Hi Segher,

on 2019/11/20 下午10:06, Segher Boessenkool wrote:
> Hi!
> 
> On Wed, Nov 20, 2019 at 03:31:36PM +0800, Kewen.Lin wrote:
> Yeah.  Just doing can_create_pseudo in the insn condition (and in the
> split condition, via &&) will work -- there just is this window of
> failure you should be aware of, and we need to do something about that,
> eventually.  It's a very general problem.  There are many places in
> rs6000 where we already do evil things with this (and some much worse
> than this even), so one more won't hurt too much.
> 
> It does need can_create_pseudo in the split condition then though.
> 

Done.

>>>   rtx comp1 = gen_rtx_fmt_ee (cond, <MODE>mode, operands[1], operands[2]);
>>>   rtx res1 = gen_reg_rtx (<MODE>mode);
>>>   emit_insn (gen_rtx_SET (res1, comp1));
>>>   rtx comp2 = gen_rtx_fmt_ee (cond, <MODE>mode, operands[2], operands[1]);
>>>   rtx res2 = gen_reg_rtx (<MODE>mode);
>>>   emit_insn (gen_rtx_SET (res2, comp2));
>>>
>>>   if (need_invert)
>>>     {
>>>       rtx comp3 = gen_rtx_fmt_ee (IOR, <MODE>mode, res1, res2);
>>>       rtx comp4 = gen_rtx_fmt_e (NOT, <MODE>mode, comp3);
>>
>> I had to add one more gen_rtx_SET for the IOR here, otherwise it gets the 
>> error
>> on the pattern can't be recognized.
> 
> Oh, sorry.  The canonical way of writing a NOR is as an ANDCC, so this
> should be
> 
>       rtx not1 = gen_rtx_fmt_e (NOT, <MODE>mode, res1);
>       rtx not2 = gen_rtx_fmt_e (NOT, <MODE>mode, res2);
>       rtx comp3 = gen_rtx_fmt_ee (AND, <MODE>mode, not1, not2);
>       emit_insn (gen_rtx_SET (operands[0], comp3));
> 

Thanks for the example, updated.

>> diff --git a/gcc/config/rs6000/vector.md b/gcc/config/rs6000/vector.md
>> index b132037..deeab9f 100644
>> --- a/gcc/config/rs6000/vector.md
>> +++ b/gcc/config/rs6000/vector.md
>> @@ -107,6 +107,12 @@
>>                               (smin "smin")
>>                               (smax "smax")])
>>  
>> +;; code iterators and attributes for vector FP comparison operators:
>> +(define_code_iterator
>> +    vector_fp_comparison_operator [lt le ne ungt unge unlt unle])
> 
> Can you change this name so it is clear it is just the simple ones?
> 

Renamed as vector_fp_comparison_simple and vector_fp_comparison_complex,
does it look better?

New version attached, compared with last version:
  1) Change code iterator names to vector_fp_comparison_simple
     and vector_fp_comparison_complex.
  2) Add can_create_pseudo in the split condition.
  3) Update NOR generation with AND (NOT, NOT).
  4) Append "DONE" at the end of define_insn_and_split.

Bootstrapped and regress tested on powerpc64le-linux-gnu.

BR,
Kewen

----
gcc/ChangeLog

2019-11-21 Kewen Lin  <li...@gcc.gnu.org>

        * config/rs6000/vector.md (vector_fp_comparison_simple):
        New code iterator.
        (vector_fp_comparison_complex): Likewise.
        (vector_<code><mode> for VEC_F and
        vector_fp_comparison_simple): New define_and_split.
        (vector_<code><mode> for VEC_F and
        vector_fp_comparison_complex): Likewise.
        (vector_lt<mode> for VEC_F): Refactor with
        vector_fp_comparison_simple.
        (vector_le<mode> for VEC_F): Likewise.
        (vector_unge<mode> for VEC_F): Likewise.
        (vector_unle<mode> for VEC_F): Likewise.
        (vector_ne<mode> for VEC_F): Likewise.
        (vector_ungt<mode> for VEC_F): Likewise.
        (vector_unlt<mode> for VEC_F): Likewise.
        (vector_ltgt<mode> for VEC_F): Refactor with
        vector_fp_comparison_complex.
        (vector_ordered<mode> for VEC_F): Likewise.
        (vector_uneq<mode> for VEC_F): Likewise.
        (vector_unordered<mode> for VEC_F): Likewise.
diff --git a/gcc/config/rs6000/vector.md b/gcc/config/rs6000/vector.md
index b132037..aa3154a 100644
--- a/gcc/config/rs6000/vector.md
+++ b/gcc/config/rs6000/vector.md
@@ -107,6 +107,12 @@
                                 (smin "smin")
                                 (smax "smax")])
 
+;; code iterators and attributes for vector FP comparison operators:
+(define_code_iterator
+       vector_fp_comparison_simple [lt le ne ungt unge unlt unle])
+(define_code_iterator
+       vector_fp_comparison_complex [ltgt uneq unordered ordered])
+
 
 ;; Vector move instructions.  Little-endian VSX loads and stores require
 ;; special handling to circumvent "element endianness."
@@ -665,88 +671,6 @@
   DONE;
 })
 
-; lt(a,b) = gt(b,a)
-(define_expand "vector_lt<mode>"
-  [(set (match_operand:VEC_F 0 "vfloat_operand")
-       (lt:VEC_F (match_operand:VEC_F 1 "vfloat_operand")
-                 (match_operand:VEC_F 2 "vfloat_operand")))]
-  "VECTOR_UNIT_ALTIVEC_OR_VSX_P (<MODE>mode)"
-{
-  emit_insn (gen_vector_gt<mode> (operands[0], operands[2], operands[1]));
-  DONE;
-})
-
-; le(a,b) = ge(b,a)
-(define_expand "vector_le<mode>"
-  [(set (match_operand:VEC_F 0 "vfloat_operand")
-       (le:VEC_F (match_operand:VEC_F 1 "vfloat_operand")
-                 (match_operand:VEC_F 2 "vfloat_operand")))]
-  "VECTOR_UNIT_ALTIVEC_OR_VSX_P (<MODE>mode)"
-{
-  emit_insn (gen_vector_ge<mode> (operands[0], operands[2], operands[1]));
-  DONE;
-})
-
-; ne(a,b) = ~eq(a,b)
-(define_expand "vector_ne<mode>"
-  [(set (match_operand:VEC_F 0 "vfloat_operand")
-       (ne:VEC_F (match_operand:VEC_F 1 "vfloat_operand")
-                 (match_operand:VEC_F 2 "vfloat_operand")))]
-  "VECTOR_UNIT_ALTIVEC_OR_VSX_P (<MODE>mode)"
-{
-  emit_insn (gen_vector_eq<mode> (operands[0], operands[1], operands[2]));
-  emit_insn (gen_one_cmpl<mode>2 (operands[0], operands[0]));
-  DONE;
-})
-
-; unge(a,b) = ~gt(b,a)
-(define_expand "vector_unge<mode>"
-  [(set (match_operand:VEC_F 0 "vfloat_operand")
-       (unge:VEC_F (match_operand:VEC_F 1 "vfloat_operand")
-                   (match_operand:VEC_F 2 "vfloat_operand")))]
-  "VECTOR_UNIT_ALTIVEC_OR_VSX_P (<MODE>mode)"
-{
-  emit_insn (gen_vector_gt<mode> (operands[0], operands[2], operands[1]));
-  emit_insn (gen_one_cmpl<mode>2 (operands[0], operands[0]));
-  DONE;
-})
-
-; ungt(a,b) = ~ge(b,a)
-(define_expand "vector_ungt<mode>"
-  [(set (match_operand:VEC_F 0 "vfloat_operand")
-       (ungt:VEC_F (match_operand:VEC_F 1 "vfloat_operand")
-                   (match_operand:VEC_F 2 "vfloat_operand")))]
-  "VECTOR_UNIT_ALTIVEC_OR_VSX_P (<MODE>mode)"
-{
-  emit_insn (gen_vector_ge<mode> (operands[0], operands[2], operands[1]));
-  emit_insn (gen_one_cmpl<mode>2 (operands[0], operands[0]));
-  DONE;
-})
-
-; unle(a,b) = ~gt(a,b)
-(define_expand "vector_unle<mode>"
-  [(set (match_operand:VEC_F 0 "vfloat_operand")
-       (unle:VEC_F (match_operand:VEC_F 1 "vfloat_operand")
-                   (match_operand:VEC_F 2 "vfloat_operand")))]
-  "VECTOR_UNIT_ALTIVEC_OR_VSX_P (<MODE>mode)"
-{
-  emit_insn (gen_vector_gt<mode> (operands[0], operands[1], operands[2]));
-  emit_insn (gen_one_cmpl<mode>2 (operands[0], operands[0]));
-  DONE;
-})
-
-; unlt(a,b) = ~ge(a,b)
-(define_expand "vector_unlt<mode>"
-  [(set (match_operand:VEC_F 0 "vfloat_operand")
-       (unlt:VEC_F (match_operand:VEC_F 1 "vfloat_operand")
-                   (match_operand:VEC_F 2 "vfloat_operand")))]
-  "VECTOR_UNIT_ALTIVEC_OR_VSX_P (<MODE>mode)"
-{
-  emit_insn (gen_vector_ge<mode> (operands[0], operands[1], operands[2]));
-  emit_insn (gen_one_cmpl<mode>2 (operands[0], operands[0]));
-  DONE;
-})
-
 (define_expand "vector_eq<mode>"
   [(set (match_operand:VEC_C 0 "vlogical_operand")
        (eq:VEC_C (match_operand:VEC_C 1 "vlogical_operand")
@@ -761,13 +685,6 @@
   "VECTOR_UNIT_ALTIVEC_OR_VSX_P (<MODE>mode)"
   "")
 
-(define_expand "vector_ge<mode>"
-  [(set (match_operand:VEC_F 0 "vlogical_operand")
-       (ge:VEC_F (match_operand:VEC_F 1 "vlogical_operand")
-                 (match_operand:VEC_F 2 "vlogical_operand")))]
-  "VECTOR_UNIT_ALTIVEC_OR_VSX_P (<MODE>mode)"
-  "")
-
 ; >= for integer vectors: swap operands and apply not-greater-than
 (define_expand "vector_nlt<mode>"
   [(set (match_operand:VEC_I 3 "vlogical_operand")
@@ -829,88 +746,122 @@
   operands[3] = gen_reg_rtx_and_attrs (operands[0]);
 })
 
-(define_insn_and_split "vector_uneq<mode>"
-  [(set (match_operand:VEC_F 0 "vfloat_operand")
-       (uneq:VEC_F (match_operand:VEC_F 1 "vfloat_operand")
-                   (match_operand:VEC_F 2 "vfloat_operand")))]
-  "VECTOR_UNIT_ALTIVEC_OR_VSX_P (<MODE>mode)"
-  "#"
-  ""
-  [(set (match_dup 3)
-       (gt:VEC_F (match_dup 1)
-                 (match_dup 2)))
-   (set (match_dup 4)
-       (gt:VEC_F (match_dup 2)
-                 (match_dup 1)))
-   (set (match_dup 0)
-       (and:VEC_F (not:VEC_F (match_dup 3))
-                  (not:VEC_F (match_dup 4))))]
-{
-  operands[3] = gen_reg_rtx (<MODE>mode);
-  operands[4] = gen_reg_rtx (<MODE>mode);
-})
+; There are 14 possible vector FP comparison operators, gt and eq of them have
+; been expanded above, so just support 12 remaining operators here.
 
-(define_insn_and_split "vector_ltgt<mode>"
-  [(set (match_operand:VEC_F 0 "vfloat_operand")
-       (ltgt:VEC_F (match_operand:VEC_F 1 "vfloat_operand")
-                   (match_operand:VEC_F 2 "vfloat_operand")))]
+; For ge:
+(define_expand "vector_ge<mode>"
+  [(set (match_operand:VEC_F 0 "vlogical_operand")
+       (ge:VEC_F (match_operand:VEC_F 1 "vlogical_operand")
+                 (match_operand:VEC_F 2 "vlogical_operand")))]
   "VECTOR_UNIT_ALTIVEC_OR_VSX_P (<MODE>mode)"
-  "#"
-  ""
-  [(set (match_dup 3)
-       (gt:VEC_F (match_dup 1)
-                 (match_dup 2)))
-   (set (match_dup 4)
-       (gt:VEC_F (match_dup 2)
-                 (match_dup 1)))
-   (set (match_dup 0)
-       (ior:VEC_F (match_dup 3)
-                  (match_dup 4)))]
-{
-  operands[3] = gen_reg_rtx (<MODE>mode);
-  operands[4] = gen_reg_rtx (<MODE>mode);
-})
+  "")
 
-(define_insn_and_split "vector_ordered<mode>"
+; For lt/le/ne/ungt/unge/unlt/unle:
+; lt(a,b)   = gt(b,a)
+; le(a,b)   = ge(b,a)
+; unge(a,b) = ~lt(a,b)
+; unle(a,b) = ~gt(a,b)
+; ne(a,b)   = ~eq(a,b)
+; ungt(a,b) = ~le(a,b)
+; unlt(a,b) = ~ge(a,b)
+(define_insn_and_split "vector_<code><mode>"
   [(set (match_operand:VEC_F 0 "vfloat_operand")
-       (ordered:VEC_F (match_operand:VEC_F 1 "vfloat_operand")
-                      (match_operand:VEC_F 2 "vfloat_operand")))]
+       (vector_fp_comparison_simple:VEC_F
+                          (match_operand:VEC_F 1 "vfloat_operand")
+                          (match_operand:VEC_F 2 "vfloat_operand")))]
   "VECTOR_UNIT_ALTIVEC_OR_VSX_P (<MODE>mode)"
   "#"
-  ""
-  [(set (match_dup 3)
-       (ge:VEC_F (match_dup 1)
-                 (match_dup 2)))
-   (set (match_dup 4)
-       (ge:VEC_F (match_dup 2)
-                 (match_dup 1)))
-   (set (match_dup 0)
-       (ior:VEC_F (match_dup 3)
-                  (match_dup 4)))]
+  "&& 1"
+  [(pc)]
 {
-  operands[3] = gen_reg_rtx (<MODE>mode);
-  operands[4] = gen_reg_rtx (<MODE>mode);
+  enum rtx_code cond = <CODE>;
+  bool need_invert = false;
+
+  if (cond == UNLE || cond == UNLT || cond == NE || cond == UNGE
+      || cond == UNGT)
+    {
+      cond = reverse_condition_maybe_unordered (cond);
+      need_invert = true;
+    }
+
+  if (cond == LT || cond == LE)
+    {
+      cond = swap_condition (cond);
+      std::swap (operands[1], operands[2]);
+    }
+
+  gcc_assert (cond == EQ || cond == GE || cond == GT);
+
+  rtx comp = gen_rtx_fmt_ee (cond, <MODE>mode, operands[1], operands[2]);
+
+  if (need_invert)
+    {
+      rtx res
+       = (!can_create_pseudo_p () ? operands[0] : gen_reg_rtx (<MODE>mode));
+      emit_insn (gen_rtx_SET (res, comp));
+      emit_insn (gen_one_cmpl<mode>2 (operands[0], res));
+    }
+  else
+    emit_insn (gen_rtx_SET (operands[0], comp));
+
+  DONE;
 })
 
-(define_insn_and_split "vector_unordered<mode>"
+; For ltgt/uneq/ordered/unordered:
+; ltgt: gt(a,b) | gt(b,a)
+; uneq: ~(gt(a,b) | gt(b,a))
+; ordered: ge(a,b) | ge(b,a)
+; unordered: ~(ge(a,b) | ge(b,a))
+(define_insn_and_split "vector_<code><mode>"
   [(set (match_operand:VEC_F 0 "vfloat_operand")
-       (unordered:VEC_F (match_operand:VEC_F 1 "vfloat_operand")
-                        (match_operand:VEC_F 2 "vfloat_operand")))]
-  "VECTOR_UNIT_ALTIVEC_OR_VSX_P (<MODE>mode)"
+       (vector_fp_comparison_complex:VEC_F
+                          (match_operand:VEC_F 1 "vfloat_operand")
+                          (match_operand:VEC_F 2 "vfloat_operand")))]
+  "VECTOR_UNIT_ALTIVEC_OR_VSX_P (<MODE>mode) && can_create_pseudo_p ()"
   "#"
-  ""
-  [(set (match_dup 3)
-       (ge:VEC_F (match_dup 1)
-                 (match_dup 2)))
-   (set (match_dup 4)
-       (ge:VEC_F (match_dup 2)
-                 (match_dup 1)))
-   (set (match_dup 0)
-        (and:VEC_F (not:VEC_F (match_dup 3))
-                   (not:VEC_F (match_dup 4))))]
+  "&& can_create_pseudo_p ()"
+  [(pc)]
 {
-  operands[3] = gen_reg_rtx (<MODE>mode);
-  operands[4] = gen_reg_rtx (<MODE>mode);
+  enum rtx_code cond = <CODE>;
+  bool need_invert = false;
+
+  if (cond == UNORDERED || cond == UNEQ)
+    {
+      cond = reverse_condition_maybe_unordered (cond);
+      need_invert = true;
+    }
+
+  switch (cond)
+    {
+    case LTGT:
+      cond = GT;
+      break;
+    case ORDERED:
+      cond = GE;
+      break;
+    default:
+      gcc_unreachable ();
+    }
+
+  rtx comp1 = gen_rtx_fmt_ee (cond, <MODE>mode, operands[1], operands[2]);
+  rtx res1 = gen_reg_rtx (<MODE>mode);
+  emit_insn (gen_rtx_SET (res1, comp1));
+  rtx comp2 = gen_rtx_fmt_ee (cond, <MODE>mode, operands[2], operands[1]);
+  rtx res2 = gen_reg_rtx (<MODE>mode);
+  emit_insn (gen_rtx_SET (res2, comp2));
+
+  if (need_invert)
+    {
+      rtx not1 = gen_rtx_fmt_e (NOT, <MODE>mode, res1);
+      rtx not2 = gen_rtx_fmt_e (NOT, <MODE>mode, res2);
+      rtx comp3 = gen_rtx_fmt_ee (AND, <MODE>mode, not1, not2);
+      emit_insn (gen_rtx_SET (operands[0], comp3));
+    }
+  else
+    emit_insn (gen_ior<mode>3 (operands[0], res1, res2));
+
+  DONE;
 })
 
 ;; Note the arguments for __builtin_altivec_vsel are op2, op1, mask

Reply via email to