
         data_b[b_idx + 4 * l + 0] = D_TYPE(db * grid.x * ((signs & 1) != 0 ? -1.0 : 1.0));
         data_b[b_idx + 4 * l + 1] = D_TYPE(db * grid.y * ((signs & 2) != 0 ? -1.0 : 1.0));
         data_b[b_idx + 4 * l + 2] = D_TYPE(db * grid.z * ((signs & 4) != 0 ? -1.0 : 1.0));
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_xxs.comp b/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_xxs.comp
index 671c1f4a0d..60c5318e3b 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_xxs.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_xxs.comp
@@ -37,8 +37,8 @@ void main() {
         const uint sign8 = sign7 | (bitCount(sign7) << 7);
         const uint qs0 = data_a[ib].qs[8 * is + 2 * l];
         const uint qs1 = data_a[ib].qs[8 * is + 2 * l + 1];
-        const u8vec4 grid0 = unpack8(iq3xxs_grid[qs0]);
-        const u8vec4 grid1 = unpack8(iq3xxs_grid[qs1]);
+        const u8vec4 grid0 = unpack8(iq3_grid[qs0]);
+        const u8vec4 grid1 = unpack8(iq3_grid[qs1]);
         data_b[b_idx + 8 * l + 0] = D_TYPE(db * grid0.x * ((sign8 & 1) != 0 ? -1.0 : 1.0));
         data_b[b_idx + 8 * l + 1] = D_TYPE(db * grid0.y * ((sign8 & 2) != 0 ? -1.0 : 1.0));
         data_b[b_idx + 8 * l + 2] = D_TYPE(db * grid0.z * ((sign8 & 4) != 0 ? -1.0 : 1.0));
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/iq_shmem_init.glsl b/ggml/src/ggml-vulkan/vulkan-shaders/iq_shmem_init.glsl
index 4c6aaedbda..5fb2ba9ab8 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/iq_shmem_init.glsl
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/iq_shmem_init.glsl
@@ -10,37 +10,37 @@ void init_iq_shmem(uvec3 wgsize) {
         }
         barrier();
     } else if (MmTypeA == GGML_TYPE_IQ2_XXS) {
-        [[unroll]] for (uint i = 0; i < iq2xxs_grid.length(); i += wgsize.x) {
+        [[unroll]] for (uint i = 0; i < iq2_grid.length(); i += wgsize.x) {
             if (iq2xxs_grid_const.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq2xxs_grid_const.length()) {
-                iq2xxs_grid[i + gl_LocalInvocationIndex.x] = iq2xxs_grid_const[i + gl_LocalInvocationIndex.x];
+                iq2_grid[i + gl_LocalInvocationIndex.x] = iq2xxs_grid_const[i + gl_LocalInvocationIndex.x];
             }
         }
         barrier();
     } else if (MmTypeA == GGML_TYPE_IQ2_XS) {
-        [[unroll]] for (uint i = 0; i < iq2xs_grid.length(); i += wgsize.x) {
+        [[unroll]] for (uint i = 0; i < iq2_grid.length(); i += wgsize.x) {
             if (iq2xs_grid_const.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq2xs_grid_const.length()) {
-                iq2xs_grid[i + gl_LocalInvocationIndex.x] = iq2xs_grid_const[i + gl_LocalInvocationIndex.x];
+                iq2_grid[i + gl_LocalInvocationIndex.x] = iq2xs_grid_const[i + gl_LocalInvocationIndex.x];
             }
         }
         barrier();
     } else if (MmTypeA == GGML_TYPE_IQ2_S) {
-        [[unroll]] for (uint i = 0; i < iq2s_grid.length(); i += wgsize.x) {
+        [[unroll]] for (uint i = 0; i < iq2_grid.length(); i += wgsize.x) {
             if (iq2s_grid_const.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq2s_grid_const.length()) {
-                iq2s_grid[i + gl_LocalInvocationIndex.x] = iq2s_grid_const[i + gl_LocalInvocationIndex.x];
+                iq2_grid[i + gl_LocalInvocationIndex.x] = iq2s_grid_const[i + gl_LocalInvocationIndex.x];
             }
         }
         barrier();
     } else if (MmTypeA == GGML_TYPE_IQ3_XXS) {
-        [[unroll]] for (uint i = 0; i < iq3xxs_grid.length(); i += wgsize.x) {
+        [[unroll]] for (uint i = 0; i < iq3_grid.length(); i += wgsize.x) {
             if (iq3xxs_grid_const.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq3xxs_grid_const.length()) {
-                iq3xxs_grid[i + gl_LocalInvocationIndex.x] = iq3xxs_grid_const[i + gl_LocalInvocationIndex.x];
+                iq3_grid[i + gl_LocalInvocationIndex.x] = iq3xxs_grid_const[i + gl_LocalInvocationIndex.x];
             }
         }
         barrier();
     } else if (MmTypeA == GGML_TYPE_IQ3_S) {
-        [[unroll]] for (uint i = 0; i < iq3s_grid.length(); i += wgsize.x) {
+        [[unroll]] for (uint i = 0; i < iq3_grid.length(); i += wgsize.x) {
             if (iq3s_grid_const.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq3s_grid_const.length()) {
-                iq3s_grid[i + gl_LocalInvocationIndex.x] = iq3s_grid_const[i + gl_LocalInvocationIndex.x];
+                iq3_grid[i + gl_LocalInvocationIndex.x] = iq3s_grid_const[i + gl_LocalInvocationIndex.x];
             }
         }
         barrier();
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_s.comp b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_s.comp
index b4f6d1d6b6..27c461d60d 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_s.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_s.comp
@@ -24,7 +24,7 @@ void calc_superblock(const uint a_offset, const uint b_offset, const uint itid,
         [[unroll]] for (uint l = 0; l < 2; ++l) {
             const uint8_t sign = sign16[l];
             const uint qs = qs16[l] | ((qh << (8 - nibble_shift - 2 * l)) & 0x300);
-            const uvec2 grid = iq2s_grid[qs];
+            const uvec2 grid = iq2_grid[qs];
             const vec4 grid0 = vec4(unpack8(grid.x));
             const vec4 grid1 = vec4(unpack8(grid.y));
 
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xs.comp b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xs.comp
index d8dafe5f70..1a202e1297 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xs.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xs.comp
@@ -31,7 +31,7 @@ void calc_superblock(const uint a_offset, const uint b_offset, const uint itid,
             const uint qs = data_a[ibi].qs[2 * itid + l];
             sign_vals[l] = qs >> 9;
             sign7_vals[l] = bitCount(sign_vals[l]);
-            const uvec2 grid_data = iq2xs_grid[qs & 511];
+            const uvec2 grid_data = iq2_grid[qs & 511];
             grid0_vals[l] = vec4(unpack8(grid_data.x));
             grid1_vals[l] = vec4(unpack8(grid_data.y));
         }
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xxs.comp b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xxs.comp
index f75dcf8331..f55ee10e8a 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xxs.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xxs.comp
@@ -22,8 +22,8 @@ void calc_superblock(const uint a_offset, const uint b_offset, const uint itid,
             const uint qs = data_a[ibi].qs[8 * ib32 + 2 * (itid & 1) + l];
             const uint sign = bitfieldExtract(signscale, 7 * int(2 * (itid & 1) + l), 7);
             const uint sign7 = bitCount(sign);
-            const vec4 grid0 = vec4(unpack8(iq2xxs_grid[qs].x));
-            const vec4 grid1 = vec4(unpack8(iq2xxs_grid[qs].y));
+            const vec4 grid0 = vec4(unpack8(iq2_grid[qs].x));
+            const vec4 grid1 = vec4(unpack8(iq2_grid[qs].y));
 
             [[unroll]] for (uint j = 0; j < NUM_COLS; ++j) {
                 const vec4 b0 = vec4(data_b_v4[(j*p.batch_stride_b + b_offset + y_idx) / 4 + 2*l + 0]);
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_s.comp b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_s.comp
index 5cdf2a89d0..840c1174bd 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_s.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_s.comp
@@ -23,8 +23,8 @@ void calc_superblock(const uint a_offset, const uint b_offset, const uint ib32,
         [[unroll]] for (uint l = 0; l < 4; ++l) {
             const u8vec2 qs = unpack8(uint32_t(data_a_packed16[ibi].qs[4 * ib32 + l])).xy; // vec4 used due to #12147
             const uint sign = data_a[ibi].signs[4 * ib32 + l];
-            const vec4 grid0 = vec4(unpack8(iq3s_grid[qs.x | ((qh << (8 - 2*l)) & 0x100)]));
-            const vec4 grid1 = vec4(unpack8(iq3s_grid[qs.y | ((qh << (7 - 2*l)) & 0x100)]));
+            const vec4 grid0 = vec4(unpack8(iq3_grid[qs.x | ((qh << (8 - 2*l)) & 0x100)]));
+            const vec4 grid1 = vec4(unpack8(iq3_grid[qs.y | ((qh << (7 - 2*l)) & 0x100)]));
 
             [[unroll]] for (uint j = 0; j < NUM_COLS; ++j) {
                 const vec4 b0 = vec4(data_b_v4[(j*p.batch_stride_b + b_offset + y_idx) / 4 + 2*l + 0]);
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_xxs.comp b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_xxs.comp
index a88898109a..5f7a89bc1c 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_xxs.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_xxs.comp
@@ -23,8 +23,8 @@ void calc_superblock(const uint a_offset, const uint b_offset, const uint itid,
             const uint qs1 = data_a[ibi].qs[8 * ib32 + 4 * (itid & 1) + 2 * l + 1];
             const uint sign = bitfieldExtract(signscale, 7 * int(2 * (itid & 1) + l), 7);
             const uint sign7 = bitCount(sign);
-            const vec4 grid0 = vec4(unpack8(iq3xxs_grid[qs0]));
-            const vec4 grid1 = vec4(unpack8(iq3xxs_grid[qs1]));
+            const vec4 grid0 = vec4(unpack8(iq3_grid[qs0]));
+            const vec4 grid1 = vec4(unpack8(iq3_grid[qs1]));
 
             [[unroll]] for (uint j = 0; j < NUM_COLS; ++j) {
                 const vec4 b0 = vec4(data_b_v4[(j*p.batch_stride_b + b_offset + y_idx) / 4 + 2*l + 0]);
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp
index 1fbdb757c0..fa4929a15f 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp
@@ -31,6 +31,11 @@
 #extension GL_EXT_shader_explicit_arithmetic_types_int16 : require
 #endif
 
+#ifdef MULMAT_QUANT
+#include "ggml_type_ids.glsl"
+layout (constant_id = 12) const uint MmTypeA = 0;
+#endif
+
 #include "types.glsl"
 #include "dot_product_funcs.glsl"
 
@@ -46,8 +51,6 @@
 layout (constant_id = 11) const uint ALIGNED = 0;
 
 #ifdef MULMAT_QUANT
-#include "ggml_type_ids.glsl"
-layout (constant_id = 12) const uint MmTypeA = 0;
 
 uint mm_load_vec_a() {
     switch (MmTypeA) {
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp
index d10006c95f..07905b3d21 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp
@@ -21,9 +21,14 @@
 #extension GL_EXT_bfloat16 : enable
 #endif
 
+#include "ggml_type_ids.glsl"
+
+#ifdef MULMAT_QUANT
+layout (constant_id = 7) const uint MmTypeA = 0;
+#endif
+
 #include "types.glsl"
 #include "utils.glsl"
-#include "ggml_type_ids.glsl"
 
 layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
 
@@ -41,7 +46,6 @@ layout (constant_id = 5) const uint ALIGNED = 0;
 layout (constant_id = 6) const uint subgroup_size = 32;
 
 #ifdef MULMAT_QUANT
-layout (constant_id = 7) const uint MmTypeA = 0;
 
 uint mm_quant_k() {
     switch (MmTypeA) {
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl
index 320cc58a4e..c50348983f 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl
@@ -364,7 +364,7 @@ void load_a_to_shmem(const uint pos_a, const uint row, const uint col, const uin
         const FLOAT_TYPE db = FLOAT_TYPE(d * 0.25 * (0.5 + (signs >> 28)));
         const uint32_t sign7 = bitfieldExtract(signs, 7 * int(ib8), 7);
         const uint sign = sign7 | (bitCount(sign7) << 7);
-        const uvec2 grid = iq2xxs_grid[qs];
+        const uvec2 grid = iq2_grid[qs];
         const vec4 grid0 = vec4(unpack8(grid.x));
         const vec4 grid1 = vec4(unpack8(grid.y));
 
@@ -390,7 +390,7 @@ void load_a_to_shmem(const uint pos_a, const uint row, const uint col, const uin
         const uint qs = a_iq2_xs.data[ib].qs[4 * ib32 + ib8];
         const uint sign7 = qs >> 9;
         const uint sign = sign7 | (bitCount(sign7) << 7);
-        const uvec2 grid = iq2xs_grid[qs & 511];
+        const uvec2 grid = iq2_grid[qs & 511];
         const vec4 grid0 = vec4(unpack8(grid.x));
         const vec4 grid1 = vec4(unpack8(grid.y));
 
@@ -418,7 +418,7 @@ void load_a_to_shmem(const uint pos_a, const uint row, const uint col, const uin
 
         const float d = float(a_iq2_s.data[ib].d);
         const FLOAT_TYPE db = FLOAT_TYPE(d * 0.25 * (0.5 + scale));
-        const uvec2 grid = iq2s_grid[qs | ((qh << (8 - qhshift)) & 0x300)];
+        const uvec2 grid = iq2_grid[qs | ((qh << (8 - qhshift)) & 0x300)];
         const vec4 grid0 = vec4(unpack8(grid.x));
         const vec4 grid1 = vec4(unpack8(grid.y));
 
@@ -447,7 +447,7 @@ void load_a_to_shmem(const uint pos_a, const uint row, const uint col, const uin
         const float db = d * 0.5 * (0.5 + (signs >> 28));
         const uint32_t sign7 = bitfieldExtract(signs, 7 * (int(iqs / 2) % 4), 7);
         const uint sign = (sign7 | (bitCount(sign7) << 7)) >> (4 * (idx % 2));
-        const uint grid = iq3xxs_grid[qs];
+        const uint grid = iq3_grid[qs];
         const vec4 v = db * vec4(unpack8(grid));
 
         buf_a[buf_idx    ] = FLOAT_TYPEV2((sign &   1) != 0 ? -v.x : v.x,
@@ -469,7 +469,7 @@ void load_a_to_shmem(const uint pos_a, const uint row, const uint col, const uin
         const uint scale = a_iq3_s.data[ib].scales[iqs / 16];
         const i8vec2 sign01 = i8vec2(1 - (2 & i8vec2(sign << 1, sign)));
         const float db = d * (1 + 2 * ((scale >> (4 * (iqh & 1))) & 0xf));
-        const uint32_t grid = iq3s_grid[qs | ((qh << (8 - (iqs % 8))) & 256)];
+        const uint32_t grid = iq3_grid[qs | ((qh << (8 - (iqs % 8))) & 256)];
         const vec4 v = db * vec4(unpack8(grid));
 
         buf_a[buf_idx    ] = FLOAT_TYPEV2((sign &   1) != 0 ? -v.x : v.x,
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/types.glsl b/ggml/src/ggml-vulkan/vulkan-shaders/types.glsl
index 6d66ac59d1..2b025a736a 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/types.glsl
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/types.glsl
@@ -891,10 +891,17 @@ const uint32_t[2048] iq1s_grid_gpu_const = {
 };
 #endif
 
+#ifdef MULMAT_QUANT
+shared uint16_t iq1s_grid[(MmTypeA == GGML_TYPE_IQ1_S || MmTypeA == GGML_TYPE_IQ1_M) ? 2048 : 1];
+#if defined(NEEDS_IQ1S_GRID_GPU)
+shared uint32_t iq1s_grid_gpu[(MmTypeA == GGML_TYPE_IQ1_S || MmTypeA == GGML_TYPE_IQ1_M) ? 2048 : 1];
+#endif
+#else
 shared uint16_t iq1s_grid[2048];
 #if defined(NEEDS_IQ1S_GRID_GPU)
 shared uint32_t iq1s_grid_gpu[2048];
 #endif
+#endif
 
 #if defined(DATA_A_IQ1_S) || defined(DATA_A_IQ1_M)
 #define NEEDS_INIT_IQ_SHMEM
@@ -922,6 +929,18 @@ void init_iq_shmem(uvec3 wgsize)
 #endif
 #endif
 
+#if defined(DATA_A_IQ2_XXS) || defined(DATA_A_IQ2_XS) || defined(DATA_A_IQ2_S) || defined(MULMAT_QUANT)
+#ifdef MULMAT_QUANT
+shared uvec2 iq2_grid[MmTypeA == GGML_TYPE_IQ2_S ? 1024 : MmTypeA == GGML_TYPE_IQ2_XS ? 512 : MmTypeA == GGML_TYPE_IQ2_XXS ? 256 : 1];
+#elif defined(DATA_A_IQ2_S)
+shared uvec2 iq2_grid[1024];
+#elif defined(DATA_A_IQ2_XS)
+shared uvec2 iq2_grid[512];
+#else
+shared uvec2 iq2_grid[256];
+#endif
+#endif
+
 #define QUANT_K_IQ2_XXS 256
 #define QUANT_R_IQ2_XXS 1
 
@@ -1006,16 +1025,14 @@ const uvec2[256] iq2xxs_grid_const = {
     uvec2(0x08080808, 0x2b2b082b), uvec2(0x08192b08, 0x2b2b1908), uvec2(0x19190808, 0x2b2b2b08), uvec2(0x08081908, 0x2b2b2b19)
 };
 
-shared uvec2 iq2xxs_grid[256];
-
 #if defined(DATA_A_IQ2_XXS)
 #define NEEDS_INIT_IQ_SHMEM
 void init_iq_shmem(uvec3 wgsize)
 {
     // copy the table into shared memory and sync
-    [[unroll]] for (uint i = 0; i < iq2xxs_grid.length(); i += wgsize.x) {
+    [[unroll]] for (uint i = 0; i < iq2_grid.length(); i += wgsize.x) {
         if (iq2xxs_grid_const.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq2xxs_grid_const.length()) {
-            iq2xxs_grid[i + gl_LocalInvocationIndex.x] = iq2xxs_grid_const[i + gl_LocalInvocationIndex.x];
+            iq2_grid[i + gl_LocalInvocationIndex.x] = iq2xxs_grid_const[i + gl_LocalInvocationIndex.x];
         }
     }
     barrier();
@@ -1180,16 +1197,14 @@ const uvec2 iq2xs_grid_const[512] = {
     uvec2(0x082b2b08, 0x2b2b2b2b), uvec2(0x082b2b2b, 0x2b2b2b2b), uvec2(0x2b190819, 0x2b2b2b2b), uvec2(0x2b2b2b2b, 0x2b2b2b2b),
 };
 
-shared uvec2 iq2xs_grid[512];
-
 #if defined(DATA_A_IQ2_XS)
 #define NEEDS_INIT_IQ_SHMEM
 void init_iq_shmem(uvec3 wgsize)
 {
     // copy the table into shared memory and sync
-    [[unroll]] for (uint i = 0; i < iq2xs_grid.length(); i += wgsize.x) {
-        if (iq2xs_grid.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq2xs_grid_const.length()) {
-            iq2xs_grid[i + gl_LocalInvocationIndex.x] = iq2xs_grid_const[i + gl_LocalInvocationIndex.x];
+    [[unroll]] for (uint i = 0; i < iq2_grid.length(); i += wgsize.x) {
+        if (iq2_grid.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq2xs_grid_const.length()) {
+            iq2_grid[i + gl_LocalInvocationIndex.x] = iq2xs_grid_const[i + gl_LocalInvocationIndex.x];
         }
     }
     barrier();
@@ -1484,16 +1499,14 @@ const uvec2 iq2s_grid_const[1024] = {
     uvec2(0x082b082b, 0x2b2b2b2b), uvec2(0x082b2b08, 0x2b2b2b2b), uvec2(0x2b082b08, 0x2b2b2b2b), uvec2(0x2b2b2b2b, 0x2b2b2b2b)
 };
 
-shared uvec2 iq2s_grid[1024];
-
 #if defined(DATA_A_IQ2_S)
 #define NEEDS_INIT_IQ_SHMEM
 void init_iq_shmem(uvec3 wgsize)
 {
     // copy the table into shared memory and sync
-    [[unroll]] for (uint i = 0; i < iq2s_grid.length(); i += wgsize.x) {
-        if (iq2s_grid.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq2s_grid_const.length()) {
-            iq2s_grid[i + gl_LocalInvocationIndex.x] = iq2s_grid_const[i + gl_LocalInvocationIndex.x];
+    [[unroll]] for (uint i = 0; i < iq2_grid.length(); i += wgsize.x) {
+        if (iq2_grid.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq2s_grid_const.length()) {
+            iq2_grid[i + gl_LocalInvocationIndex.x] = iq2s_grid_const[i + gl_LocalInvocationIndex.x];
         }
     }
     barrier();
@@ -1508,6 +1521,16 @@ void init_iq_shmem(uvec3 wgsize)
 #endif
 #endif
 
+#if defined(DATA_A_IQ3_XXS) || defined(DATA_A_IQ3_S) || defined(MULMAT_QUANT)
+#ifdef MULMAT_QUANT
+shared uint32_t iq3_grid[MmTypeA == GGML_TYPE_IQ3_S ? 512 : MmTypeA == GGML_TYPE_IQ3_XXS ? 256 : 1];
+#elif defined(DATA_A_IQ3_S)
+shared uint32_t iq3_grid[512];
+#else
+shared uint32_t iq3_grid[256];
+#endif
+#endif
+
 #define QUANT_K_IQ3_XXS 256
 #define QUANT_R_IQ3_XXS 1
 
@@ -1560,16 +1583,14 @@ const uint32_t iq3xxs_grid_const[256] = {
     0x3e1c1c1c, 0x3e1c3404, 0x3e24140c, 0x3e24240c, 0x3e2c0404, 0x3e2c0414, 0x3e2c1424, 0x3e341c04,
 };
 
-shared uint32_t iq3xxs_grid[256];
-
 #if defined(DATA_A_IQ3_XXS)
 #define NEEDS_INIT_IQ_SHMEM
 void init_iq_shmem(uvec3 wgsize)
 {
     // copy the table into shared memory and sync
-    [[unroll]] for (uint i = 0; i < iq3xxs_grid.length(); i += wgsize.x) {
-        if (iq3xxs_grid.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq3xxs_grid.length()) {
-            iq3xxs_grid[i + gl_LocalInvocationIndex.x] = iq3xxs_grid_const[i + gl_LocalInvocationIndex.x];
+    [[unroll]] for (uint i = 0; i < iq3_grid.length(); i += wgsize.x) {
+        if (iq3_grid.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq3_grid.length()) {
+            iq3_grid[i + gl_LocalInvocationIndex.x] = iq3xxs_grid_const[i + gl_LocalInvocationIndex.x];
         }
     }
     barrier();
@@ -1674,16 +1695,14 @@ const uint32_t iq3s_grid_const[512] = {
     0x0f090307, 0x0f090501, 0x0f090b01, 0x0f0b0505, 0x0f0b0905, 0x0f0d0105, 0x0f0d0703, 0x0f0f0101,
 };
 
-shared uint32_t iq3s_grid[512];
-
 #if defined(DATA_A_IQ3_S)
 #define NEEDS_INIT_IQ_SHMEM
 void init_iq_shmem(uvec3 wgsize)
 {
     // copy the table into shared memory and sync
-    [[unroll]] for (uint i = 0; i < iq3s_grid.length(); i += wgsize.x) {
-        if (iq3s_grid.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq3s_grid.length()) {
-            iq3s_grid[i + gl_LocalInvocationIndex.x] = iq3s_grid_const[i + gl_LocalInvocationIndex.x];
+    [[unroll]] for (uint i = 0; i < iq3_grid.length(); i += wgsize.x) {
+        if (iq3_grid.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq3_grid.length()) {
+            iq3_grid[i + gl_LocalInvocationIndex.x] = iq3s_grid_const[i + gl_LocalInvocationIndex.x];
         }
     }
     barrier();
@@ -1808,7 +1827,11 @@ const int8_t kvalues_iq4nl_const[16] = {
     int8_t(1), int8_t(13), int8_t(25), int8_t(38), int8_t(53), int8_t(69), int8_t(89), int8_t(113)
 };
 
+#ifdef MULMAT_QUANT
+shared FLOAT_TYPE kvalues_iq4nl[(MmTypeA == GGML_TYPE_IQ4_NL || MmTypeA == GGML_TYPE_IQ4_XS) ? 16 : 1];
+#else
 shared FLOAT_TYPE kvalues_iq4nl[16];
+#endif
 
 #if defined(DATA_A_IQ4_NL) || defined(DATA_A_IQ4_XS)
 #define NEEDS_INIT_IQ_SHMEM
@@ -1830,12 +1853,19 @@ const int8_t kvalues_mxfp4_const[16] = {
     int8_t(0), int8_t(-1), int8_t(-2), int8_t(-3), int8_t(-4), int8_t(-6), int8_t(-8), int8_t(-12),
 };
 
+#ifdef MULMAT_QUANT
+shared int8_t kvalues_mxfp4[(MmTypeA == GGML_TYPE_MXFP4 || MmTypeA == GGML_TYPE_NVFP4) ? 16 : 1];
+#else
 shared int8_t kvalues_mxfp4[16];
 #endif
+#endif
 
 #if (defined(DATA_A_NVFP4) || defined(MULMAT_QUANT)) && !defined(USE_OCP_FP4)
-// UE4M3 scale in NVFP4 blocks use only 7 bits; sign (bit 7) is always zero.
+#ifdef MULMAT_QUANT
+shared float ue4m3_fp32_lut[MmTypeA == GGML_TYPE_NVFP4 ? 128 : 1];
+#else
 shared float ue4m3_fp32_lut[128];
+#endif
 
 float ue4m3_to_fp32_build(uint u) {
     if (u == 0u || u == 127u) {