Fix non-comforming expander for floatv2div2sf2,floatunsv2div2sf2,fix_truncv2sfv2di...

2020-05-26  Hongtao Liu  <hongtao.liu@intel.com>
	    Uroš Bizjak  <ubizjak@gmail.com>

gcc/ChangeLog:
	PR target/95211
	PR target/95256
	* config/i386/sse.md (<floatunssuffix>v2div2sf2): New expander.
	(fix<fixunssuffix>_truncv2sfv2di2): Ditto.
	(avx512dq_float<floatunssuffix>v2div2sf2): Renaming from
	float<floatunssuffix>v2div2sf2.
	(avx512dq_fix<fixunssuffix>_truncv2sfv2di2<mask_name>):
	Renaming from fix<fixunssuffix>_truncv2sfv2di2<mask_name>.
	(vec_pack<floatprefix>_float_<mode>): Adjust icode name.
	(vec_unpack_<fixprefix>fix_trunc_lo_<mode>): Ditto.
	(vec_unpack_<fixprefix>fix_trunc_hi_<mode>): Ditto.
	* config/i386/i386-builtin.def: Ditto.
	* emit-rtl.c (validate_subreg): Allow use of *paradoxical* vector
	subregs when both omode and imode are vector mode and
	have the same inner mode.

gcc/testsuite/ChangeLog
	* gcc.target/i386/pr95211.c: New test.

diff --git a/gcc/config/i386/i386-builtin.def b/gcc/config/i386/i386-builtin.def
index fa123788a8e49a023ea17c69f7cd56ecd3acca47..b873498f3ab976aa1c2629c107171427b1a4a7b7 100644 (file)
--- a/gcc/config/i386/i386-builtin.def
+++ b/gcc/config/i386/i386-builtin.def
@@ -1649,9 +1649,9 @@ BDESC (OPTION_MASK_ISA_AVX512DQ | OPTION_MASK_ISA_AVX512VL, 0, CODE_FOR_ufix_not
 BDESC (OPTION_MASK_ISA_AVX512VL, 0, CODE_FOR_ufix_notruncv4dfv4si2_mask, "__builtin_ia32_cvtpd2udq256_mask", IX86_BUILTIN_CVTPD2UDQ256_MASK, UNKNOWN, (int) V4SI_FTYPE_V4DF_V4SI_UQI)
 BDESC (OPTION_MASK_ISA_AVX512VL, 0, CODE_FOR_ufix_notruncv2dfv2si2_mask, "__builtin_ia32_cvtpd2udq128_mask", IX86_BUILTIN_CVTPD2UDQ128_MASK, UNKNOWN, (int) V4SI_FTYPE_V2DF_V4SI_UQI)
 BDESC (OPTION_MASK_ISA_AVX512DQ | OPTION_MASK_ISA_AVX512VL, 0, CODE_FOR_fix_truncv4sfv4di2_mask, "__builtin_ia32_cvttps2qq256_mask", IX86_BUILTIN_CVTTPS2QQ256, UNKNOWN, (int) V4DI_FTYPE_V4SF_V4DI_UQI)
-BDESC (OPTION_MASK_ISA_AVX512DQ | OPTION_MASK_ISA_AVX512VL, 0, CODE_FOR_fix_truncv2sfv2di2_mask, "__builtin_ia32_cvttps2qq128_mask", IX86_BUILTIN_CVTTPS2QQ128, UNKNOWN, (int) V2DI_FTYPE_V4SF_V2DI_UQI)
+BDESC (OPTION_MASK_ISA_AVX512DQ | OPTION_MASK_ISA_AVX512VL, 0, CODE_FOR_avx512dq_fix_truncv2sfv2di2_mask, "__builtin_ia32_cvttps2qq128_mask", IX86_BUILTIN_CVTTPS2QQ128, UNKNOWN, (int) V2DI_FTYPE_V4SF_V2DI_UQI)
 BDESC (OPTION_MASK_ISA_AVX512DQ | OPTION_MASK_ISA_AVX512VL, 0, CODE_FOR_fixuns_truncv4sfv4di2_mask, "__builtin_ia32_cvttps2uqq256_mask", IX86_BUILTIN_CVTTPS2UQQ256, UNKNOWN, (int) V4DI_FTYPE_V4SF_V4DI_UQI)
-BDESC (OPTION_MASK_ISA_AVX512DQ | OPTION_MASK_ISA_AVX512VL, 0, CODE_FOR_fixuns_truncv2sfv2di2_mask, "__builtin_ia32_cvttps2uqq128_mask", IX86_BUILTIN_CVTTPS2UQQ128, UNKNOWN, (int) V2DI_FTYPE_V4SF_V2DI_UQI)
+BDESC (OPTION_MASK_ISA_AVX512DQ | OPTION_MASK_ISA_AVX512VL, 0, CODE_FOR_avx512dq_fixuns_truncv2sfv2di2_mask, "__builtin_ia32_cvttps2uqq128_mask", IX86_BUILTIN_CVTTPS2UQQ128, UNKNOWN, (int) V2DI_FTYPE_V4SF_V2DI_UQI)
 BDESC (OPTION_MASK_ISA_AVX512VL, 0, CODE_FOR_fix_truncv8sfv8si2_mask, "__builtin_ia32_cvttps2dq256_mask", IX86_BUILTIN_CVTTPS2DQ256_MASK, UNKNOWN, (int) V8SI_FTYPE_V8SF_V8SI_UQI)
 BDESC (OPTION_MASK_ISA_AVX512VL, 0, CODE_FOR_fix_truncv4sfv4si2_mask, "__builtin_ia32_cvttps2dq128_mask", IX86_BUILTIN_CVTTPS2DQ128_MASK, UNKNOWN, (int) V4SI_FTYPE_V4SF_V4SI_UQI)
 BDESC (OPTION_MASK_ISA_AVX512VL, 0, CODE_FOR_ufix_truncv8sfv8si2_mask, "__builtin_ia32_cvttps2udq256_mask", IX86_BUILTIN_CVTTPS2UDQ256, UNKNOWN, (int) V8SI_FTYPE_V8SF_V8SI_UQI)

diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index cee2b2890a5d396ce66a22c1f23a85e6b9548d6b..fde65391d7d4279aa01e9656b63ef633ecfb679b 100644 (file)
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -5795,7 +5795,7 @@
    (set_attr "prefix" "evex")
    (set_attr "mode" "<MODE>")])
 
-(define_expand "float<floatunssuffix>v2div2sf2"
+(define_expand "avx512dq_float<floatunssuffix>v2div2sf2"
   [(set (match_operand:V4SF 0 "register_operand" "=v")
        (vec_concat:V4SF
            (any_float:V2SF (match_operand:V2DI 1 "nonimmediate_operand" "vm"))
@@ -5803,7 +5803,7 @@
   "TARGET_AVX512DQ && TARGET_AVX512VL"
   "operands[2] = CONST0_RTX (V2SFmode);")
 
-(define_insn "*float<floatunssuffix>v2div2sf2"
+(define_insn "*avx512dq_float<floatunssuffix>v2div2sf2"
   [(set (match_operand:V4SF 0 "register_operand" "=v")
        (vec_concat:V4SF
            (any_float:V2SF (match_operand:V2DI 1 "nonimmediate_operand" "vm"))
@@ -5814,6 +5814,17 @@
    (set_attr "prefix" "evex")
    (set_attr "mode" "V4SF")])
 
+(define_expand "float<floatunssuffix>v2div2sf2"
+  [(set (match_operand:V2SF 0 "register_operand")
+       (any_float:V2SF (match_operand:V2DI 1 "nonimmediate_operand")))]
+  "TARGET_AVX512DQ && TARGET_AVX512VL"
+{
+  operands[0] = simplify_gen_subreg (V4SFmode, operands[0], V2SFmode, 0);
+  emit_insn (gen_avx512dq_float<floatunssuffix>v2div2sf2
+            (operands[0], operands[1]));
+  DONE;
+})
+
 (define_mode_attr vpckfloat_concat_mode
   [(V8DI "v16sf") (V4DI "v8sf") (V2DI "v8sf")])
 (define_mode_attr vpckfloat_temp_mode
@@ -5830,7 +5841,12 @@
 {
   rtx r1 = gen_reg_rtx (<vpckfloat_temp_mode>mode);
   rtx r2 = gen_reg_rtx (<vpckfloat_temp_mode>mode);
-  rtx (*gen) (rtx, rtx) = gen_float<floatunssuffix><mode><vpckfloat_op_mode>2;
+  rtx (*gen) (rtx, rtx);
+
+  if (<MODE>mode == V2DImode)
+    gen = gen_avx512dq_float<floatunssuffix>v2div2sf2;
+  else
+    gen = gen_float<floatunssuffix><mode><vpckfloat_op_mode>2;
   emit_insn (gen (r1, operands[1]));
   emit_insn (gen (r2, operands[2]));
   if (<MODE>mode == V2DImode)
@@ -6217,7 +6233,7 @@
    (set_attr "prefix" "evex")
    (set_attr "mode" "<sseintvecmode3>")])
 
-(define_insn "fix<fixunssuffix>_truncv2sfv2di2<mask_name>"
+(define_insn "avx512dq_fix<fixunssuffix>_truncv2sfv2di2<mask_name>"
   [(set (match_operand:V2DI 0 "register_operand" "=v")
        (any_fix:V2DI
          (vec_select:V2SF
@@ -6229,6 +6245,18 @@
    (set_attr "prefix" "evex")
    (set_attr "mode" "TI")])
 
+(define_expand "fix<fixunssuffix>_truncv2sfv2di2"
+  [(set (match_operand:V2DI 0 "register_operand")
+       (any_fix:V2DI
+         (match_operand:V2SF 1 "register_operand")))]
+  "TARGET_AVX512DQ && TARGET_AVX512VL"
+{
+  operands[1] = simplify_gen_subreg (V4SFmode, operands[1], V2SFmode, 0);
+  emit_insn (gen_avx512dq_fix<fixunssuffix>_truncv2sfv2di2
+            (operands[0], operands[1]));
+  DONE;
+})
+
 (define_mode_attr vunpckfixt_mode
   [(V16SF "V8DI") (V8SF "V4DI") (V4SF "V2DI")])
 (define_mode_attr vunpckfixt_model
@@ -6243,14 +6271,18 @@
   "TARGET_AVX512DQ"
 {
   rtx tem = operands[1];
+  rtx (*gen) (rtx, rtx);
+
   if (<MODE>mode != V4SFmode)
     {
       tem = gen_reg_rtx (<ssehalfvecmode>mode);
       emit_insn (gen_vec_extract_lo_<vunpckfixt_extract_mode> (tem,
                                                               operands[1]));
+      gen = gen_fix<fixunssuffix>_trunc<ssehalfvecmodelower><vunpckfixt_model>2;
     }
-  rtx (*gen) (rtx, rtx)
-    = gen_fix<fixunssuffix>_trunc<ssehalfvecmodelower><vunpckfixt_model>2;
+  else
+    gen = gen_avx512dq_fix<fixunssuffix>_truncv2sfv2di2;
+
   emit_insn (gen (operands[0], tem));
   DONE;
 })
@@ -6262,19 +6294,22 @@
   "TARGET_AVX512DQ"
 {
   rtx tem;
+  rtx (*gen) (rtx, rtx);
+
   if (<MODE>mode != V4SFmode)
     {
       tem = gen_reg_rtx (<ssehalfvecmode>mode);
       emit_insn (gen_vec_extract_hi_<vunpckfixt_extract_mode> (tem,
                                                               operands[1]));
+      gen = gen_fix<fixunssuffix>_trunc<ssehalfvecmodelower><vunpckfixt_model>2;
     }
   else
     {
       tem = gen_reg_rtx (V4SFmode);
       emit_insn (gen_avx_vpermilv4sf (tem, operands[1], GEN_INT (0x4e)));
+      gen = gen_avx512dq_fix<fixunssuffix>_truncv2sfv2di2;
     }
-  rtx (*gen) (rtx, rtx)
-    = gen_fix<fixunssuffix>_trunc<ssehalfvecmodelower><vunpckfixt_model>2;
+
   emit_insn (gen (operands[0], tem));
   DONE;
 })

diff --git a/gcc/emit-rtl.c b/gcc/emit-rtl.c
index c49628040f84f7f076b98690332b6aa00dd40a38..2b79063636682ddd61605e64276f2bd484145ed7 100644 (file)
--- a/gcc/emit-rtl.c
+++ b/gcc/emit-rtl.c
@@ -939,11 +939,13 @@ validate_subreg (machine_mode omode, machine_mode imode,
           && GET_MODE_INNER (imode) == omode)
     ;
   /* ??? x86 sse code makes heavy use of *paradoxical* vector subregs,
-     i.e. (subreg:V4SF (reg:SF) 0).  This surely isn't the cleanest way to
-     represent this.  It's questionable if this ought to be represented at
-     all -- why can't this all be hidden in post-reload splitters that make
-     arbitrarily mode changes to the registers themselves.  */
-  else if (VECTOR_MODE_P (omode) && GET_MODE_INNER (omode) == imode)
+     i.e. (subreg:V4SF (reg:SF) 0) or (subreg:V4SF (reg:V2SF) 0).  This
+     surely isn't the cleanest way to represent this.  It's questionable
+     if this ought to be represented at all -- why can't this all be hidden
+     in post-reload splitters that make arbitrarily mode changes to the
+     registers themselves.  */
+  else if (VECTOR_MODE_P (omode)
+          && GET_MODE_INNER (omode) == GET_MODE_INNER (imode))
     ;
   /* Subregs involving floating point modes are not allowed to
      change size.  Therefore (subreg:DI (reg:DF) 0) is fine, but

diff --git a/gcc/testsuite/gcc.target/i386/pr95211.c b/gcc/testsuite/gcc.target/i386/pr95211.c
new file mode 100644 (file)
index 0000000..dc10f8f
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr95211.c
@@ -0,0 +1,38 @@
+/* PR target/95211 target/95256 */
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-options "-O2 -ftree-slp-vectorize -march=skylake-avx512" } */
+
+extern float f[4];
+extern long long l[2];
+extern long long ul[2];
+
+void
+fix_128 (void)
+{
+  l[0] = f[0];
+  l[1] = f[1];
+}
+
+void
+fixuns_128 (void)
+{
+  ul[0] = f[0];
+  ul[1] = f[1];
+}
+
+void
+float_128 (void)
+{
+  f[0] = l[0];
+  f[1] = l[1];
+}
+
+void
+floatuns_128 (void)
+{
+  f[0] = ul[0];
+  f[1] = ul[1];
+}
+
+/* { dg-final { scan-assembler-times "vcvttps2qq" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2ps" 2 } } */