meta-intel/recipes-oneapi/crypto/intel-crypto-mb/0002-cmake-exclude-Yocto-build-flags.patch
Naveen Saini f8ce4df052 intel-crypto-mb: upgrade 2021.3 -> 2021.5
Intel IPP Cryptography uses multiple implementations of each function,
optimized for various CPUs, and the library version targeted
for any CPU contains all of these implementations.

With the dispatcher, the library detects an available CPU in a runtime
and chooses the best for the current hardware version of a function, hence
ignore Yocto march, mtune values and let the project
pass those values along with the right optimization flags.

https://github.com/intel/ipp-crypto/blob/ippcp_2021.5/OVERVIEW.md#dispatcher

Signed-off-by: Naveen Saini <naveen.kumar.saini@intel.com>
Signed-off-by: Anuj Mittal <anuj.mittal@intel.com>
2022-03-21 21:42:52 +08:00

46 lines
2.2 KiB
Diff

From b4549bb765d279b5ba042c6340e1dd69d0890b64 Mon Sep 17 00:00:00 2001
From: Naveen Saini <naveen.kumar.saini@intel.com>
Date: Thu, 10 Mar 2022 14:30:01 +0800
Subject: [PATCH 2/2] cmake: exclude Yocto build flags
Ipp-crypto has its own set of flags and -march values,
which causes conflict with default -march=nehalem in gcc
Upstream-Status: Inappropriate
Signed-off-by: Naveen Saini <naveen.kumar.saini@intel.com>
---
sources/ippcp/crypto_mb/src/CMakeLists.txt | 2 +-
sources/ippcp/crypto_mb/src/cmake/linux/GNU.cmake | 2 +-
2 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/sources/ippcp/crypto_mb/src/CMakeLists.txt b/sources/ippcp/crypto_mb/src/CMakeLists.txt
index 1ca1a8c..5284d62 100644
--- a/sources/ippcp/crypto_mb/src/CMakeLists.txt
+++ b/sources/ippcp/crypto_mb/src/CMakeLists.txt
@@ -81,7 +81,7 @@ endif()
set(MB_LIB_TARGET ${MB_DYN_LIB_TARGET})
set_source_files_properties(${CRYPTO_MB_SOURCES} PROPERTIES COMPILE_DEFINITIONS "${AVX512_LIBRARY_DEFINES}"
- COMPILE_FLAGS "${AVX512_CFLAGS} ${CMAKE_ASM_FLAGS} ${CMAKE_C_FLAGS_SECURITY}")
+ COMPILE_FLAGS "${AVX512_CFLAGS} ${CMAKE_C_FLAGS_SECURITY}")
# Don't specify architectural flags for the assembler for this sources, because of the bug in Intel® C Compiler under MacOS: error: invalid instruction mnemonic 'vkmovb'
# The bug has been fixed since version 2021.3. This is a workaround to support older versions of Intel® C Compiler.
diff --git a/sources/ippcp/crypto_mb/src/cmake/linux/GNU.cmake b/sources/ippcp/crypto_mb/src/cmake/linux/GNU.cmake
index a2abeeb..aadd6e2 100644
--- a/sources/ippcp/crypto_mb/src/cmake/linux/GNU.cmake
+++ b/sources/ippcp/crypto_mb/src/cmake/linux/GNU.cmake
@@ -58,7 +58,7 @@ set(LINK_FLAGS_DYNAMIC "${LINK_FLAGS_DYNAMIC} ${CRYPTO_MB_SOURCES_DIR}/cmake/dll
# Compiler flags
# Tells the compiler to align functions and loops
-set(CMAKE_C_FLAGS " -falign-functions=32 -falign-loops=32")
+set(CMAKE_C_FLAGS " -falign-functions=32 -falign-loops=32 ${TOOLCHAIN_OPTIONS}")
# Ensures that compilation takes place in a freestanding environment
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -ffreestanding")
--
2.17.1