i386: Fix blend vector permutation for 8-byte modes

i386: Fix blend vector permutation for 8-byte modes

8-byte modes should be processed only for TARGET_MMX_WITH_SSE. Handle
V2SFmode and fix V2HImode handling. The resulting BLEND instructions
are always faster than MOVSS/MOVSD, so prioritize them w.r.t MOVSS/MOVSD
for TARGET_SSE4_1.

gcc/ChangeLog:

	* config/i386/i386-expand.cc (expand_vec_perm_blend):
	Handle 8-byte modes only with TARGET_MMX_WITH_SSE. Handle V2SFmode
	and fix V2HImode handling.
	(expand_vec_perm_1): Try to emit BLEND instruction
	before MOVSS/MOVSD.
	* config/i386/mmx.md (*mmx_blendps): New insn pattern.

gcc/testsuite/ChangeLog:

	* gcc.target/i386/merge-1.c (dg-options): Use -mno-sse4.
	* gcc.target/i386/sse2-mmx-21.c (dg-options): Ditto.
	* gcc.target/i386/sse-movss-4.c (dg-options):
	Use -mno-sse4.  Simplify scan-assembler-not strings.
	* gcc.target/i386/sse2-movsd-3.c (dg-options): Ditto.
	* gcc.target/i386/sse2-mmx-movss-1.c: New test.