diff --git a/src/gpu/ganesh/vk/GrVkCaps.cpp b/src/gpu/ganesh/vk/GrVkCaps.cpp
index 23b969a..a04c1f8 100644
--- a/src/gpu/ganesh/vk/GrVkCaps.cpp
+++ b/src/gpu/ganesh/vk/GrVkCaps.cpp
@@ -606,6 +606,12 @@
     if (kImagination_VkVendor == properties.vendorID) {
         fShaderCaps->fAtan2ImplementedAsAtanYOverX = true;
     }
+
+    // ARM GPUs calculate `matrix * vector` in SPIR-V at full precision, even when the inputs are
+    // RelaxedPrecision. Rewriting the multiply as a sum of vector*scalar fixes this. (skia:11769)
+    if (kARM_VkVendor == properties.vendorID) {
+        fShaderCaps->fRewriteMatrixVectorMultiply = true;
+    }
 }
 
 void GrVkCaps::initGrCaps(const GrVkInterface* vkInterface,
@@ -703,10 +709,6 @@
 
     shaderCaps->fShaderDerivativeSupport = true;
 
-    // ARM GPUs calculate `matrix * vector` in SPIR-V at full precision, even when the inputs are
-    // RelaxedPrecision. Rewriting the multiply as a sum of vector*scalar fixes this. (skia:11769)
-    shaderCaps->fRewriteMatrixVectorMultiply = (kARM_VkVendor == properties.vendorID);
-
     shaderCaps->fDualSourceBlendingSupport = features.features.dualSrcBlend;
 
     shaderCaps->fIntegerSupport = true;