Work around Clang failing to disambiguate SVE intrinsics and add AppleClang crossbuild to MacOS/arm64 DYNAMIC_ARCH in AzureCI (#4140)

* Add AppleClang crossbuild to MacOS/arm64 DYNAMIC_ARCH

* add casts to disambiguate svwhilelt for clang

azure-pipelines.yml

@@ -271,6 +271,19 @@ jobs:
   - script: |
      make TARGET=ARMV7 DYNAMIC_ARCH=1 NUM_THREADS=32 HOSTCC=clang NOFORTRAN=1
 
+- job: OSX_xbuild_DYNAMIC_ARM64
+  pool:
+     vmImage: 'macOS-11'
+  variables:
+     CC: /Applications/Xcode_12.5.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
+     CFLAGS: -O2 -Wno-macro-redefined -isysroot /Applications/Xcode_12.5.1.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX11.3.sdk -arch arm64
+  steps:
+  - script: |
+     ls /Applications/Xcode_12.5.1.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs
+     /Applications/Xcode_12.5.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang -arch arm64 --print-supported-cpus
+     /Applications/Xcode_11.7.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang --version
+     make TARGET=ARMV8 DYNAMIC_ARCH=1 NUM_THREADS=32 HOSTCC=clang NOFORTRAN=1
+
 - job: ALPINE_MUSL
   pool:
      vmImage: 'ubuntu-latest'

kernel/arm64/dot_kernel_sve.c

@@ -50,8 +50,8 @@ static FLOAT dot_kernel_sve(BLASLONG n, FLOAT *x, FLOAT *y) {
         BLASLONG sve_width = SVE_WIDTH;
 
         for (BLASLONG i = 0; i < n; i += sve_width * 2) {
-                svbool_t pg_a = SVE_WHILELT(i, n);
-                svbool_t pg_b = SVE_WHILELT(i + sve_width, n);
+                svbool_t pg_a = SVE_WHILELT((uint64_t)i, (uint64_t)n);
+                svbool_t pg_b = SVE_WHILELT((uint64_t)(i + sve_width), (uint64_t)n);
 
                 SVE_TYPE x_vec_a = svld1(pg_a, &x[i]);
                 SVE_TYPE y_vec_a = svld1(pg_a, &y[i]);

kernel/arm64/gemm_ncopy_sve_v1x8.c

@@ -107,7 +107,7 @@ int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b) {
     BLASLONG remaining_n = n - single_vectors_n;
     if (remaining_n) {
         a_offset_inner = a_offset;
-        svbool_t pg = SV_WHILE(0L, remaining_n);
+        svbool_t pg = SV_WHILE((uint64_t)0L, (uint64_t)remaining_n);
         uint64_t active = remaining_n;
         uint64_t i_cnt = m >> 2;
         while (i_cnt--) {

kernel/arm64/gemm_tcopy_sve_v1x8.c

@@ -100,7 +100,7 @@ int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
     BLASLONG remaining_n = n - single_vectors_n;
     if (remaining_n) {
         a_offset_inner = a_offset;
-        svbool_t pg = SV_WHILE(0L, remaining_n);
+        svbool_t pg = SV_WHILE((uint64_t)0L, (uint64_t)remaining_n);
         uint64_t active = remaining_n;
         uint64_t i_cnt = m >> 2;
         while (i_cnt--) {

kernel/arm64/trmm_lncopy_sve_v1.c

@@ -52,11 +52,11 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
     FLOAT *ao;
 #ifdef DOUBLE
     svint64_t index = svindex_s64(0LL, lda);
-    svbool_t pn = svwhilelt_b64(js, n);
+    svbool_t pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
     int n_active = svcntp_b64(svptrue_b64(), pn);
 #else
     svint32_t index = svindex_s32(0, lda);
-    svbool_t pn = svwhilelt_b32(js, n);
+    svbool_t pn = svwhilelt_b32((uint64_t)js, (uint64_t)n);
     int n_active = svcntp_b32(svptrue_b32(), pn);
 #endif
     do
@@ -123,11 +123,11 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
         posY += n_active;
         js += n_active;
 #ifdef DOUBLE
-        pn = svwhilelt_b64(js, n);
+        pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
         n_active = svcntp_b64(svptrue_b64(), pn);
     } while (svptest_any(svptrue_b64(), pn));
 #else
-        pn = svwhilelt_b32(js, n);
+        pn = svwhilelt_b32((uint64_t)js, (uint64_t)n);
         n_active = svcntp_b32(svptrue_b32(), pn);
     } while (svptest_any(svptrue_b32(), pn));
 #endif

kernel/arm64/trmm_ltcopy_sve_v1.c

@@ -51,10 +51,10 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
     FLOAT *ao;
     js = 0;
 #ifdef DOUBLE
-    svbool_t pn = svwhilelt_b64(js, n);
+    svbool_t pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
     int n_active = svcntp_b64(svptrue_b64(), pn);
 #else
-    svbool_t pn = svwhilelt_b32(js, n);
+    svbool_t pn = svwhilelt_b32((uint64_t)js, (uint64_t)n);
     int n_active = svcntp_b32(svptrue_b32(), pn);
 #endif
     do
@@ -122,11 +122,11 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
         posY += n_active;
         js += n_active;
 #ifdef DOUBLE
-        pn = svwhilelt_b64(js, n);
+        pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
         n_active = svcntp_b64(svptrue_b64(), pn);
     } while (svptest_any(svptrue_b64(), pn));
 #else
-        pn = svwhilelt_b32(js, n);
+        pn = svwhilelt_b32((uint64_t)js, (uint64_t)n);
         n_active = svcntp_b32(svptrue_b32(), pn);
     } while (svptest_any(svptrue_b32(), pn));
 #endif

kernel/arm64/trmm_uncopy_sve_v1.c

@@ -52,11 +52,11 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
     FLOAT *ao;
 #ifdef DOUBLE
     svint64_t index = svindex_s64(0LL, lda);
-    svbool_t pn = svwhilelt_b64(js, n);
+    svbool_t pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
     int n_active = svcntp_b64(svptrue_b64(), pn);
 #else
     svint32_t index = svindex_s32(0, lda);
-    svbool_t pn = svwhilelt_b32(js, n);
+    svbool_t pn = svwhilelt_b32((uint64_t)js, (uint64_t)n);
     int n_active = svcntp_b32(svptrue_b32(), pn);
 #endif
     do
@@ -123,11 +123,11 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
         posY += n_active;
         js += n_active;
 #ifdef DOUBLE
-        pn = svwhilelt_b64(js, n);
+        pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
         n_active = svcntp_b64(svptrue_b64(), pn);
     } while (svptest_any(svptrue_b64(), pn));
 #else
-        pn = svwhilelt_b32(js, n);
+        pn = svwhilelt_b32((uint64_t)js, (uint64_t)n);
         n_active = svcntp_b32(svptrue_b32(), pn);
     } while (svptest_any(svptrue_b32(), pn));
 #endif

kernel/arm64/trmm_utcopy_sve_v1.c

@@ -51,10 +51,10 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
     FLOAT *ao;
     js = 0;
 #ifdef DOUBLE
-    svbool_t pn = svwhilelt_b64(js, n);
+    svbool_t pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
     int n_active = svcntp_b64(svptrue_b64(), pn);
 #else
-    svbool_t pn = svwhilelt_b32(js, n);
+    svbool_t pn = svwhilelt_b32((uint64_t)js, (uint64_t)n);
     int n_active = svcntp_b32(svptrue_b32(), pn);
 #endif
     do
@@ -121,11 +121,11 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
         posY += n_active;
         js += n_active;
 #ifdef DOUBLE
-        pn = svwhilelt_b64(js, n);
+        pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
         n_active = svcntp_b64(svptrue_b64(), pn);
     } while (svptest_any(svptrue_b64(), pn));
 #else
-        pn = svwhilelt_b32(js, n);
+        pn = svwhilelt_b32((uint64_t)js, (uint64_t)n);
         n_active = svcntp_b32(svptrue_b32(), pn);
     } while (svptest_any(svptrue_b32(), pn));
 #endif

kernel/arm64/trsm_lncopy_sve.c

@@ -56,13 +56,13 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG offset, FLOAT
 #ifdef DOUBLE
   int64_t js = 0;
   svint64_t index = svindex_s64(0LL, lda);
-  svbool_t pn = svwhilelt_b64(js, n);
+  svbool_t pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
   int n_active = svcntp_b64(svptrue_b64(), pn);
 #else
   int32_t N = n;
   int32_t js = 0;
   svint32_t index = svindex_s32(0, lda);
-  svbool_t pn = svwhilelt_b32(js, N);
+  svbool_t pn = svwhilelt_b32((uint32_t)js, (uint32_t)N);
   int n_active = svcntp_b32(svptrue_b32(), pn);
 #endif
   do {
@@ -106,11 +106,11 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG offset, FLOAT
 
     js += n_active;
 #ifdef DOUBLE
-    pn = svwhilelt_b64(js, n);
+    pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
     n_active = svcntp_b64(svptrue_b64(), pn);
   } while (svptest_any(svptrue_b64(), pn));
 #else
-    pn = svwhilelt_b32(js, N);
+    pn = svwhilelt_b32((uint32_t)js, (uint32_t)N);
     n_active = svcntp_b32(svptrue_b32(), pn);
   } while (svptest_any(svptrue_b32(), pn));
 #endif