[PATCH, middle-end/RTL, i386]: Simplify nested VEC_SELECT RTX

["Uros Bizjak" <ubizjak at gmail dot com>]

Hello!

This patch optimizes VEC_SELECT nested RTL expressions (with optional
VEC_CONCAT), when top VEC_SELECT extracts scalar operand.

Currently, following testcase

--cut here--
int fooSI_1(__v4si *val)
{
 return __builtin_ia32_vec_ext_v4si(*val, 1);
}

int fooSI_2(__v4si *val)
{
 return __builtin_ia32_vec_ext_v4si(*val, 2);
}
--cut here--

compiles with '-O2 -msse (-fomit-frame-pointer)' into:

fooSI_2:
       subl    $4, %esp
       movl    8(%esp), %eax
       movdqa  (%eax), %xmm0
       punpckhdq       %xmm0, %xmm0
       movd    %xmm0, (%esp)
       movl    (%esp), %eax
       addl    $4, %esp
       ret

fooSI_1:
       subl    $4, %esp
       movl    8(%esp), %eax
       pshufd  $85, (%eax), %xmm0
       movd    %xmm0, (%esp)
       movl    (%esp), %eax
       addl    $4, %esp
       ret

Attached patch simplifies this into:

fooSI_2:
       movl    4(%esp), %eax
       movl    8(%eax), %eax
       ret

fooSI_1:
       movl    4(%esp), %eax
       movl    4(%eax), %eax
       ret

Similar effects can be obtained for SFmode operands.

Patch was bootstrapped and regression tested on i686-pc-linux-gnu.

2007-07-13  Uros Bizjak  <ubizjak@gmail.com>

	PR target/32661
	* simplify-rtx.c (simplify_binary_operation_1) [VEC_SELECT]:
	Simplify nested VEC_SELECT (with optional VEC_CONCAT operator as
	operand) when top VEC_SELECT extracts scalar element.
	* config/i386/sse.md (*vec_extract_v4si_mem): New.
	(*vec_extract_v4sf_mem): Ditto.

testsuite/ChangeLog:

2007-07-13  Uros Bizjak  <ubizjak@gmail.com>

	PR target/32661
	* gcc.target/i386/pr32661.c: New test.

OK for mainline (patch needs middle-end/RTL approval) ?

Uros.

Attachment: pr32661.diff.txt
Description: Text document