Merge pull request #242 from intel/develop

Develop

README.md

@@ -27,14 +27,20 @@ YASK contains a domain-specific compiler to convert stencil-equation specificati
   for multi-socket and multi-node operation or
   Intel(R) Parallel Studio XE Composer Edition for C++ Linux
   for single-socket only
-  (2018 or later; 2019 update 5 (2019.0.5 or 2019u5) or later recommended).
+  (2018 or later; 2019 update 5 (2019.0.5) recommended;
+   2019.1.x has an issue that prevents compilation).
      * There was an issue in Intel(R) MPI versions 2019u1 and 2019u2 that
        caused the application to crash when allocating very
        large shared-memory (shm) regions, so those
        versions are not recommended when using the `-use_shm` feature.
        This issue was resolved in MPI version 2019u3.
+     * There is an issue in the Intel C++ compiler 2019.1.x that causes
+       an internal error when building YASK kernels.
+       Use 2019.0.5 to work around until a patch is released.
      * If you are using g++ version 8.x or later, Intel(R) C++ version 2019
        is required.
+     * Building a YASK kernel with clang is possible; however,
+       SIMD operations for functions such as sin() is not supported.
      * Building a YASK kernel with the Gnu C++ compiler is possible.
        Limited testing with g++ 8.2.0 shows the "iso3dfd" kernel
        runs about 30% slower compared to the same kernel built with

src/common/common_utils.cpp

@@ -43,7 +43,7 @@ namespace yask {
     // for numbers above 9 (at least up to 99).
 
     // Format: "major.minor.patch".
-    const string version = "3.03.00";
+    const string version = "3.03.01";
 
     string yask_get_version_string() {
         return version;

src/kernel/lib/alloc.cpp

@@ -35,7 +35,7 @@ namespace yask {
     static inline int getnode() {
         #ifdef SYS_getcpu
         int node, status;
-       status = syscall(SYS_getcpu, NULL, &node, NULL);
+        status = syscall(SYS_getcpu, NULL, &node, NULL);
         return (status == -1) ? status : node;
         #else
         return -1; // unavailable

src/kernel/lib/realv.hpp

@@ -218,9 +218,9 @@ namespace yask {
         // copy whole vector.
         ALWAYS_INLINE real_vec_t& operator=(const real_vec_t& rhs) {
             #ifdef NO_INTRINSICS
-            REAL_VEC_LOOP(i) u.r[i] = rhs[i];
+            REAL_VEC_LOOP(i) u.ci[i] = rhs.u.ci[i];
             #else
-            u.mr = rhs.u.mr;
+            u.mi = rhs.u.mi;
             #endif
             return *this;
         }
@@ -233,6 +233,10 @@ namespace yask {
             u.mr = rhs;
             return *this;
         }
+        ALWAYS_INLINE real_vec_t& operator=(const isimd_t& rhs) {
+            u.mi = rhs;
+            return *this;
+        }
         #endif
 
         // assignment: single value broadcast.
@@ -360,38 +364,37 @@ namespace yask {
         }
 
         // less-than comparator.
+        // uses int comparison for reliable results, even w/NaNs.
         bool operator<(const real_vec_t& rhs) const {
             for (int j = 0; j < VLEN; j++) {
-                if (u.r[j] < rhs.u.r[j])
+                if (u.ci[j] < rhs.u.ci[j])
                     return true;
-                else if (u.r[j] > rhs.u.r[j])
+                else if (u.ci[j] > rhs.u.ci[j])
                     return false;
             }
             return false;
         }
 
         // greater-than comparator.
+        // uses int comparison for reliable results, even w/NaNs.
         bool operator>(const real_vec_t& rhs) const {
             for (int j = 0; j < VLEN; j++) {
-                if (u.r[j] > rhs.u.r[j])
+                if (u.ci[j] > rhs.u.ci[j])
                     return true;
-                else if (u.r[j] < rhs.u.r[j])
+                else if (u.ci[j] < rhs.u.ci[j])
                     return false;
             }
             return false;
         }
 
         // equal-to comparator.
+        // uses int comparison for reliable results, even w/NaNs.
         ALWAYS_INLINE bool operator==(const real_vec_t& rhs) const {
-            #ifdef NO_SIMD_COMPARE
             for (int j = 0; j < VLEN; j++) {
-                if (u.r[j] != rhs.u.r[j])
+                if (u.ci[j] != rhs.u.ci[j])
                     return false;
             }
             return true;
-            #else
-            return u.r == rhs.u.r;
-            #endif
         }
 
         // not-equal-to comparator.