From: Atharva Dubey <redacted>
Date: Thu, 3 Apr 2025 12:12:39 +0000 (+0100)
Subject: ci : add env variable in ggml-ci and document the same in SYCL.md (#12736)
X-Git-Tag: gguf-v0.16.1~114
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=2004644b7a5da6fe080e51861ab583480280f1d3;p=pkg%2Fggml%2Fsources%2Fllama.cpp

ci : add env variable in ggml-ci and document the same in SYCL.md (#12736)
---

diff --git a/ci/run.sh b/ci/run.sh
index 4d9c2dc4..ebd662b3 100755
--- a/ci/run.sh
+++ b/ci/run.sh
@@ -59,6 +59,8 @@ if [ ! -z ${GG_BUILD_SYCL} ]; then
     export ONEAPI_DEVICE_SELECTOR="level_zero:0"
     # Enable sysman for correct memory reporting
     export ZES_ENABLE_SYSMAN=1
+    # to circumvent precision issues on CPY operations
+    export SYCL_PROGRAM_COMPILE_OPTIONS="-cl-fp32-correctly-rounded-divide-sqrt"
     CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_SYCL=1 -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL_F16=ON"
 fi
 
diff --git a/docs/backend/SYCL.md b/docs/backend/SYCL.md
index 19fe8a9d..f1204dde 100644
--- a/docs/backend/SYCL.md
+++ b/docs/backend/SYCL.md
@@ -302,6 +302,10 @@ cmake -B build -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -
 cmake --build build --config Release -j -v
 ```
 
+It is possible to come across some precision issues when running tests that stem from using faster
+instructions, which can be circumvented by setting the environment variable `SYCL_PROGRAM_COMPILE_OPTIONS`
+as `-cl-fp32-correctly-rounded-divide-sqrt`
+
 #### Nvidia GPU
 
 The SYCL backend depends on [oneMath](https://github.com/uxlfoundation/oneMath) for Nvidia and AMD devices.
@@ -322,6 +326,9 @@ cmake -B build -DGGML_SYCL=ON -DGGML_SYCL_TARGET=NVIDIA -DGGML_SYCL_DEVICE_ARCH=
 cmake --build build --config Release -j -v
 ```
 
+It is possible to come across some precision issues when running tests that stem from using faster
+instructions, which can be circumvented by passing the `-fno-fast-math` flag to the compiler.
+
 #### AMD GPU
 
 The SYCL backend depends on [oneMath](https://github.com/uxlfoundation/oneMath) for Nvidia and AMD devices.