vulkan: Split large mul_mat_id to fit in shared memory

Author: jeffbolznv

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

@@ -5964,7 +5964,30 @@ static void ggml_vk_mul_mat_id(ggml_backend_vk_context * ctx, vk_context& subctx
     if (src2->ne[1] == 1 && (src0->type == GGML_TYPE_F32 || src0->type == GGML_TYPE_F16 || ggml_is_quantized(src0->type))) {
         ggml_vk_mul_mat_vec_id_q_f16(ctx, subctx, src0, src1, src2, dst, dryrun);
     } else {
-        ggml_vk_mul_mat_id_q_f16(ctx, subctx, src0, src1, src2, dst, dryrun);
+        // Split based on number of ids, to fix in shared memory
+        const uint32_t nei0 = (uint32_t)src2->ne[0];
+        const uint32_t nei1 = (uint32_t)src2->ne[1];
+
+        GGML_ASSERT(nei0 <= 4096);
+        const uint32_t split_size = std::min(nei1, 4096u / nei0);
+
+        for (uint32_t token_start = 0; token_start < nei1; token_start += split_size) {
+            const uint32_t n_tokens = std::min(split_size, nei1 - token_start);
+
+            ggml_tensor src1_copy = *src1;
+            ggml_tensor src2_copy = *src2;
+            ggml_tensor dst_copy = *dst;
+
+            src1_copy.view_offs += token_start * src1_copy.nb[2];
+            src2_copy.view_offs += token_start * src2_copy.nb[1];
+            dst_copy.view_offs += token_start * dst_copy.nb[2];
+
+            src1_copy.ne[2] = n_tokens;
+            src2_copy.ne[1] = n_tokens;
+            dst_copy.ne[2] = n_tokens;
+
+            ggml_vk_mul_mat_id_q_f16(ctx, subctx, src0, &src1_copy, &src2_copy, &dst_copy, dryrun);
+        }
     }
 }
 
@@ -10127,9 +10150,15 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
                 ggml_type src0_type = op->src[0]->type;
                 ggml_backend_vk_device_context * ctx = (ggml_backend_vk_device_context *)dev->context;
                 const vk_device& device = ggml_vk_get_device(ctx->device);
-                if (op->op == GGML_OP_MUL_MAT_ID && !device->mul_mat_id_s[src0_type] && !device->mul_mat_id_m[src0_type] && !device->mul_mat_id_l[src0_type]) {
-                    // If there's not enough shared memory for row_ids and the result tile, fallback to CPU
-                    return false;
+                if (op->op == GGML_OP_MUL_MAT_ID) {
+                    if (!device->mul_mat_id_s[src0_type] && !device->mul_mat_id_m[src0_type] && !device->mul_mat_id_l[src0_type]) {
+                        // If there's not enough shared memory for row_ids and the result tile, fallback to CPU
+                        return false;
+                    }
+                    // Check against size of shared memory variable
+                    if (op->src[2]->ne[0] > 4096) {
+                        return false;
+                    }
                 }
                 switch (src0_type) {
                     case GGML_TYPE_F32:

tests/test-backend-ops.cpp

@@ -4623,6 +4623,11 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
     // this case is verified (pass) in Intel(R) Data Center GPU Max 1100 (sycl backend) and NV A30 (cuda backend)
     // test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F16, 512, 262144, 9216, {1, 1}, {1, 1}));
 
+    // test large experts*tokens
+    for (bool b : {false, true}) {
+        test_cases.emplace_back(new test_mul_mat_id(GGML_TYPE_F16, GGML_TYPE_F32, 16, 16, b, 32, 1024, 16));
+    }
+
     for (ggml_type type_a : base_types) {
         for (ggml_type type_b : {GGML_TYPE_F32 /*, GGML_TYPE_F16 */}) {
             for (int n_mats : {4, 8}) {