Merge branch 'OpenMathLib:develop' into fix_dyn_armv9sme

Author: martin-frbg

.cirrus.yml

@@ -58,8 +58,8 @@ task:
   - export VALID_ARCHS="i386 x86_64"
   - xcrun --sdk macosx --show-sdk-path
   - xcodebuild -version
-  - export CC=/Applications/Xcode_15.4.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
-  - export CFLAGS="-O2 -unwindlib=none -Wno-macro-redefined -isysroot /Applications/Xcode_15.4.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX14.5.sdk -arch x86_64"
+  - export CC=/Applications/Xcode_16.3.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
+  - export CFLAGS="-O2 -unwindlib=none -Wno-macro-redefined -isysroot /Applications/Xcode_16.3.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX15.4.sdk -arch x86_64"
   - make TARGET=CORE2 DYNAMIC_ARCH=1 NUM_THREADS=32 HOSTCC=clang NOFORTRAN=1 RANLIB="ls -l"
   always:
     config_artifacts:
@@ -78,8 +78,8 @@ task:
   - export #PATH=/opt/homebrew/opt/llvm/bin:$PATH
   - export #LDFLAGS="-L/opt/homebrew/opt/llvm/lib"
   - export #CPPFLAGS="-I/opt/homebrew/opt/llvm/include" 
-  - export CC=/Applications/Xcode_15.4.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
-  - export CFLAGS="-O2 -unwindlib=none -Wno-macro-redefined -isysroot /Applications/Xcode_15.4.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS17.5.sdk -arch arm64 -miphoneos-version-min=10.0"
+  - export CC=/Applications/Xcode_16.3.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
+  - export CFLAGS="-O2 -unwindlib=none -Wno-macro-redefined -isysroot /Applications/Xcode_16.3.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS18.4.sdk -arch arm64 -miphoneos-version-min=10.0"
   - xcrun --sdk iphoneos --show-sdk-path
   - ls -l /Applications
   - make TARGET=ARMV8 NUM_THREADS=32 HOSTCC=clang NOFORTRAN=1 CROSS=1

kernel/arm64/sgemm_direct_arm64_sme1.c

@@ -7,7 +7,6 @@
 #include <stdlib.h>
 #include <inttypes.h>
 #include <math.h>
-
 #if defined(HAVE_SME)
 
 /* Function prototypes */
@@ -44,15 +43,31 @@ void CNAME (BLASLONG M, BLASLONG N, BLASLONG K, float * __restrict A,\
         m_mod = ceil((double)M/(double)vl_elms) * vl_elms;
 
         float *A_mod = (float *) malloc(m_mod*K*sizeof(float));
-		
+	    
+	    /* Prevent compiler optimization by reading from memory instead
+	     * of reading directly from vector (z) registers.
+	     * */
+        asm volatile("" : : :"p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7",
+                         "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15",
+                         "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7",
+                         "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15",
+                         "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23",
+                         "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31");
+      
         /* Pre-process the left matrix to make it suitable for 
            matrix sum of outer-product calculation
          */
         sgemm_direct_sme1_preprocess(M, K, A, A_mod);
 
         /* Calculate C = A*B */
         sgemm_direct_sme1_2VLx2VL(M, K, N, A_mod, B, R);
-        
+       
+        asm volatile("" : : :"p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7",
+                         "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15",
+                         "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7",
+                         "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15",
+                         "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23",
+                         "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31");
         free(A_mod);
 }