[RFC PATCH, vectorizer]: Vectorize int -> double conversions

["Uros Bizjak" <ubizjak at gmail dot com>]

Hello!

Attached (RFC!) patch implements vectorization of int -> double
conversion. The testcase:

 /* int -> double */
 for (i = 0; i < N; i++)
   {
     da[i] = (double) ib[i];
   }

compiles on i686 -msse2 target into:

.L3:
       movdqa  (%eax,%ecx), %xmm0
       cvtdq2pd        %xmm0, %xmm1
       pshufd  $238, %xmm0, %xmm0
       movapd  %xmm1, (%edx,%eax,2)
       cvtdq2pd        %xmm0, %xmm0
       movapd  %xmm0, 16(%edx,%eax,2)
       addl    $16, %eax
       cmpl    $128, %eax
       jne     .L3

(pshufd is there to shuffle SImode vector from x0x1x2x3 into x2x3x2x3.

Regarding the patch: in vectorizable_conversion() we detect
(nunits_out == nunits_in / 2) as EXPAND case and handle conversion in
the same way as vectorizable_type_promotion(). Unfortunatelly, we need
a couple of new tree codes and 4 new optabs to handle signed and
unsigned conversion.

Unfortunatelly, if correct optabs are not defined, compilation aborts
in vect_transform_stmt():

4351    case type_conversion_vec_info_type:
4352      done = vectorizable_conversion (stmt, bsi, &vec_stmt);
4353      gcc_assert (done);
4354      break;

vectorizable_conversion() returns false when it can't handle EXPAND
case due to missing optabs. Sure, we will abort then, but
vectorizable_convresion() has many other exits with "false". I'm kind
of out of ideas here how transformation should be skipped in this
case. This is why the patch is RFC only.

The code is also prepared for SHRINK case,  but it is not handled ATM.

2007-04-23  Uros Bizjak  <ubizjak@gmail.com>

	PR tree-optimization/24659
       * optabs.h (enum optab_index): Add OTI_vec_unpacks_float_hi,
	OTI_vec_unpacks_float_lo, OTI_vec_unpacku_float_hi and
	OTI_vec_unpacku_float_lo.
	(vec_unpacks_float_hi_optab): Define new macro.
	(vec_unpacks_float_lo_optab): Ditto.
	(vec_unpacku_float_hi_optab): Ditto.
	(vec_unpacku_float_lo_optab): Ditto.
	* genopinit.c (optabs): Implement vec_unpack[s|u]_[hi|lo]_optab
	using vec_unpack[s|u]_[hi\lo]_* patterns.
	* tree-vect-transform.c (vect_gen_widened_results_half): Prototype.
	(vectorizable_conversion): Handle (nunits_in == nunits_out / 2)
	and (nunits_out == nunits_in / 2) cases.
	* optabs.c (optab_for_tree_code) Return vec_unpack[s|u]_float_hi_optab
	for VEC_UNPACK_FLOAT_HI_EXPR and vec_unpack[s|u]_float_lo_optab
	for VEC_UNPACK_FLOAT_LO_EXPR.
	(init_optabs): Initialize vec_unpack[s|u]_[hi|lo]_optab.

	* tree.def (VEC_UNPACK_FLOAT_HI_EXPR, VEC_UNPACK_FLOAT_LO_EXPR):
	New tree codes.
	* tree-pretty-print.c (dump_generic_node): Handle
	VEC_UNPACK_FLOAT_HI_EXPR and VEC_UNPACK_FLOAT_LO_EXPR.
	(op_prio): Ditto.
	* expr.c (expand_expr_real_1): Ditto.
	* tree-inline.c (estimate_num_insns_1): Ditto.
	* tree-vect-generic.c (expand_vector_operations_1): Ditto.

	* config/i386/sse.md (vec_unpacks_float_hi_v4si): New expander.
	(vec_unpacks_float_lo_v4si): Ditto.

testsuite/ChangeLog:

2007-04-23  Uros Bizjak  <ubizjak@gmail.com>

	PR tree-optimization/24659
	* gcc.dg/vect/vect-float-intfloat-3.c: New test.

Uros.

Attachment: vect-intfloat.diff
Description: Binary data