diff mbox

Power/GCC: Implement little-endian SPE operations

Message ID alpine.DEB.1.10.1407071530220.16254@tp.orcam.me.uk
State Accepted
Headers show

Commit Message

Maciej W. Rozycki July 7, 2014, 2:33 p.m. UTC 
On Mon, 7 Jul 2014, David Edelsohn wrote:

> gcc/
> * config/rs6000/rs6000.c (output_vec_const_move): Handle
> little-endian code generation.
> * config/rs6000/spe.md (spe_evmergehi): Rename to...
> (vec_perm00_v2si): ... this.  Handle little-endian code
> generation.
> (spe_evmergehilo): Rename to...
> (vec_perm01_v2si): ... this.  Handle little-endian code
> generation.
> (spe_evmergelo): Rename to...
> (vec_perm11_v2si): ... this.  Handle little-endian code
> generation.
> (spe_evmergelohi): Rename to...
> (vec_perm10_v2si): ... this.  Handle little-endian code
> generation.
> (spe_evmergehi, spe_evmergehilo): New expanders.
> (spe_evmergelo, spe_evmergelohi): Likewise.
> (*frob_<SPE64:mode>_<DITI:mode>): Handle little-endian code
> generation.
> (*frob_tf_ti): Likewise.
> (*frob_<mode>_di_2): Likewise.
> (*frob_tf_di_8_2): Likewise.
> (*frob_di_<mode>): Likewise.
> (*frob_ti_tf): Likewise.
> (*frob_<DITI:mode>_<SPE64:mode>_2): Likewise.
> (*frob_ti_<mode>_8_2): Likewise.
> (*frob_ti_tf_2): Likewise.
> (mov_si<mode>_e500_subreg0): Rename to...
> (mov_si<mode>_e500_subreg0_be): ... this.  Restrict to the big
> endianness only.
> (*mov_si<mode>_e500_subreg0_le): New instruction pattern.
> (*mov_si<mode>_e500_subreg0_elf_low): Rename to...
> (*mov_si<mode>_e500_subreg0_elf_low_be): ... this.  Restrict to
> the big endianness only.
> (*mov_si<mode>_e500_subreg0_elf_low_le): New instruction pattern.
> (*mov_si<mode>_e500_subreg0_2): Rename to...
> (*mov_si<mode>_e500_subreg0_2_be): ... this.  Restrict to the
> big big endianness only.
> (*mov_si<mode>_e500_subreg0_2_le): New instruction pattern.
> (*mov_si<mode>_e500_subreg4): Rename to...
> (*mov_si<mode>_e500_subreg4_be): ... this.  Restrict to the big
> endianness only.
> (mov_si<mode>_e500_subreg4_le): New instruction pattern.
> (*mov_si<mode>_e500_subreg4_elf_low): Rename to...
> (*mov_si<mode>_e500_subreg4_elf_low_be): ... this.  Restrict to
> the big endianness only.
> (*mov_si<mode>_e500_subreg4_elf_low_le): New instruction/splitter
> pattern.
> (*mov_si<mode>_e500_subreg4_2): Rename to...
> (*mov_si<mode>_e500_subreg4_2_be): ... this.  Restrict to the big
> endianness only.
> (*mov_si<mode>_e500_subreg4_2_le): New instruction pattern.
> (*mov_sitf_e500_subreg8): Rename to...
> (*mov_sitf_e500_subreg8_be): ... this.  Restrict to the big
> endianness only.
> (*mov_sitf_e500_subreg8_le): New instruction pattern.
> (*mov_sitf_e500_subreg8_2): Rename to...
> (*mov_sitf_e500_subreg8_2_be): ... this.  Restrict to the big
> endianness only.
> (*mov_sitf_e500_subreg8_2_le): New instruction pattern.
> (*mov_sitf_e500_subreg12): Rename to...
> (*mov_sitf_e500_subreg12_be): ... this.  Restrict to the big
> endianness only.
> (*mov_sitf_e500_subreg12_le): New instruction pattern.
> (*mov_sitf_e500_subreg12_2): Rename to...
> (*mov_sitf_e500_subreg12_2_be): ... this.  Restrict to the big
> endianness only.
> (*mov_sitf_e500_subreg12_2_le): New instruction pattern.
> 
> gcc/testsuite/
> * gcc.target/powerpc/spe-evmerge.c: New file.
> 
> Okay.
> 
> Could you add a short comment explaining what the "0" and "1" labels
> in vec_perm[01][01]_v2si mean?

 Like this?

  Maciej

gcc-ppc-spe-le-update.diff

Comments

David Edelsohn July 7, 2014, 3:05 p.m. UTC | #1 
On Mon, Jul 7, 2014 at 10:33 AM, Maciej W. Rozycki
<macro@codesourcery.com> wrote:

>> Could you add a short comment explaining what the "0" and "1" labels
>> in vec_perm[01][01]_v2si mean?
>
>  Like this?
>
>   Maciej
>
> gcc-ppc-spe-le-update.diff
> Index: gcc-fsf-trunk-quilt/gcc/config/rs6000/spe.md
> ===================================================================
> --- gcc-fsf-trunk-quilt.orig/gcc/config/rs6000/spe.md   2014-07-07 15:27:54.258937029 +0100
> +++ gcc-fsf-trunk-quilt/gcc/config/rs6000/spe.md        2014-07-07 15:26:43.748937008 +0100
> @@ -438,6 +438,11 @@
>    [(set_attr "type" "vecload")
>     (set_attr  "length" "4")])
>
> +;; Integer vector permutation instructions.  The pairs of digits in the
> +;; names of these instructions indicate the indices, in the memory vector
> +;; element ordering, of the vector elements permuted to the output vector
> +;; from the first and the second input vector respectively.

Yes, that's good. It helps if someone reading the code doesn't need to
reverse engineer the numbering convention of the name.

Thanks, David
Maciej W. Rozycki July 7, 2014, 3:49 p.m. UTC | #2 
On Mon, 7 Jul 2014, David Edelsohn wrote:

> Yes, that's good. It helps if someone reading the code doesn't need to
> reverse engineer the numbering convention of the name.

 Sure, I haven't questioned your request.  Thanks for your review, applied 
now.

  Maciej
diff mbox

Patch

Index: gcc-fsf-trunk-quilt/gcc/config/rs6000/spe.md
===================================================================
--- gcc-fsf-trunk-quilt.orig/gcc/config/rs6000/spe.md	2014-07-07 15:27:54.258937029 +0100
+++ gcc-fsf-trunk-quilt/gcc/config/rs6000/spe.md	2014-07-07 15:26:43.748937008 +0100
@@ -438,6 +438,11 @@ 
   [(set_attr "type" "vecload")
    (set_attr  "length" "4")])
 
+;; Integer vector permutation instructions.  The pairs of digits in the
+;; names of these instructions indicate the indices, in the memory vector
+;; element ordering, of the vector elements permuted to the output vector
+;; from the first and the second input vector respectively.
+
 (define_insn "vec_perm00_v2si"
   [(set (match_operand:V2SI 0 "gpc_reg_operand" "=r")
 	(vec_select:V2SI
@@ -571,6 +576,8 @@ 
   DONE;
 })
 
+;; End of integer vector permutation instructions.
+
 (define_insn "spe_evnand"
   [(set (match_operand:V2SI 0 "gpc_reg_operand" "=r")
         (not:V2SI (and:V2SI (match_operand:V2SI 1 "gpc_reg_operand" "r")