Update on "Optimize Vulkan command buffer submission rate."

Differential Revision: [D25729889](https://our.internmc.facebook.com/intern/diff/D25729889) [ghstack-poisoned]
pytorch · Jan 8, 2021 · 9c1750d · 9c1750d
2 parents 8742826 + bd4c29c
commit 9c1750d
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -207,7 +207,7 @@ cmake_dependent_option(
     USE_VALGRIND "Use Valgrind. Only available on Linux." ON
     "LINUX" OFF)
 option(USE_VULKAN "Use Vulkan GPU backend" OFF)
-option(USE_VULKAN_FP16_INFERENCE "Vulkan - Use fp16 inference even on fp32 tensors" ON)
+option(USE_VULKAN_FP16_INFERENCE "Vulkan - Use fp16 inference even on fp32 tensors" OFF)
 option(USE_VULKAN_RELAXED_PRECISION "Vulkan - Use relaxed precision math in the kernels (mediump)" OFF)
 option(USE_VULKAN_SHADERC_RUNTIME "Vulkan - Use runtime shader compilation (needs libshaderc)" OFF)
 option(USE_VULKAN_WRAPPER "Vulkan - Dynamically load Vulkan functions" ON)

diff --git a/aten/src/ATen/native/vulkan/ops/Utils.h b/aten/src/ATen/native/vulkan/ops/Utils.h
@@ -10,7 +10,7 @@ namespace vulkan {
 namespace ops {
 namespace utils {
 
-int64_t normalize(
+inline int64_t normalize(
     const int64_t dimension,
     const int64_t n) {
   return (dimension % n + n) % n;