vulkan: fix quantizing issue when tensor is not divisible by 128

Author: 0cc4m

ggml/src/ggml-vulkan/ggml-vulkan.cpp

@@ -2792,7 +2792,7 @@ static void ggml_vk_load_shaders(vk_device& device) {
         uint32_t wg_size_subgroup16 = (w == DMMV_WG_SIZE_SUBGROUP) ? subgroup_size_16 : (subgroup_size_16 * 4);
         uint32_t wg_size_subgroup   = (w == DMMV_WG_SIZE_SUBGROUP) ? device->subgroup_size : (device->subgroup_size * 4);
 
-        const bool s = device->subgroup_add && device->architecture != vk_device_architecture::AMD_GCN;
+        const bool s = device->subgroup_arithmetic && device->architecture != vk_device_architecture::AMD_GCN;
 
         for (uint32_t i = 0; i < mul_mat_vec_max_cols; ++i) {
             ggml_vk_create_pipeline(device, device->pipeline_dequant_mul_mat_vec_f32_f32[w][GGML_TYPE_F32 ][i], "mul_mat_vec_f32_f32_f32",  arr_dmmv_f32_f32_f32_len[s],  arr_dmmv_f32_f32_f32_data[s],  "main", 3, sizeof(vk_mat_vec_push_constants), {2, 1, 1}, {wg_size_subgroup, 2, i+1}, 1);
@@ -2843,8 +2843,10 @@ static void ggml_vk_load_shaders(vk_device& device) {
             ggml_vk_create_pipeline(device, device->pipeline_dequant_mul_mat_vec_f16_f32[w][GGML_TYPE_IQ4_NL][i],  "mul_mat_vec_iq4_nl_f16_f32",  arr_dmmv_iq4_nl_f16_f32_len[s],  arr_dmmv_iq4_nl_f16_f32_data[s],  "main", 3, sizeof(vk_mat_vec_push_constants), {rm_iq, 1, 1}, {wg_size_subgroup16, rm_iq, i+1}, 1, true);
             ggml_vk_create_pipeline(device, device->pipeline_dequant_mul_mat_vec_f16_f32[w][GGML_TYPE_MXFP4][i],   "mul_mat_vec_mxfp4_f16_f32",   arr_dmmv_mxfp4_f16_f32_len[s],   arr_dmmv_mxfp4_f16_f32_data[s],   "main", 3, sizeof(vk_mat_vec_push_constants), {rm_iq, 1, 1}, {wg_size_subgroup16, rm_iq, i+1}, 1, true);
         }
+    }
 
 #if defined(GGML_VULKAN_INTEGER_DOT_GLSLC_SUPPORT)
+    for (uint32_t i = 0; i < mul_mat_vec_max_cols; ++i) {
         if (device->integer_dot_product) {
             const uint32_t subgroup_size = (device->subgroup_size_control && device->vendor_id == VK_VENDOR_ID_INTEL) ? device->subgroup_min_size : device->subgroup_size;
             if (device->subgroup_arithmetic && device->subgroup_require_full_support) {
@@ -2861,8 +2863,8 @@ static void ggml_vk_load_shaders(vk_device& device) {
                 ggml_vk_create_pipeline(device, device->pipeline_dequant_mul_mat_vec_q8_1_f32[GGML_TYPE_Q8_0][i], "mul_mat_vec_q8_0_q8_1_f32_"+std::to_string(i+1), mul_mat_vec_q8_0_q8_1_f32_len, mul_mat_vec_q8_0_q8_1_f32_data, "main", 3, sizeof(vk_mat_vec_push_constants), {1*rm_stdq, 1, 1}, {subgroup_size, 1*rm_stdq, i+1}, 1, true);
             }
         }
-#endif // GGML_VULKAN_INTEGER_DOT_GLSLC_SUPPORT
     }
+#endif // GGML_VULKAN_INTEGER_DOT_GLSLC_SUPPORT
 
     ggml_vk_create_pipeline(device, device->pipeline_dequant_mul_mat_vec_id_f32[GGML_TYPE_F32 ], "mul_mat_vec_id_f32_f32",  mul_mat_vec_id_f32_f32_len,  mul_mat_vec_id_f32_f32_data,  "main", 4, sizeof(vk_mat_vec_id_push_constants), {2, 1, 1}, {device->subgroup_size, 2}, 1);
     ggml_vk_create_pipeline(device, device->pipeline_dequant_mul_mat_vec_id_f32[GGML_TYPE_F16 ], "mul_mat_vec_id_f16_f32",  mul_mat_vec_id_f16_f32_len,  mul_mat_vec_id_f16_f32_data,  "main", 4, sizeof(vk_mat_vec_id_push_constants), {2, 1, 1}, {device->subgroup_size, 2}, 1);
@@ -5622,7 +5624,10 @@ static void ggml_vk_mul_mat_q_f16(ggml_backend_vk_context * ctx, vk_context& sub
 
     if (dryrun) {
         const uint64_t x_sz_upd = x_sz * ne02 * ne03;
-        const uint64_t y_sz_upd = y_sz * ne12 * ne13;
+        uint64_t y_sz_upd = y_sz * ne12 * ne13;
+        if (quantize_y) {
+            y_sz_upd = CEIL_DIV(y_sz_upd, 144) * 144;
+        }
         const uint64_t split_k_size = split_k > 1 ? d_sz * ne12 * ne13 * split_k : 0;
         if (
                 (qx_needs_dequant && x_sz_upd > ctx->device->max_memory_allocation_size) ||
@@ -5634,7 +5639,7 @@ static void ggml_vk_mul_mat_q_f16(ggml_backend_vk_context * ctx, vk_context& sub
             ctx->prealloc_size_x = x_sz_upd;
         }
         if ((qy_needs_dequant || quantize_y) && ctx->prealloc_size_y < y_sz_upd) {
-            ctx->prealloc_size_y = CEIL_DIV(y_sz_upd, 128) * 128;
+            ctx->prealloc_size_y = y_sz_upd;
         }
         if (split_k > 1 && ctx->prealloc_size_split_k < split_k_size) {
             ctx->prealloc_size_split_k = split_k_size;
@@ -5688,7 +5693,7 @@ static void ggml_vk_mul_mat_q_f16(ggml_backend_vk_context * ctx, vk_context& sub
         GGML_ASSERT(d_Y->size >= y_sz * ne12 * ne13);
     } else if (quantize_y) {
         d_Y = ctx->prealloc_y;
-        GGML_ASSERT(d_Y->size >= y_ne * ggml_type_size(GGML_TYPE_Q8_1) / ggml_blck_size(GGML_TYPE_Q8_1));
+        GGML_ASSERT(d_Y->size >= CEIL_DIV(y_sz * ne12 * ne13, 144) * 144);
     } else {
         d_Y = d_Qy;
         y_buf_offset = qy_buf_offset;
@@ -5730,10 +5735,15 @@ static void ggml_vk_mul_mat_q_f16(ggml_backend_vk_context * ctx, vk_context& sub
         stride_batch_y = src1->nb[0] / ggml_type_size(src1->type);
     }
 
+    uint32_t y_sz_total = y_sz * ne12 * ne13;
+    if (quantize_y) {
+        y_sz_total = CEIL_DIV(y_sz_total, 144) * 144;
+    }
+
     // compute
     ggml_vk_matmul(
         ctx, subctx, pipeline,
-        { d_X, x_buf_offset, x_sz * ne02 * ne03 }, { d_Y, y_buf_offset, y_sz * ne12 * ne13 },
+        { d_X, x_buf_offset, x_sz * ne02 * ne03 }, { d_Y, y_buf_offset, y_sz_total },
         { d_D, d_buf_offset, d_sz * ne12 * ne13 }, { ctx->prealloc_split_k, 0, d_sz * ne12 * ne13 * split_k },
         ne01, ne11, ne10,
         ne10, ne10, ne01, stride_batch_x, stride_batch_y, ne20*ne21,
@@ -5844,7 +5854,10 @@ static void ggml_vk_mul_mat_vec_q_f16(ggml_backend_vk_context * ctx, vk_context&
 
     if (dryrun) {
         const uint64_t x_sz_upd = x_sz * ne02 * ne03;
-        const uint64_t y_sz_upd = y_sz * ne12 * ne13;
+        uint64_t y_sz_upd = y_sz * ne12 * ne13;
+        if (quantize_y) {
+            y_sz_upd = CEIL_DIV(y_sz_upd, 144) * 144;
+        }
         if (
                 (qx_needs_dequant && x_sz_upd > ctx->device->max_memory_allocation_size) ||
                 (qy_needs_dequant && y_sz_upd > ctx->device->max_memory_allocation_size)) {
@@ -5854,7 +5867,7 @@ static void ggml_vk_mul_mat_vec_q_f16(ggml_backend_vk_context * ctx, vk_context&
             ctx->prealloc_size_x = x_sz_upd;
         }
         if ((qy_needs_dequant || quantize_y) && ctx->prealloc_size_y < y_sz_upd) {
-            ctx->prealloc_size_y = CEIL_DIV(y_sz_upd, 128) * 128;
+            ctx->prealloc_size_y = y_sz_upd;
         }
 
         // Request descriptor sets
@@ -5899,7 +5912,7 @@ static void ggml_vk_mul_mat_vec_q_f16(ggml_backend_vk_context * ctx, vk_context&
         d_Y = ctx->prealloc_y;
     } else if (quantize_y) {
         d_Y = ctx->prealloc_y;
-        GGML_ASSERT(d_Y->size >= CEIL_DIV(y_sz * ne12 * ne13, 128) * 128);
+        GGML_ASSERT(d_Y->size >= CEIL_DIV(y_sz * ne12 * ne13, 144) * 144);
     } else {
         d_Y = d_Qy;
         y_buf_offset = qy_buf_offset;
@@ -5946,6 +5959,12 @@ static void ggml_vk_mul_mat_vec_q_f16(ggml_backend_vk_context * ctx, vk_context&
         groups_x = CEIL_DIV(groups_x, groups_z);
     }
 
+    // TODO: Clean up this whole sz * ne_2 * ne_3 thing, it hasn't been necessary for a long time
+    uint32_t y_sz_total = y_sz * ne12 * ne13;
+    if (quantize_y) {
+        y_sz_total = CEIL_DIV(y_sz_total, 144) * 144;
+    }
+
     // compute
     const vk_mat_vec_push_constants pc = {
         (uint32_t)ne00, (uint32_t)ne10, (uint32_t)ne10, (uint32_t)ne01,
@@ -5954,7 +5973,7 @@ static void ggml_vk_mul_mat_vec_q_f16(ggml_backend_vk_context * ctx, vk_context&
     };
     ggml_vk_sync_buffers(subctx);
     ggml_vk_dispatch_pipeline(ctx, subctx, dmmv,
-                              { vk_subbuffer{ d_X, x_buf_offset, x_sz * ne02 * ne03 }, vk_subbuffer{ d_Y, y_buf_offset, y_sz * ne12 * ne13 }, vk_subbuffer{ d_D, d_buf_offset, d_sz * ne22 * ne23} },
+                              { vk_subbuffer{ d_X, x_buf_offset, x_sz * ne02 * ne03 }, vk_subbuffer{ d_Y, y_buf_offset, y_sz_total }, vk_subbuffer{ d_D, d_buf_offset, d_sz * ne22 * ne23} },
                               pc, { groups_x, (uint32_t)(ne12 * ne13), groups_z });
 }
 

ggml/src/ggml-vulkan/vulkan-shaders/quantize_q8_1.comp

@@ -45,13 +45,18 @@ void quantize() {
     const uint ib = wgid * blocks_per_group + block_in_wg;
     const uint iqs = tid % 8;
 
+#ifndef QBLOCK_X4
     if (ib >= gl_NumWorkGroups.x * blocks_per_group) {
         return;
     }
-
-#ifdef QBLOCK_X4
+#else
     const uint ibx4_outer = ib / 4;
     const uint ibx4_inner = ib % 4;
+
+    const uint required_x4_blocks = (p.ne + 127) / 128;
+    if (ibx4_outer >= required_x4_blocks) {
+        return;
+    }
 #endif
 
     const uint a_idx = ib * 8 + iqs;