Re: [PATCH] Add ISA 3.0 PowerPC support for VEXTU{B,H,W}{L,R}X instructions

[Michael Meissner <meissner at linux dot vnet dot ibm dot com>]

This patch should address the comments in the last patch.

I have tested this patch with bootstrap builds and make check regression tests
on a little endian Power8 64-bit system and a big endian Power7 32/64-bit
system with no regressions.  Can I check this into the trunk?

[gcc]
2016-12-13  Michael Meissner  <meissner@linux.vnet.ibm.com>

	* config/rs6000/rs6000.c (rs6000_split_vec_extract_var): On ISA
	3.0/power9, add support to use the VEXTU{B,H,W}{L,R}X extract
	instructions.
	* config/rs6000/vsx.md (VSr2): Add IEEE 128-bit floating point
	type constraint registers.
	(VSr3): Likewise.
	(FL_CONV): New mode iterator for binary floating types that have a
	direct conversion from 64-bit integer to floating point.
	(vsx_extract_<mode>_p9): Add support for the ISA 3.0/power9
	VEXTU{B,H,W}{L,R}X extract instructions.
	(vsx_extract_<mode>_p9 splitter): Add splitter to load up the
	extract byte position into the GPR if we are using the
	VEXTU{B,H,W}{L,R}X extract instructions.
	(vsx_extract_<mode>_di_p9): Support extracts to GPRs.
	(vsx_extract_<mode>_store_p9): Support extracting to GPRs so that
	we can use reg+offset address instructions.
	(vsx_extract_<mode>_var): Support extracts to GPRs.
	(vsx_extract_<VSX_EXTRACT_I:mode>_<SDI:mode>_var): New combiner
	insn to combine vector extracts with zero_extend.
	(vsx_ext_<VSX_EXTRACT_I:VS_scalar>_fl_<FL_CONV:mode>): Optimize
	extracting a small integer vector element and converting it to a
	floating point type.
	(vsx_ext_<VSX_EXTRACT_I:VS_scalar>_ufl_<FL_CONV:mode>): Likewise.

[gcc/testsuite]
2016-12-13  Michael Meissner  <meissner@linux.vnet.ibm.com>

	* gcc/testsuite/gcc.target/powerpc/vec-extract.h: If DO_TRACE is
	defined, add tracing of the various extracts to stderr.  Add
	support for tests that convert the result to another type.
	* gcc/testsuite/gcc.target/powerpc/vec-extract-v2df.c: Likewise.
	* gcc/testsuite/gcc.target/powerpc/vec-extract-v4sf.c: Likewise.
	* gcc/testsuite/gcc.target/powerpc/vec-extract-v4si-df.c: Add new
	tests that do an extract and then convert the values double.
	* gcc/testsuite/gcc.target/powerpc/vec-extract-v4siu-df.c: Likewise.
	* gcc/testsuite/gcc.target/powerpc/vec-extract-v16qiu-df.c: Likewise.
	* gcc/testsuite/gcc.target/powerpc/vec-extract-v16qi-df.c: Likewise.
	* gcc/testsuite/gcc.target/powerpc/vec-extract-v8hiu-df.c: Likewise.
	* gcc/testsuite/gcc.target/powerpc/vec-extract-v8hi-df.c: Likewise.
	* gcc.target/powerpc/p9-extract-1.c: Update test to check for
	VEXTU{B,H,W}{L,R}X instructions being generated by default instead
	of VEXTRACTU{B,H} and XXEXTRACTUW.
	* gcc.target/powerpc/p9-extract-3.c: New test for combination of
	vec_extract and convert to floating point.

-- 
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: meissner@linux.vnet.ibm.com, phone: +1 (978) 899-4797

Attachment: gcc-stage7.intvsx007b
Description: Text document