i386: Optimize psubusw compared to 0 into pminuw compared to op0 [PR96906]

The following patch renames VI12_AVX2 iterator to VI12_AVX2_AVX512BW
for consistency with some other iterators, as I need VI12_AVX2 without
AVX512BW for this change.
The real meat is a combiner split which combine
can use to optimize psubusw compared to 0 into pminuw compared to op0
(and similarly for psubusb compared to 0 into pminub compared to op0).
According to Agner Fog's tables, psubus[bw] and pminu[bw] timings
are the same, but the advantage of pminu[bw] is that the comparison
doesn't need a zero operand, so e.g. for -msse4.1 it causes changes like
-       psubusw %xmm1, %xmm0
-       pxor    %xmm1, %xmm1
+       pminuw  %xmm0, %xmm1
        pcmpeqw %xmm1, %xmm0
and similarly for avx2:
-       vpsubusb        %ymm1, %ymm0, %ymm0
-       vpxor   %xmm1, %xmm1, %xmm1
-       vpcmpeqb        %ymm1, %ymm0, %ymm0
+       vpminub %ymm1, %ymm0, %ymm1
+       vpcmpeqb        %ymm0, %ymm1, %ymm0

I haven't done the AVX512{BW,VL} define_split(s), they'll need
to match the UNSPEC_PCMP which are used for avx512 comparisons.

2020-11-26  Jakub Jelinek  <jakub@redhat.com>

	PR target/96906
	* config/i386/sse.md (VI12_AVX2): Remove V64QI/V32HI modes.
	(VI12_AVX2_AVX512BW): New mode iterator.
	(<sse2_avx2>_<plusminus_insn><mode>3<mask_name>,
	uavg<mode>3_ceil, <sse2_avx2>_uavg<mode>3<mask_name>): Use
	VI12_AVX2_AVX512BW iterator instead of VI12_AVX2.
	(*<sse2_avx2>_<plusminus_insn><mode>3<mask_name>): Likewise.
	(*<sse2_avx2>_uavg<mode>3<mask_name>): Likewise.
	(*<sse2_avx2>_<plusminus_insn><mode>3<mask_name>): Add a new
	define_split after this insn.

	* gcc.target/i386/pr96906-1.c: New test.

2 files changed, 103 insertions, 20 deletions

diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index 1193680..4aad462 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -466,6 +466,10 @@
   [(V4TI "TARGET_AVX512BW") (V2TI "TARGET_AVX2") TI])
 
 (define_mode_iterator VI12_AVX2
+  [(V32QI "TARGET_AVX2") V16QI
+   (V16HI "TARGET_AVX2") V8HI])
+
+(define_mode_iterator VI12_AVX2_AVX512BW
   [(V64QI "TARGET_AVX512BW") (V32QI "TARGET_AVX2") V16QI
    (V32HI "TARGET_AVX512BW") (V16HI "TARGET_AVX2") V8HI])
 
@@ -11395,18 +11399,18 @@
    (set_attr "mode" "<sseinsnmode>")])
 
 (define_expand "<sse2_avx2>_<plusminus_insn><mode>3<mask_name>"
-  [(set (match_operand:VI12_AVX2 0 "register_operand")
-	(sat_plusminus:VI12_AVX2
-	  (match_operand:VI12_AVX2 1 "vector_operand")
-	  (match_operand:VI12_AVX2 2 "vector_operand")))]
+  [(set (match_operand:VI12_AVX2_AVX512BW 0 "register_operand")
+	(sat_plusminus:VI12_AVX2_AVX512BW
+	  (match_operand:VI12_AVX2_AVX512BW 1 "vector_operand")
+	  (match_operand:VI12_AVX2_AVX512BW 2 "vector_operand")))]
   "TARGET_SSE2 && <mask_mode512bit_condition> && <mask_avx512bw_condition>"
   "ix86_fixup_binary_operands_no_copy (<CODE>, <MODE>mode, operands);")
 
 (define_insn "*<sse2_avx2>_<plusminus_insn><mode>3<mask_name>"
-  [(set (match_operand:VI12_AVX2 0 "register_operand" "=x,v")
-	(sat_plusminus:VI12_AVX2
-	  (match_operand:VI12_AVX2 1 "vector_operand" "<comm>0,v")
-	  (match_operand:VI12_AVX2 2 "vector_operand" "xBm,vm")))]
+  [(set (match_operand:VI12_AVX2_AVX512BW 0 "register_operand" "=x,v")
+	(sat_plusminus:VI12_AVX2_AVX512BW
+	  (match_operand:VI12_AVX2_AVX512BW 1 "vector_operand" "<comm>0,v")
+	  (match_operand:VI12_AVX2_AVX512BW 2 "vector_operand" "xBm,vm")))]
   "TARGET_SSE2 && <mask_mode512bit_condition> && <mask_avx512bw_condition>
    && ix86_binary_operator_ok (<CODE>, <MODE>mode, operands)"
   "@
@@ -11418,6 +11422,23 @@
    (set_attr "prefix" "orig,maybe_evex")
    (set_attr "mode" "TI")])
 
+;; PR96906 - optimize psubusw compared to 0 into pminuw compared to op0.
+(define_split
+  [(set (match_operand:VI12_AVX2 0 "register_operand")
+	(eq:VI12_AVX2
+	  (us_minus:VI12_AVX2
+	    (match_operand:VI12_AVX2 1 "vector_operand")
+	    (match_operand:VI12_AVX2 2 "vector_operand"))
+	  (match_operand:VI12_AVX2 3 "const0_operand")))]
+  "TARGET_SSE2
+   && (<MODE>mode != V8HImode || TARGET_SSE4_1)
+   && ix86_binary_operator_ok (US_MINUS, <MODE>mode, operands)"
+  [(set (match_dup 4)
+        (umin:VI12_AVX2 (match_dup 1) (match_dup 2)))
+   (set (match_dup 0)
+	(eq:VI12_AVX2 (match_dup 4) (match_dup 1)))]
+  "operands[4] = gen_reg_rtx (<MODE>mode);")
+
 (define_expand "mulv8qi3"
   [(set (match_operand:V8QI 0 "register_operand")
 	(mult:V8QI (match_operand:V8QI 1 "register_operand")
@@ -12022,15 +12043,15 @@
 })
 
 (define_expand "uavg<mode>3_ceil"
-  [(set (match_operand:VI12_AVX2 0 "register_operand")
-	(truncate:VI12_AVX2
+  [(set (match_operand:VI12_AVX2_AVX512BW 0 "register_operand")
+	(truncate:VI12_AVX2_AVX512BW
 	  (lshiftrt:<ssedoublemode>
 	    (plus:<ssedoublemode>
 	      (plus:<ssedoublemode>
 		(zero_extend:<ssedoublemode>
-		  (match_operand:VI12_AVX2 1 "vector_operand"))
+		  (match_operand:VI12_AVX2_AVX512BW 1 "vector_operand"))
 		(zero_extend:<ssedoublemode>
-		  (match_operand:VI12_AVX2 2 "vector_operand")))
+		  (match_operand:VI12_AVX2_AVX512BW 2 "vector_operand")))
 	      (match_dup 3))
 	    (const_int 1))))]
   "TARGET_SSE2"
@@ -15744,15 +15765,15 @@
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 
 (define_expand "<sse2_avx2>_uavg<mode>3<mask_name>"
-  [(set (match_operand:VI12_AVX2 0 "register_operand")
-	(truncate:VI12_AVX2
+  [(set (match_operand:VI12_AVX2_AVX512BW 0 "register_operand")
+	(truncate:VI12_AVX2_AVX512BW
 	  (lshiftrt:<ssedoublemode>
 	    (plus:<ssedoublemode>
 	      (plus:<ssedoublemode>
 		(zero_extend:<ssedoublemode>
-		  (match_operand:VI12_AVX2 1 "vector_operand"))
+		  (match_operand:VI12_AVX2_AVX512BW 1 "vector_operand"))
 		(zero_extend:<ssedoublemode>
-		  (match_operand:VI12_AVX2 2 "vector_operand")))
+		  (match_operand:VI12_AVX2_AVX512BW 2 "vector_operand")))
 	      (match_dup <mask_expand_op3>))
 	    (const_int 1))))]
   "TARGET_SSE2 && <mask_mode512bit_condition> && <mask_avx512bw_condition>"
@@ -15762,15 +15783,15 @@
 })
 
 (define_insn "*<sse2_avx2>_uavg<mode>3<mask_name>"
-  [(set (match_operand:VI12_AVX2 0 "register_operand" "=x,v")
-	(truncate:VI12_AVX2
+  [(set (match_operand:VI12_AVX2_AVX512BW 0 "register_operand" "=x,v")
+	(truncate:VI12_AVX2_AVX512BW
 	  (lshiftrt:<ssedoublemode>
 	    (plus:<ssedoublemode>
 	      (plus:<ssedoublemode>
 		(zero_extend:<ssedoublemode>
-		  (match_operand:VI12_AVX2 1 "vector_operand" "%0,v"))
+		  (match_operand:VI12_AVX2_AVX512BW 1 "vector_operand" "%0,v"))
 		(zero_extend:<ssedoublemode>
-		  (match_operand:VI12_AVX2 2 "vector_operand" "xBm,vm")))
+		  (match_operand:VI12_AVX2_AVX512BW 2 "vector_operand" "xBm,vm")))
 	      (match_operand:<ssedoublemode> <mask_expand_op3> "const1_operand"))
 	    (const_int 1))))]
   "TARGET_SSE2 && <mask_mode512bit_condition> && <mask_avx512bw_condition>
diff --git a/gcc/testsuite/gcc.target/i386/pr96906-1.c b/gcc/testsuite/gcc.target/i386/pr96906-1.c
new file mode 100644
index 0000000..9d836eb
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr96906-1.c
@@ -0,0 +1,62 @@
+/* PR target/96906 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx2" } */
+/* { dg-final { scan-assembler-times "\tvpminub\[^\n\r]*xmm" 2 } } */
+/* { dg-final { scan-assembler-times "\tvpminuw\[^\n\r]*xmm" 2 } } */
+/* { dg-final { scan-assembler-times "\tvpminub\[^\n\r]*ymm" 2 } } */
+/* { dg-final { scan-assembler-times "\tvpminuw\[^\n\r]*ymm" 2 } } */
+/* { dg-final { scan-assembler-times "\tvpcmpeqb\[^\n\r]*xmm" 2 } } */
+/* { dg-final { scan-assembler-times "\tvpcmpeqw\[^\n\r]*xmm" 2 } } */
+/* { dg-final { scan-assembler-times "\tvpcmpeqb\[^\n\r]*ymm" 2 } } */
+/* { dg-final { scan-assembler-times "\tvpcmpeqw\[^\n\r]*ymm" 2 } } */
+/* { dg-final { scan-assembler-not "\tvpsubus\[bw]" } } */
+
+#include <x86intrin.h>
+
+__m128i
+f1 (__m128i x, __m128i y)
+{
+  return _mm_cmpeq_epi16 (_mm_subs_epu16 (x, y), _mm_setzero_si128 ());
+}
+
+__m128i
+f2 (__m128i x, __m128i y)
+{
+  return _mm_cmpeq_epi16 (_mm_min_epu16 (x, y), x);
+}
+
+__m128i
+f3 (__m128i x, __m128i y)
+{
+  return _mm_cmpeq_epi8 (_mm_subs_epu8 (x, y), _mm_setzero_si128 ());
+}
+
+__m128i
+f4 (__m128i x, __m128i y)
+{
+  return _mm_cmpeq_epi8 (_mm_min_epu8 (x, y), x);
+}
+
+__m256i
+f5 (__m256i x, __m256i y)
+{
+  return _mm256_cmpeq_epi16 (_mm256_subs_epu16 (x, y), _mm256_setzero_si256 ());
+}
+
+__m256i
+f6 (__m256i x, __m256i y)
+{
+  return _mm256_cmpeq_epi16 (_mm256_min_epu16 (x, y), x);
+}
+
+__m256i
+f7 (__m256i x, __m256i y)
+{
+  return _mm256_cmpeq_epi8 (_mm256_subs_epu8 (x, y), _mm256_setzero_si256 ());
+}
+
+__m256i
+f8 (__m256i x, __m256i y)
+{
+  return _mm256_cmpeq_epi8 (_mm256_min_epu8 (x, y), x);
+}