[PATCH] The use of ARM NEON vshll_n_u8 intrinsic results in compile error on valid code (Bugzilla ID 41196)

[PATCH] The use of ARM NEON vshll_n_u8 intrinsic results in compile error on valid code (Bugzilla ID 41196)

[Daniel Gutson]

[-- Attachment #1: Type: text/plain, Size: 374 bytes --]

Hi,
   the attached patch attempts to fix bug 41196 (the attachment here is 
the same as the attachment in bugzilla).

(http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41196)

Please refer to comment #3 for the description of the fix and comment #4 
for the change log.

Please let me know if OK to commit.

Thanks,
	Daniel.

-- 
Daniel Gutson
CodeSourcery
www.codesourcery.com

[-- Attachment #2: neon.patch --]
[-- Type: text/x-diff, Size: 1888 bytes --]

Index: src/gcc-mainline/gcc/testsuite/gcc.target/arm/neon/vfp-shift-a2t2.c
===================================================================
--- src/gcc-mainline/gcc/testsuite/gcc.target/arm/neon/vfp-shift-a2t2.c	(revision 0)
+++ src/gcc-mainline/gcc/testsuite/gcc.target/arm/neon/vfp-shift-a2t2.c	(revision 0)
@@ -0,0 +1,27 @@
+/* Check that NEON vector shifts support immediate values == size.  /*
+
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_neon_ok } */
+/* { dg-options "-save-temps -mfpu=neon -mfloat-abi=softfp" } */
+
+#include <arm_neon.h>
+
+uint16x8_t test_vshll_n_u8 (uint8x8_t a)
+{
+    return vshll_n_u8(a, 8);
+}
+
+uint32x4_t test_vshll_n_u16 (uint16x4_t a)
+{   
+    return vshll_n_u16(a, 16);
+}
+
+uint64x2_t test_vshll_n_u32 (uint32x2_t a)
+{
+    return vshll_n_u32(a, 32);
+}
+
+/* { dg-final { scan-assembler "vshll\.u16\[ 	\]+\[qQ\]\[0-9\]+, \[dD\]\[0-9\]+, #\[0-9\]+!?\(\[ 	\]+@\[a-zA-Z0-9 \]+\)?\n" } } */
+/* { dg-final { scan-assembler "vshll\.u32\[ 	\]+\[qQ\]\[0-9\]+, \[dD\]\[0-9\]+, #\[0-9\]+!?\(\[ 	\]+@\[a-zA-Z0-9 \]+\)?\n" } } */
+/* { dg-final { scan-assembler "vshll\.u8\[ 	\]+\[qQ\]\[0-9\]+, \[dD\]\[0-9\]+, #\[0-9\]+!?\(\[ 	\]+@\[a-zA-Z0-9 \]+\)?\n" } } */
+/* { dg-final { cleanup-saved-temps } } */
Index: src/gcc-mainline/gcc/config/arm/neon.md
===================================================================
--- src/gcc-mainline/gcc/config/arm/neon.md	(revision 152180)
+++ src/gcc-mainline/gcc/config/arm/neon.md	(working copy)
@@ -3655,7 +3655,8 @@ (define_insn "neon_vshll_n<mode>"
 			  UNSPEC_VSHLL_N))]
   "TARGET_NEON"
 {
-  neon_const_bounds (operands[2], 0, neon_element_bits (<MODE>mode));
+  /* The boundaries are: 0 < imm <= size.  */
+  neon_const_bounds (operands[2], 0, neon_element_bits (<MODE>mode) + 1);
   return "vshll.%T3%#<V_sz_elem>\t%q0, %P1, %2";
 }
   [(set_attr "neon_type" "neon_shift_1")]

Re: [PATCH] The use of ARM NEON vshll_n_u8 intrinsic results in compile error on valid code (Bugzilla ID 41196)

[Richard Earnshaw]

On Fri, 2009-10-09 at 16:02 -0300, Daniel Gutson wrote:
> Hi,
>    the attached patch attempts to fix bug 41196 (the attachment here is 
> the same as the attachment in bugzilla).
> 
> (http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41196)
> 
> Please refer to comment #3 for the description of the fix and comment #4 
> for the change log.
> 
> Please let me know if OK to commit.
> 

OK.

R.