re PR target/35767 (x86 backend uses aligned load on unaligned memory)

gcc/

2008-05-27  H.J. Lu  <hongjiu.lu@intel.com>

	PR target/35767
	PR target/35771
	* config/i386/i386.c (ix86_function_arg_boundary): Use
	alignment of canonical type.
	(ix86_expand_vector_move): Check unaligned memory access for
	all SSE modes.

gcc/testsuite/

2008-05-27  H.J. Lu  <hongjiu.lu@intel.com>

	PR target/35767
	PR target/35771
	* gcc.target/i386/pr35767-1.c: New.
	* gcc.target/i386/pr35767-1d.c: Likewise.
	* gcc.target/i386/pr35767-1i.c: Likewise.
	* gcc.target/i386/pr35767-2.c: Likewise.
	* gcc.target/i386/pr35767-2d.c: Likewise.
	* gcc.target/i386/pr35767-2i.c: Likewise.
	* gcc.target/i386/pr35767-3.c: Likewise.
	* gcc.target/i386/pr35767-4.c: Likewise.
	* gcc.target/i386/pr35767-5.c: Likewise.

From-SVN: r136054

12 files changed, 239 insertions, 6 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 92c82d0..c25c438 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,5 +1,14 @@
 2008-05-27  H.J. Lu  <hongjiu.lu@intel.com>
 
+	PR target/35767
+	PR target/35771
+	* config/i386/i386.c (ix86_function_arg_boundary): Use
+	alignment of canonical type.
+	(ix86_expand_vector_move): Check unaligned memory access for
+	all SSE modes.
+
+2008-05-27  H.J. Lu  <hongjiu.lu@intel.com>
+
 	* dwarf2out.c (current_fde): Change return type to dw_fde_ref.
 	Moved to the front of file.
 
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index 6cac18a..cbb122a 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -4638,7 +4638,12 @@ ix86_function_arg_boundary (enum machine_mode mode, tree type)
 {
   int align;
   if (type)
-    align = TYPE_ALIGN (type);
+    {
+      if (TYPE_STRUCTURAL_EQUALITY_P (type))
+	align = TYPE_ALIGN (type);
+      else
+	align = TYPE_ALIGN (TYPE_CANONICAL (type));
+    }
   else
     align = GET_MODE_ALIGNMENT (mode);
   if (align < PARM_BOUNDARY)
@@ -10408,12 +10413,10 @@ ix86_expand_vector_move (enum machine_mode mode, rtx operands[])
       && standard_sse_constant_p (op1) <= 0)
     op1 = validize_mem (force_const_mem (mode, op1));
 
-  /* TDmode values are passed as TImode on the stack.  TImode values
-     are moved via xmm registers, and moving them to stack can result in
-     unaligned memory access.  Use ix86_expand_vector_move_misalign()
-     if memory operand is not aligned correctly.  */
+  /* We need to check memory alignment for SSE mode since attribute
+     can make operands unaligned.  */
   if (can_create_pseudo_p ()
-      && (mode == TImode) && !TARGET_64BIT
+      && SSE_REG_MODE_P (mode)
       && ((MEM_P (op0) && (MEM_ALIGN (op0) < align))
 	  || (MEM_P (op1) && (MEM_ALIGN (op1) < align))))
     {
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 4983df2..59bf3a9 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,17 @@
+2008-05-27  H.J. Lu  <hongjiu.lu@intel.com>
+
+	PR target/35767
+	PR target/35771
+	* gcc.target/i386/pr35767-1.c: New.
+	* gcc.target/i386/pr35767-1d.c: Likewise.
+	* gcc.target/i386/pr35767-1i.c: Likewise.
+	* gcc.target/i386/pr35767-2.c: Likewise.
+	* gcc.target/i386/pr35767-2d.c: Likewise.
+	* gcc.target/i386/pr35767-2i.c: Likewise.
+	* gcc.target/i386/pr35767-3.c: Likewise.
+	* gcc.target/i386/pr35767-4.c: Likewise.
+	* gcc.target/i386/pr35767-5.c: Likewise.
+
 2008-05-27  Tobias Burnus  <burnus@net-b.de>
 
 	* gfortran.dg/assignment_3.f90: Add missing cleanup-modules.
diff --git a/gcc/testsuite/gcc.target/i386/pr35767-1.c b/gcc/testsuite/gcc.target/i386/pr35767-1.c
new file mode 100644
index 0000000..5ed5b85
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr35767-1.c
@@ -0,0 +1,25 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -msse2" } */
+
+#include "sse2-check.h"
+
+typedef struct { __m128 f __attribute__((packed)); } packed;
+
+__m128  __attribute__((noinline))
+foo (__m128 a1, __m128 a2, __m128 a3, __m128 a4,
+     __m128 a5, __m128 a6, __m128 a7, __m128 a8,
+     int b1, int b2, int b3, int b4, int b5, int b6, int b7, packed y)
+{
+  return y.f;
+}
+
+void
+sse2_test (void)
+{
+  packed x;
+  __m128 y = { 0 };
+  x.f = y; 
+  y = foo (y, y, y, y, y, y, y, y, 1, 2, 3, 4, 5, 6, -1, x);
+  if (__builtin_memcmp (&y, &x.f, sizeof (y)) != 0)
+    abort ();
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr35767-1d.c b/gcc/testsuite/gcc.target/i386/pr35767-1d.c
new file mode 100644
index 0000000..cdf17fa
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr35767-1d.c
@@ -0,0 +1,25 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -msse2" } */
+
+#include "sse2-check.h"
+
+typedef struct { __m128d f __attribute__((packed)); } packed;
+
+__m128d  __attribute__((noinline))
+foo (__m128d a1, __m128d a2, __m128d a3, __m128d a4,
+     __m128d a5, __m128d a6, __m128d a7, __m128d a8,
+     int b1, int b2, int b3, int b4, int b5, int b6, int b7, packed y)
+{
+  return y.f;
+}
+
+void
+sse2_test (void)
+{
+  packed x;
+  __m128d y = { 0 };
+  x.f = y; 
+  y = foo (y, y, y, y, y, y, y, y, 1, 2, 3, 4, 5, 6, -1, x);
+  if (__builtin_memcmp (&y, &x.f, sizeof (y)) != 0)
+    abort ();
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr35767-1i.c b/gcc/testsuite/gcc.target/i386/pr35767-1i.c
new file mode 100644
index 0000000..188e8e7
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr35767-1i.c
@@ -0,0 +1,25 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -msse2" } */
+
+#include "sse2-check.h"
+
+typedef struct { __m128i f __attribute__((packed)); } packed;
+
+__m128i  __attribute__((noinline))
+foo (__m128i a1, __m128i a2, __m128i a3, __m128i a4,
+     __m128i a5, __m128i a6, __m128i a7, __m128i a8,
+     int b1, int b2, int b3, int b4, int b5, int b6, int b7, packed y)
+{
+  return y.f;
+}
+
+void
+sse2_test (void)
+{
+  packed x;
+  __m128i y = { 0 };
+  x.f = y; 
+  y = foo (y, y, y, y, y, y, y, y, 1, 2, 3, 4, 5, 6, -1, x);
+  if (__builtin_memcmp (&y, &x.f, sizeof (y)) != 0)
+    abort ();
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr35767-2.c b/gcc/testsuite/gcc.target/i386/pr35767-2.c
new file mode 100644
index 0000000..82062ff
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr35767-2.c
@@ -0,0 +1,25 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -msse2" } */
+
+#include "sse2-check.h"
+
+typedef  __m128 __attribute__((aligned(1))) unaligned;
+
+__m128  __attribute__((noinline))
+foo (__m128 a1, __m128 a2, __m128 a3, __m128 a4,
+     __m128 a5, __m128 a6, __m128 a7, __m128 a8,
+     int b1, int b2, int b3, int b4, int b5, int b6, int b7, unaligned y)
+{
+  return y;
+}
+
+void
+sse2_test (void)
+{
+  unaligned x;
+  __m128 y = { 0 };
+  x = y; 
+  y = foo (y, y, y, y, y, y, y, y, 1, 2, 3, 4, 5, 6, -1, x);
+  if (__builtin_memcmp (&y, &x, sizeof (y)) != 0)
+    abort ();
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr35767-2d.c b/gcc/testsuite/gcc.target/i386/pr35767-2d.c
new file mode 100644
index 0000000..ae96cd8
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr35767-2d.c
@@ -0,0 +1,25 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -msse2" } */
+
+#include "sse2-check.h"
+
+typedef  __m128d __attribute__((aligned(1))) unaligned;
+
+__m128d  __attribute__((noinline))
+foo (__m128d a1, __m128d a2, __m128d a3, __m128d a4,
+     __m128d a5, __m128d a6, __m128d a7, __m128d a8,
+     int b1, int b2, int b3, int b4, int b5, int b6, int b7, unaligned y)
+{
+  return y;
+}
+
+void
+sse2_test (void)
+{
+  unaligned x;
+  __m128d y = { 0 };
+  x = y; 
+  y = foo (y, y, y, y, y, y, y, y, 1, 2, 3, 4, 5, 6, -1, x);
+  if (__builtin_memcmp (&y, &x, sizeof (y)) != 0)
+    abort ();
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr35767-2i.c b/gcc/testsuite/gcc.target/i386/pr35767-2i.c
new file mode 100644
index 0000000..d241644
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr35767-2i.c
@@ -0,0 +1,25 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -msse2" } */
+
+#include "sse2-check.h"
+
+typedef  __m128i __attribute__((aligned(1))) unaligned;
+
+__m128i  __attribute__((noinline))
+foo (__m128i a1, __m128i a2, __m128i a3, __m128i a4,
+     __m128i a5, __m128i a6, __m128i a7, __m128i a8,
+     int b1, int b2, int b3, int b4, int b5, int b6, int b7, unaligned y)
+{
+  return y;
+}
+
+void
+sse2_test (void)
+{
+  unaligned x;
+  __m128i y = { 0 };
+  x = y; 
+  y = foo (y, y, y, y, y, y, y, y, 1, 2, 3, 4, 5, 6, -1, x);
+  if (__builtin_memcmp (&y, &x, sizeof (y)) != 0)
+    abort ();
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr35767-3.c b/gcc/testsuite/gcc.target/i386/pr35767-3.c
new file mode 100644
index 0000000..e7592ff
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr35767-3.c
@@ -0,0 +1,26 @@
+/* { dg-do run } */
+/* { dg-require-effective-target dfp } */
+/* { dg-options "-O -msse2 -std=gnu99" } */
+
+#include "sse2-check.h"
+
+typedef _Decimal128 unaligned __attribute__((aligned(1)));
+
+_Decimal128 __attribute__((noinline))
+foo (_Decimal128 a1, _Decimal128 a2, _Decimal128 a3, _Decimal128 a4,
+     _Decimal128 a5, _Decimal128 a6, _Decimal128 a7, _Decimal128 a8,
+     int b1, int b2, int b3, int b4, int b5, int b6, int b7, unaligned y)
+{
+  return y;
+}
+
+void
+sse2_test (void)
+{
+  unaligned x;
+  _Decimal128 y = -1;
+  x = y;
+  y = foo (0, 0, 0, 0, 0, 0, 0, 0, 1, 2, 3, 4, 5, 6, -1, x);
+  if (__builtin_memcmp (&y, &x, sizeof (y)))
+    abort ();
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr35767-4.c b/gcc/testsuite/gcc.target/i386/pr35767-4.c
new file mode 100644
index 0000000..e12f64f
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr35767-4.c
@@ -0,0 +1,14 @@
+/* Test that we generate aligned load when memory is aligned.  */
+/* { dg-do compile } */
+/* { dg-require-effective-target dfp } */
+/* { dg-options "-O -march=x86-64 -mtune=generic -std=gnu99" } */
+/* { dg-final { scan-assembler-not "movdqu" } } */
+/* { dg-final { scan-assembler "movdqa" } } */
+
+extern _Decimal128 foo (_Decimal128, _Decimal128, _Decimal128);
+
+void
+bar (void)
+{
+  foo (0, 0, 0);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr35767-5.c b/gcc/testsuite/gcc.target/i386/pr35767-5.c
new file mode 100644
index 0000000..4372d2e
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr35767-5.c
@@ -0,0 +1,17 @@
+/* Test that we generate aligned load when memory is aligned.  */
+/* { dg-do compile } */
+/* { dg-options "-O -msse2 -mtune=generic" } */
+/* { dg-final { scan-assembler-not "movups" } } */
+/* { dg-final { scan-assembler "movaps" } } */
+
+typedef float v4sf __attribute__ ((__vector_size__ (16)));
+
+extern void foo(v4sf, v4sf, v4sf, v4sf, v4sf, v4sf, v4sf, v4sf, v4sf);
+
+int test(void)
+{
+  v4sf x = { 0.0, 1.0, 2.0, 3.0 };
+
+  foo (x, x, x, x, x, x, x, x, x);
+  return 0;
+}