i386: Improve HImode interunit moves

i386: Improve HImode interunit moves

Currently, the compiler moves HImode values between GPR and XMM registers with:

	%vpinsrw\t{$0, %k1, %d0|%d0, %k1, 0}
	%vpextrw\t{$0, %1, %k0|%k0, %1, 0}

but it could use slightly faster and shorter:

	%vmovd\t{%k1, %0|%0, %k1}
	%vmovd\t{%1, %k0|%k0, %1}

2022-01-06  Uroš Bizjak  <ubizjak@gmail.com>

gcc/ChangeLog:

	* config/i386/i386.c (ix86_output_ssemov) <MODE_DI>:
	Add %q modifier for operands in general registers.
	<MODE_SI>: Add %q modifier for operands in general registers.
	* config/i386/i386.md (*movhi_internal): Change type attribute of
	xmm-gpr interunit alternatives 9,10 to ssemov and mode attribute
	to SImode for non-avx512fp16 targets.
	(*movhf_internal): Ditto for xmm-gpr interunit alternatives 6,8.
	* config/i386/mmx.md (*movv2qi_internal):
	Ditto for xmm-gpr interunit alternatives 8,9.

gcc/testsuite/ChangeLog:

	* gcc.target/i386/pr102811-2.c (dg-final):
	Update scan-assembler-times directives.
	* gcc.target/i386/sse2-float16-2.c (dg-final):
	Update scan-assembler directives.