7 files changed, 387 insertions, 0 deletions
diff --git a/gcc/testsuite/gcc.dg/dfp/pr39902.c b/gcc/testsuite/gcc.dg/dfp/pr39902.c
new file mode 100644
index 00000000000..8c4e7b24380
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/dfp/pr39902.c
@@ -0,0 +1,239 @@
+/* { dg-options "--std=gnu99" } */
+
+/* Check that optimizations like (x * 1) to x, or (x * -1) to -x,
+   do not apply to decimal float computations where trailing zeroes
+   are significant.  */
+
+extern void abort (void);
+int failcnt;
+
+#ifdef DBG
+extern int printf (const char *, ...);
+#define FAILURE { printf ("failed at line %d\n", __LINE__); failcnt++; }
+#else
+#define FAILURE abort ();
+#endif
+
+#define COMPARE32(A,B) \
+  A.i == B.i
+
+#define COMPARE64(A,B) \
+  A.i[0] == B.i[0] && A.i[1] == B.i[1]
+
+#define COMPARE128(A,B) \
+  A.i[0] == B.i[0] && A.i[1] == B.i[1] && A.i[2] == B.i[2] && A.i[3] == B.i[3]
+
+typedef union {
+  _Decimal32 d;
+  unsigned int i;
+} u32;
+
+typedef union {
+  _Decimal64 d;
+  unsigned int i[2];
+} u64;
+
+typedef union {
+  _Decimal128 d;
+  unsigned int i[4];
+} u128;
+
+volatile u32 p32_1;
+volatile u32 p32_1_0;
+volatile u32 p32_2_0;
+volatile u32 m32_1;
+volatile u32 m32_1_0;
+volatile u32 m32_2_0;
+volatile u32 a32;
+
+volatile u64 p64_1;
+volatile u64 p64_1_0;
+volatile u64 p64_2_0;
+volatile u64 m64_1;
+volatile u64 m64_1_0;
+volatile u64 m64_2_0;
+volatile u64 a64;
+
+volatile u128 p128_1;
+volatile u128 p128_1_0;
+volatile u128 p128_2_0;
+volatile u128 m128_1;
+volatile u128 m128_1_0;
+volatile u128 m128_2_0;
+volatile u128 a128;
+
+void
+init32 (void)
+{
+  p32_1.d = 1.DF;
+  p32_1_0.d = 1.0DF;
+  p32_2_0.d = 2.0DF;
+  m32_1.d = -1.DF;
+  m32_1_0.d = -1.0DF;
+  m32_2_0.d = -2.0DF;
+}
+
+void
+init64 (void)
+{
+  p64_1.d = 1.DD;
+  p64_1_0.d = 1.0DD;
+  p64_2_0.d = 2.0DD;
+  m64_1.d = -1.DD;
+  m64_1_0.d = -1.0DD;
+  m64_2_0.d = -2.0DD;
+}
+
+void
+init128 (void)
+{
+  p128_1.d = 1.DL;
+  p128_1_0.d = 1.0DL;
+  p128_2_0.d = 2.0DL;
+  m128_1.d = -1.DL;
+  m128_1_0.d = -1.0DL;
+  m128_2_0.d = -2.0DL;
+}
+
+void
+doit32 (void)
+{
+  /* Multiplying by a value with no trailing zero should not change the
+     quantum exponent.  */
+
+  a32.d = p32_2_0.d * p32_1.d;
+  if (! (COMPARE32 (a32, p32_2_0)))
+    FAILURE
+
+  a32.d = p32_2_0.d * 1.DF;
+  if (! (COMPARE32 (a32, p32_2_0)))
+    FAILURE
+
+  a32.d = p32_2_0.d * m32_1.d;
+  if (! (COMPARE32 (a32, m32_2_0)))
+    FAILURE
+
+  a32.d = p32_2_0.d * -1.DF;
+  if (! (COMPARE32 (a32, m32_2_0)))
+    FAILURE
+
+  /* Multiplying by a value with a trailing zero should change the
+     quantum exponent.  */
+
+  a32.d = p32_2_0.d * p32_1_0.d;
+  if (COMPARE32 (a32, p32_2_0))
+    FAILURE
+
+  a32.d = p32_2_0.d * 1.0DF;
+  if (COMPARE32 (a32, p32_2_0))
+    FAILURE
+
+  a32.d = p32_2_0.d * m32_1_0.d;
+  if (COMPARE32 (a32, m32_2_0))
+    FAILURE
+
+  a32.d = p32_2_0.d * -1.0DF;
+  if (COMPARE32 (a32, m32_2_0))
+    FAILURE
+}
+
+void
+doit64 (void)
+{
+  /* Multiplying by a value with no trailing zero should not change the
+     quantum exponent.  */
+
+  a64.d = p64_2_0.d * p64_1.d;
+  if (! (COMPARE64 (a64, p64_2_0)))
+    FAILURE
+
+  a64.d = p64_2_0.d * 1.DD;
+  if (! (COMPARE64 (a64, p64_2_0)))
+    FAILURE
+
+  a64.d = p64_2_0.d * m64_1.d;
+  if (! (COMPARE64 (a64, m64_2_0)))
+    FAILURE
+
+  a64.d = p64_2_0.d * -1.DD;
+  if (! (COMPARE64 (a64, m64_2_0)))
+    FAILURE
+
+  /* Multiplying by a value with a trailing zero should change the
+     quantum exponent.  */
+
+  a64.d = p64_2_0.d * p64_1_0.d;
+  if (COMPARE64 (a64, p64_2_0))
+    FAILURE
+
+  a64.d = p64_2_0.d * 1.0DD;
+  if (COMPARE64 (a64, p64_2_0))
+    FAILURE
+
+  a64.d = p64_2_0.d * m64_1_0.d;
+  if (COMPARE64 (a64, m64_2_0))
+    FAILURE
+
+  a64.d = p64_2_0.d * -1.0DD;
+  if (COMPARE64 (a64, m64_2_0))
+    FAILURE
+}
+
+void
+doit128 (void)
+{
+  /* Multiplying by a value with no trailing zero should not change the
+     quantum exponent.  */
+
+  a128.d = p128_2_0.d * p128_1_0.d;
+  if (COMPARE128 (a128, p128_2_0))
+    FAILURE
+
+  a128.d = p128_2_0.d * 1.0DD;
+  if (COMPARE128 (a128, p128_2_0))
+    FAILURE
+
+  a128.d = p128_2_0.d * m128_1_0.d;
+  if (COMPARE128 (a128, m128_2_0))
+    FAILURE
+
+  a128.d = p128_2_0.d * -1.0DD;
+  if (COMPARE128 (a128, m128_2_0))
+    FAILURE
+
+  /* Multiplying by a value with a trailing zero should change the
+     quantum exponent.  */
+
+  a128.d = p128_2_0.d * p128_1.d;
+  if (! (COMPARE128 (a128, p128_2_0)))
+    FAILURE
+
+  a128.d = p128_2_0.d * 1.DD;
+  if (! (COMPARE128 (a128, p128_2_0)))
+    FAILURE
+
+  a128.d = p128_2_0.d * m128_1.d;
+  if (! (COMPARE128 (a128, m128_2_0)))
+    FAILURE
+
+  a128.d = p128_2_0.d * -1.DD;
+  if (! (COMPARE128 (a128, m128_2_0)))
+    FAILURE
+}
+
+int
+main (void)
+{
+  init32 ();
+  init64 ();
+  init128 ();
+
+  doit32 ();
+  doit64 ();
+  doit128 ();
+
+  if (failcnt != 0)
+    abort ();
+
+  return 0;
+}
diff --git a/gcc/testsuite/gcc.dg/pr40550.c b/gcc/testsuite/gcc.dg/pr40550.c
new file mode 100644
index 00000000000..27935ab005a
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/pr40550.c
@@ -0,0 +1,34 @@
+/* { dg-do run } */
+/* { dg-options "-msse" { target { { i?86-*-* x86_64-*-* } && ilp32 } } } */
+
+#ifdef __i386__
+#include "cpuid.h"
+#endif
+
+typedef float v2sf __attribute__ ((vector_size (2 * sizeof(float))));
+
+static void test (void)
+{
+  v2sf a = {1.0, 0.0};
+  v2sf b = {0.0, 1.0};
+  v2sf d;
+  d = a + b;
+}
+
+int main ()
+{
+
+#ifdef __i386__
+  unsigned int eax, ebx, ecx, edx;
+
+  if (!__get_cpuid (1, &eax, &ebx, &ecx, &edx))
+    return 0;
+
+  if (!(edx & bit_SSE))
+    return 0;
+#endif
+
+  test ();
+
+  return 0;
+}
diff --git a/gcc/testsuite/gcc.dg/torture/pr40470-1.c b/gcc/testsuite/gcc.dg/torture/pr40470-1.c
new file mode 100644
index 00000000000..cb2510e883b
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/torture/pr40470-1.c
@@ -0,0 +1,22 @@
+/* { dg-do compile { target i?86-*-* x86_64-*-* } } */
+/* { dg-options "-msse4" } */
+#include <nmmintrin.h>
+__m128i load (char *);
+char *
+foo (char *p1, char *p2,
+     int bmsk, __m128i mask1, __m128i mask2)
+{
+  int len = 0;
+  __m128i frag1, frag2;
+  int  cmp_s;
+  if( !p2[0]) return p1;
+  if( !p1[0] ) return NULL;
+  frag2 = load (p2); 
+  frag1 = load (p1);
+  frag2 = _mm_blendv_epi8(frag2, mask2, mask1);
+  frag1 = _mm_blendv_epi8(frag1, mask2, mask1);
+  cmp_s = _mm_cmpistrs(frag2, frag1, 0x0c);
+  if( cmp_s )
+    __asm("bsfl %[bmsk], %[len]" : [len] "=r" (len) : [bmsk] "r" (bmsk) );
+  return p1 + len;
+}
diff --git a/gcc/testsuite/gcc.dg/torture/pr40470-2.c b/gcc/testsuite/gcc.dg/torture/pr40470-2.c
new file mode 100644
index 00000000000..b0414bbea91
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/torture/pr40470-2.c
@@ -0,0 +1,27 @@
+/* { dg-do compile { target i?86-*-* x86_64-*-* } } */
+/* { dg-options "-msse4" } */
+#include <nmmintrin.h>
+__m128i load (char *);
+char *
+foo (const unsigned char *s1, const unsigned char *s2,
+     int bmsk, __m128i frag2)
+{
+  int len = 0;
+  char *p1 = (char  *) s1;
+  char *p2 = (char  *) s2;
+  __m128i frag1, fruc, mask;
+  int cmp_c, cmp_s;
+  if( !p2[0]) return (char *) s1;
+  if( !p1[0] ) return NULL;
+  if( p2[1]) frag2 = load (p2); 
+  frag1 = load (p1);
+  fruc = _mm_loadu_si128 ((__m128i *) s1);
+  mask = _mm_cmpistrm(fruc, frag2, 0x44);
+  frag2 = _mm_blendv_epi8(frag2, mask, mask);
+  frag1 = _mm_blendv_epi8(frag1, mask, mask);
+  cmp_c = _mm_cmpistrc(frag2, frag1, 0x0c);
+  cmp_s = _mm_cmpistrs(frag2, frag1, 0x0c);
+  if( cmp_s  & cmp_c  )
+    __asm("bsfl %[bmsk], %[len]" : [len] "=r" (len) : [bmsk] "r" (bmsk) );
+  return p2 + len;
+}
diff --git a/gcc/testsuite/gcc.dg/torture/pr40470-3.c b/gcc/testsuite/gcc.dg/torture/pr40470-3.c
new file mode 100644
index 00000000000..5c4c361de55
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/torture/pr40470-3.c
@@ -0,0 +1,22 @@
+/* { dg-do compile { target i?86-*-* x86_64-*-* } } */
+/* { dg-options "-msse4" } */
+#include <nmmintrin.h>
+__m128i load (char *);
+char *
+foo (char *p1, char *p2,
+     int bmsk, __m128i mask1, __m128i mask2)
+{
+  int len = 0;
+  __m128i frag1, frag2;
+  int  cmp_s;
+  if( !p2[0]) return p1;
+  if( !p1[0] ) return NULL;
+  frag2 = load (p2); 
+  frag1 = load (p1);
+  frag2 = _mm_blendv_epi8(frag2, mask2, mask1);
+  frag1 = _mm_blendv_epi8(frag1, mask1, mask2);
+  cmp_s = _mm_cmpistrs(frag2, frag1, 0x0c);
+  if( cmp_s )
+    __asm("bsfl %[bmsk], %[len]" : [len] "=r" (len) : [bmsk] "r" (bmsk) );
+  return p1 + len;
+}
diff --git a/gcc/testsuite/gcc.dg/torture/pr40470-4.c b/gcc/testsuite/gcc.dg/torture/pr40470-4.c
new file mode 100644
index 00000000000..f9d0e509500
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/torture/pr40470-4.c
@@ -0,0 +1,29 @@
+/* { dg-do compile { target i?86-*-* x86_64-*-* } } */
+/* { dg-options "-msse4" } */
+#include <nmmintrin.h>
+__m128i load (char *);
+char *
+foo (const unsigned char *s1, const unsigned char *s2,
+     int bmsk, __m128i frag2)
+{
+  int len = 0;
+  char *p1 = (char  *) s1;
+  char *p2 = (char  *) s2;
+  __m128i frag1, fruc1, fruc2, mask1, mask2;
+  int cmp_c, cmp_s;
+  if( !p2[0]) return (char *) s1;
+  if( !p1[0] ) return NULL;
+  if( p2[1]) frag2 = load (p2); 
+  frag1 = load (p1);
+  fruc1 = _mm_loadu_si128 ((__m128i *) s1);
+  fruc2 = _mm_loadu_si128 ((__m128i *) s2);
+  mask1 = _mm_cmpistrm(fruc1, frag2, 0x44);
+  mask2 = _mm_cmpistrm(fruc2, frag1, 0x14);
+  frag2 = _mm_blendv_epi8(frag2, mask1, mask2);
+  frag1 = _mm_blendv_epi8(frag1, mask2, mask1);
+  cmp_c = _mm_cmpistrc(frag2, frag1, 0x0c);
+  cmp_s = _mm_cmpistrs(frag2, frag1, 0x0c);
+  if( cmp_s  & cmp_c  )
+    __asm("bsfl %[bmsk], %[len]" : [len] "=r" (len) : [bmsk] "r" (bmsk) );
+  return p2 + len;
+}
diff --git a/gcc/testsuite/gcc.dg/vect/pr40542.c b/gcc/testsuite/gcc.dg/vect/pr40542.c
new file mode 100644
index 00000000000..0a827724dc8
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/pr40542.c
@@ -0,0 +1,14 @@
+/* { dg-do compile } */
+
+void
+volarr_cpy(char *d, volatile char *s)
+{
+  int i;
+  
+  for (i = 0; i < 16; i++)
+    d[i] = s[i];
+}
+
+/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { xfail *-*-* } } } */
+/* { dg-final { cleanup-tree-dump "vect" } } */
+