diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index b45f29a12bb3..e544a23210a4 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,10 @@
+2004-04-23  Daniel Jacobowitz  <drow@mvista.com>
+
+	* config/arm/arm.c (arm_output_epilogue): Reverse the order of
+	loading iWMMXt registers with a frame pointer.  Use post-increment
+	without a frame pointer.
+	(arm_expand_prologue): Reverse the order of saving iWMMXt registers.
+
 2004-04-23  Paolo Bonzini  <bonzini@gnu.org>
 
 	* doc/invoke.texi (Optimize Options): Refer to "unit-at-a-time
diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
index d6c0c3d0421a..1c2f2b2b4949 100644
--- a/gcc/config/arm/arm.c
+++ b/gcc/config/arm/arm.c
@@ -9492,7 +9492,7 @@ arm_output_epilogue (rtx sibling)
 	     the live_regs_mask.  */
 	  lrm_count += (lrm_count % 2 ? 2 : 1);
 	      
-	  for (reg = FIRST_IWMMXT_REGNUM; reg <= LAST_IWMMXT_REGNUM; reg++)
+	  for (reg = LAST_IWMMXT_REGNUM; reg >= FIRST_IWMMXT_REGNUM; reg--)
 	    if (regs_ever_live[reg] && !call_used_regs[reg])
 	      {
 		asm_fprintf (f, "\twldrd\t%r, [%r, #-%d]\n", 
@@ -9613,7 +9613,7 @@ arm_output_epilogue (rtx sibling)
       if (TARGET_IWMMXT)
 	for (reg = FIRST_IWMMXT_REGNUM; reg <= LAST_IWMMXT_REGNUM; reg++)
 	  if (regs_ever_live[reg] && !call_used_regs[reg])
-	    asm_fprintf (f, "\twldrd\t%r, [%r, #+8]!\n", reg, SP_REGNUM);
+	    asm_fprintf (f, "\twldrd\t%r, [%r], #8\n", reg, SP_REGNUM);
 
       /* If we can, restore the LR into the PC.  */
       if (ARM_FUNC_TYPE (func_type) == ARM_FT_NORMAL
@@ -10339,7 +10339,7 @@ arm_expand_prologue (void)
     }
 
   if (TARGET_IWMMXT)
-    for (reg = FIRST_IWMMXT_REGNUM; reg <= LAST_IWMMXT_REGNUM; reg++)
+    for (reg = LAST_IWMMXT_REGNUM; reg >= FIRST_IWMMXT_REGNUM; reg--)
       if (regs_ever_live[reg] && ! call_used_regs [reg])
 	{
 	  insn = gen_rtx_PRE_DEC (V2SImode, stack_pointer_rtx);
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index b57694e76f40..f25bfd1b0c76 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,7 @@
+2004-04-23  Daniel Jacobowitz  <drow@mvista.com>
+
+	* gcc.c-torture/execute/simd-5.c: New test.
+
 2004-04-23  Giovanni Bajo  <giovannibajo@gcc.gnu.org>
 
 	PR c++/15064
diff --git a/gcc/testsuite/gcc.c-torture/execute/simd-5.c b/gcc/testsuite/gcc.c-torture/execute/simd-5.c
new file mode 100644
index 000000000000..f0584328716b
--- /dev/null
+++ b/gcc/testsuite/gcc.c-torture/execute/simd-5.c
@@ -0,0 +1,59 @@
+/* Test saving and restoring of SIMD registers.  */
+
+typedef short Q __attribute__((vector_size(8)));
+
+Q q1 = {1, 2}, q2 = {3, 4}, q3 = {5, 6}, q4 = {7, 8};
+
+Q w1, w2, w3, w4;
+Q z1, z2, z3, z4;
+
+volatile int dummy;
+
+void  __attribute__((__noinline__))
+func0 (void)
+{
+  dummy = 1;
+}
+
+void __attribute__((__noinline__))
+func1 (void)
+{
+  Q a, b;
+  a = q1 * q2;
+  b = q3 * q4;
+  w1 = a;
+  w2 = b;
+  func0 ();
+  w3 = a;
+  w4 = b;
+}
+
+void __attribute__((__noinline__))
+func2 (void)
+{
+  Q a, b;
+  a = q1 + q2;
+  b = q3 - q4;
+  z1 = a;
+  z2 = b;
+  func1 ();
+  z3 = a;
+  z4 = b;
+}
+
+int
+main (void)
+{
+  func2 ();
+
+  if (memcmp (&w1, &w3, sizeof (Q)) != 0)
+    abort ();
+  if (memcmp (&w2, &w4, sizeof (Q)) != 0)
+    abort ();
+  if (memcmp (&z1, &z3, sizeof (Q)) != 0)
+    abort ();
+  if (memcmp (&z2, &z4, sizeof (Q)) != 0)
+    abort ();
+
+  return 0;
+}