From: 0cc4m <redacted>
Date: Mon, 5 Aug 2024 05:52:55 +0000 (+0200)
Subject: vulkan : fix Qantized Mat-Vec Mul on AMD GPUs for ncols < 64 (llama/8855)
X-Git-Tag: upstream/0.0.1642~467
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=a3b2059cacacd97985b4d814f464779d6e1545f2;p=pkg%2Fggml%2Fsources%2Fggml

vulkan : fix Qantized Mat-Vec Mul on AMD GPUs for ncols < 64 (llama/8855)

* Fix Vulkan mul mat vec invalid results when ncols < warp size

* Only run backend ops mul mat vec block size test if block size not already covered
---

diff --git a/tests/test-backend-ops.cpp b/tests/test-backend-ops.cpp
index f5065f14..54cef05c 100644
--- a/tests/test-backend-ops.cpp
+++ b/tests/test-backend-ops.cpp
@@ -2271,9 +2271,10 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op
 
     for (ggml_type type_a : other_types) {
         for (ggml_type type_b : {GGML_TYPE_F32}) {
-
-            test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, ggml_blck_size(type_a), { 1,  1}, {1, 1}));
-            test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, { 1,  1}, {1, 1}));
+            if (ggml_blck_size(type_a) != 256) {
+                test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, ggml_blck_size(type_a), {1,  1}, {1, 1}));
+            }
+            test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {1,  1}, {1, 1}));
         }
     }