Merge pull request #80 from xianyi/develop

rebase

Author: martin-frbg

Makefile.zarch

@@ -8,3 +8,9 @@ ifeq ($(CORE), Z14)
 CCOMMON_OPT += -march=z14 -mzvector -O3
 FCOMMON_OPT += -march=z14 -mzvector
 endif
+
+# Enable floating-point expression contraction for clang, since it is the
+# default for gcc
+ifeq ($(C_COMPILER), CLANG)
+CCOMMON_OPT += -ffp-contract=fast
+endif

c_check

@@ -8,7 +8,7 @@ $hostos   = `uname -s | sed -e s/\-.*//`;    chop($hostos);
 $hostarch = `uname -m | sed -e s/i.86/x86/`;chop($hostarch);
 $hostarch = `uname -p` if ($hostos eq "AIX");
 $hostarch = "x86_64" if ($hostarch eq "amd64");
-$hostarch = "arm" if ($hostarch =~ /^arm.*/);
+$hostarch = "arm" if ($hostarch ne "arm64" && $hostarch =~ /^arm.*/);
 $hostarch = "arm64" if ($hostarch eq "aarch64");
 $hostarch = "power" if ($hostarch =~ /^(powerpc|ppc).*/);
 $hostarch = "zarch" if ($hostarch eq "s390x");

cmake/system.cmake

@@ -110,6 +110,11 @@ if (NO_AVX2)
   set(GETARCH_FLAGS "${GETARCH_FLAGS} -DNO_AVX2")
 endif ()
 
+if (NO_AVX512)
+  message(STATUS "Disabling Advanced Vector Extensions 512 (AVX512).")
+  set(GETARCH_FLAGS "${GETARCH_FLAGS} -DNO_AVX512")
+endif ()
+
 if (CMAKE_BUILD_TYPE STREQUAL "Debug")
   set(GETARCH_FLAGS "${GETARCH_FLAGS} ${CMAKE_C_FLAGS_DEBUG}")
 endif ()

cmake/system_check.cmake

@@ -109,13 +109,15 @@ else()
 endif()
 
 if (X86_64 OR X86)
+if (NOT NO_AVX512)
   file(WRITE ${PROJECT_BINARY_DIR}/avx512.c "#include <immintrin.h>\n\nint main(void){ __asm__ volatile(\"vbroadcastss -4 * 4(%rsi), %zmm2\"); }")
 execute_process(COMMAND ${CMAKE_C_COMPILER} -march=skylake-avx512 -c -v -o ${PROJECT_BINARY_DIR}/avx512.o ${PROJECT_BINARY_DIR}/avx512.c OUTPUT_QUIET ERROR_QUIET RESULT_VARIABLE NO_AVX512)
 if (NO_AVX512 EQUAL 1)
 set (CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX512")
 endif()
   file(REMOVE "avx512.c" "avx512.o")
 endif()
+endif()
 
 include(CheckIncludeFile)
 CHECK_INCLUDE_FILE("stdatomic.h" HAVE_C11)

kernel/zarch/camax.c

@@ -136,7 +136,7 @@ static FLOAT camax_kernel_32(BLASLONG n, FLOAT *x) {
     "wfmaxsb %%v0,%%v0,%%v16,0\n\t"
     "ler    %[amax],%%f0"
     : [amax] "=f"(amax),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v16", "v17", "v18", "v19", "v20",
        "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");

kernel/zarch/camin.c

@@ -136,7 +136,7 @@ static FLOAT camin_kernel_32(BLASLONG n, FLOAT *x) {
     "wfminsb %%v0,%%v0,%%v16,0\n\t"
     "ler    %[amin],%%f0"
     : [amin] "=f"(amin),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v16", "v17", "v18", "v19", "v20",
        "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");

kernel/zarch/casum.c

@@ -108,7 +108,7 @@ static FLOAT casum_kernel_32(BLASLONG n, FLOAT *x) {
     "vfasb   %%v24,%%v24,%%v25\n\t"
     "vstef   %%v24,%[asum],0"
     : [asum] "=Q"(asum),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x)
     : "cc", "r1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
        "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/caxpy.c

@@ -99,9 +99,9 @@ static void caxpy_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha) {
     "vst %%v19,112(%%r1,%[y])\n\t"
     "agfi  %%r1,128\n\t"
     "brctg %[n],0b"
-    : "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
+    : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v8", "v9", "v10", "v11", "v12", "v13",
        "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
        "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");

kernel/zarch/ccopy.c

@@ -36,9 +36,9 @@ static void ccopy_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y) {
     "la  %[x],256(%[x])\n\t"
     "la  %[y],256(%[y])\n\t"
     "brctg %[n],0b"
-    : "=m"(*(struct { FLOAT x[n * 2]; } *) y),[x] "+&a"(x),[y] "+&a"(y),
+    : "=m"(*(FLOAT (*)[n * 2]) y),[x] "+&a"(x),[y] "+&a"(y),
        [n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n * 2]; } *) x)
+    : "m"(*(const FLOAT (*)[n * 2]) x)
     : "cc");
 }
 

kernel/zarch/cdot.c

@@ -97,9 +97,9 @@ static void cdot_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *d) {
     "vstef  %%v24,4(%[d]),1\n\t"
     "vstef  %%v25,8(%[d]),1\n\t"
     "vstef  %%v25,12(%[d]),0"
-    : "=m"(*(struct { FLOAT x[4]; } *) d),[n] "+&r"(n)
-    : [d] "a"(d), "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) y),[y] "a"(y)
+    : "=m"(*(FLOAT (*)[4]) d),[n] "+&r"(n)
+    : [d] "a"(d), "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[n * 2]) y),[y] "a"(y)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v16", "v17", "v18", "v19", "v20",
        "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");