[AArch64] Support zero-extended move to FP register

The popcount expansion uses SIMD instructions acting on 64-bit values.
As a result a popcount of a 32-bit integer requires zero-extension before 
moving the zero-extended value into an FP register.  This patch adds
support for zero-extended int->FP moves to avoid the redundant uxtw.
Similarly, add support for 32-bit zero-extending load->FP register
and 32-bit zero-extending FP->FP and FP->int moves.
Add a missing 'fp' arch attribute to the related 8/16-bit pattern and
fix an incorrect type attribute.

To complete zero-extended load support, add a new alternative to 
load_pair_zero_extendsidi2_aarch64 to support LDP into FP registers too.

int f (int a)
{
  return __builtin_popcount (a);
}

Before:
	uxtw	x0, w0
	fmov	d0, x0
	cnt	v0.8b, v0.8b
	addv	b0, v0.8b
	fmov	w0, s0
	ret

After:
	fmov	s0, w0
	cnt	v0.8b, v0.8b
	addv	b0, v0.8b
	fmov	w0, s0
	ret

Passes regress & bootstrap on AArch64.

    gcc/
	* config/aarch64/aarch64.md (zero_extendsidi2_aarch64): Add alternatives
	to zero-extend between int and floating-point registers.
	(load_pair_zero_extendsidi2_aarch64): Add alternative for zero-extended
	ldp into floating-point registers.  Add type and arch attributes.
	(zero_extend<SHORT:mode><GPI:mode>2_aarch64): Add arch attribute.
	Use f_loads for type attribute.

    testsuite/
	* gcc.target/aarch64/popcnt.c: Test zero-extended popcount.
	* gcc.target/aarch64/vec_zeroextend.c: Test zero-extended vectors.

From-SVN: r265079

0 files changed, 0 insertions, 0 deletions