[AArch64][SVE2] Support for EOR3 and variants of BSL

2019-10-17  Yuliang Wang  <yuliang.wang@arm.com>

gcc/
	* config/aarch64/aarch64-sve2.md (aarch64_sve2_eor3<mode>)
	(aarch64_sve2_nor<mode>, aarch64_sve2_nand<mode>)
	(aarch64_sve2_bsl<mode>, aarch64_sve2_nbsl<mode>)
	(aarch64_sve2_bsl1n<mode>, aarch64_sve2_bsl2n<mode>):
	New combine patterns.
	* config/aarch64/iterators.md (BSL_DUP): New int iterator for the
	above.
	(bsl_1st, bsl_2nd, bsl_dup, bsl_mov): Attributes for the above.

gcc/testsuite/
	* gcc.target/aarch64/sve2/eor3_1.c: New test.
	* gcc.target/aarch64/sve2/nlogic_1.c: As above.
	* gcc.target/aarch64/sve2/nlogic_2.c: As above.
	* gcc.target/aarch64/sve2/bitsel_1.c: As above.
	* gcc.target/aarch64/sve2/bitsel_2.c: As above.
	* gcc.target/aarch64/sve2/bitsel_3.c: As above.
	* gcc.target/aarch64/sve2/bitsel_4.c: As above.


git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@277110 138bc75d-0d04-0410-961f-82ee72b054a4

11 files changed, 351 insertions, 0 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index efbc5c6a8a1..7eb331d60f5 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,14 @@
+2019-10-17  Yuliang Wang  <yuliang.wang@arm.com>
+
+	* config/aarch64/aarch64-sve2.md (aarch64_sve2_eor3<mode>)
+	(aarch64_sve2_nor<mode>, aarch64_sve2_nand<mode>)
+	(aarch64_sve2_bsl<mode>, aarch64_sve2_nbsl<mode>)
+	(aarch64_sve2_bsl1n<mode>, aarch64_sve2_bsl2n<mode>):
+	New combine patterns.
+	* config/aarch64/iterators.md (BSL_DUP): New int iterator for the
+	above.
+	(bsl_1st, bsl_2nd, bsl_dup, bsl_mov): Attributes for the above.
+
 2019-10-17  Aldy Hernandez  <aldyh@redhat.com>
 
 	* tree-vrp.c (value_range_base::dump): Display +INF for both
diff --git a/gcc/config/aarch64/aarch64-sve2.md b/gcc/config/aarch64/aarch64-sve2.md
index b018f5b0bc9..ecbee9733f0 100644
--- a/gcc/config/aarch64/aarch64-sve2.md
+++ b/gcc/config/aarch64/aarch64-sve2.md
@@ -142,3 +142,187 @@
   }
 )
 
+;; Unpredicated 3-way exclusive OR.
+(define_insn "*aarch64_sve2_eor3<mode>"
+  [(set (match_operand:SVE_I 0 "register_operand" "=w, w, w, ?&w")
+	(xor:SVE_I
+	  (xor:SVE_I
+	    (match_operand:SVE_I 1 "register_operand" "0, w, w, w")
+	    (match_operand:SVE_I 2 "register_operand" "w, 0, w, w"))
+	  (match_operand:SVE_I 3 "register_operand" "w, w, 0, w")))]
+  "TARGET_SVE2"
+  "@
+  eor3\t%0.d, %0.d, %2.d, %3.d
+  eor3\t%0.d, %0.d, %1.d, %3.d
+  eor3\t%0.d, %0.d, %1.d, %2.d
+  movprfx\t%0, %1\;eor3\t%0.d, %0.d, %2.d, %3.d"
+  [(set_attr "movprfx" "*,*,*,yes")]
+)
+
+;; Use NBSL for vector NOR.
+(define_insn_and_rewrite "*aarch64_sve2_nor<mode>"
+  [(set (match_operand:SVE_I 0 "register_operand" "=w, ?&w")
+	(unspec:SVE_I
+	  [(match_operand 3)
+	   (and:SVE_I
+	     (not:SVE_I
+	       (match_operand:SVE_I 1 "register_operand" "%0, w"))
+	     (not:SVE_I
+	       (match_operand:SVE_I 2 "register_operand" "w, w")))]
+	  UNSPEC_PRED_X))]
+  "TARGET_SVE2"
+  "@
+  nbsl\t%0.d, %0.d, %2.d, %0.d
+  movprfx\t%0, %1\;nbsl\t%0.d, %0.d, %2.d, %0.d"
+  "&& !CONSTANT_P (operands[3])"
+  {
+    operands[3] = CONSTM1_RTX (<VPRED>mode);
+  }
+  [(set_attr "movprfx" "*,yes")]
+)
+
+;; Use NBSL for vector NAND.
+(define_insn_and_rewrite "*aarch64_sve2_nand<mode>"
+  [(set (match_operand:SVE_I 0 "register_operand" "=w, ?&w")
+	(unspec:SVE_I
+	  [(match_operand 3)
+	   (ior:SVE_I
+	     (not:SVE_I
+	       (match_operand:SVE_I 1 "register_operand" "%0, w"))
+	     (not:SVE_I
+	       (match_operand:SVE_I 2 "register_operand" "w, w")))]
+	  UNSPEC_PRED_X))]
+  "TARGET_SVE2"
+  "@
+  nbsl\t%0.d, %0.d, %2.d, %2.d
+  movprfx\t%0, %1\;nbsl\t%0.d, %0.d, %2.d, %2.d"
+  "&& !CONSTANT_P (operands[3])"
+  {
+    operands[3] = CONSTM1_RTX (<VPRED>mode);
+  }
+  [(set_attr "movprfx" "*,yes")]
+)
+
+;; Unpredicated bitwise select.
+;; (op3 ? bsl_mov : bsl_dup) == (((bsl_mov ^ bsl_dup) & op3) ^ bsl_dup)
+(define_insn "*aarch64_sve2_bsl<mode>"
+  [(set (match_operand:SVE_I 0 "register_operand" "=w, ?&w")
+	(xor:SVE_I
+	  (and:SVE_I
+	    (xor:SVE_I
+	      (match_operand:SVE_I 1 "register_operand" "<bsl_1st>, w")
+	      (match_operand:SVE_I 2 "register_operand" "<bsl_2nd>, w"))
+	    (match_operand:SVE_I 3 "register_operand" "w, w"))
+	  (match_dup BSL_DUP)))]
+  "TARGET_SVE2"
+  "@
+  bsl\t%0.d, %0.d, %<bsl_dup>.d, %3.d
+  movprfx\t%0, %<bsl_mov>\;bsl\t%0.d, %0.d, %<bsl_dup>.d, %3.d"
+  [(set_attr "movprfx" "*,yes")]
+)
+
+;; Unpredicated bitwise inverted select.
+;; (~(op3 ? bsl_mov : bsl_dup)) == (~(((bsl_mov ^ bsl_dup) & op3) ^ bsl_dup))
+(define_insn_and_rewrite "*aarch64_sve2_nbsl<mode>"
+  [(set (match_operand:SVE_I 0 "register_operand" "=w, ?&w")
+	(unspec:SVE_I
+	  [(match_operand 4)
+	   (not:SVE_I
+	     (xor:SVE_I
+	       (and:SVE_I
+		 (xor:SVE_I
+		   (match_operand:SVE_I 1 "register_operand" "<bsl_1st>, w")
+		   (match_operand:SVE_I 2 "register_operand" "<bsl_2nd>, w"))
+		 (match_operand:SVE_I 3 "register_operand" "w, w"))
+	       (match_dup BSL_DUP)))]
+	  UNSPEC_PRED_X))]
+  "TARGET_SVE2"
+  "@
+  nbsl\t%0.d, %0.d, %<bsl_dup>.d, %3.d
+  movprfx\t%0, %<bsl_mov>\;nbsl\t%0.d, %0.d, %<bsl_dup>.d, %3.d"
+  "&& !CONSTANT_P (operands[4])"
+  {
+    operands[4] = CONSTM1_RTX (<VPRED>mode);
+  }
+  [(set_attr "movprfx" "*,yes")]
+)
+
+;; Unpredicated bitwise select with inverted first operand.
+;; (op3 ? ~bsl_mov : bsl_dup) == ((~(bsl_mov ^ bsl_dup) & op3) ^ bsl_dup)
+(define_insn_and_rewrite "*aarch64_sve2_bsl1n<mode>"
+  [(set (match_operand:SVE_I 0 "register_operand" "=w, ?&w")
+	(xor:SVE_I
+	  (and:SVE_I
+	    (unspec:SVE_I
+	      [(match_operand 4)
+	       (not:SVE_I
+		 (xor:SVE_I
+		   (match_operand:SVE_I 1 "register_operand" "<bsl_1st>, w")
+		   (match_operand:SVE_I 2 "register_operand" "<bsl_2nd>, w")))]
+	      UNSPEC_PRED_X)
+	    (match_operand:SVE_I 3 "register_operand" "w, w"))
+	  (match_dup BSL_DUP)))]
+  "TARGET_SVE2"
+  "@
+  bsl1n\t%0.d, %0.d, %<bsl_dup>.d, %3.d
+  movprfx\t%0, %<bsl_mov>\;bsl1n\t%0.d, %0.d, %<bsl_dup>.d, %3.d"
+  "&& !CONSTANT_P (operands[4])"
+  {
+    operands[4] = CONSTM1_RTX (<VPRED>mode);
+  }
+  [(set_attr "movprfx" "*,yes")]
+)
+
+;; Unpredicated bitwise select with inverted second operand.
+;; (bsl_dup ? bsl_mov : ~op3) == ((bsl_dup & bsl_mov) | (~op3 & ~bsl_dup))
+(define_insn_and_rewrite "*aarch64_sve2_bsl2n<mode>"
+  [(set (match_operand:SVE_I 0 "register_operand" "=w, ?&w")
+	(ior:SVE_I
+	  (and:SVE_I
+	    (match_operand:SVE_I 1 "register_operand" "<bsl_1st>, w")
+	    (match_operand:SVE_I 2 "register_operand" "<bsl_2nd>, w"))
+	  (unspec:SVE_I
+	    [(match_operand 4)
+	     (and:SVE_I
+	       (not:SVE_I
+		 (match_operand:SVE_I 3 "register_operand" "w, w"))
+	       (not:SVE_I
+		 (match_dup BSL_DUP)))]
+	    UNSPEC_PRED_X)))]
+  "TARGET_SVE2"
+  "@
+  bsl2n\t%0.d, %0.d, %3.d, %<bsl_dup>.d
+  movprfx\t%0, %<bsl_mov>\;bsl2n\t%0.d, %0.d, %3.d, %<bsl_dup>.d"
+  "&& !CONSTANT_P (operands[4])"
+  {
+    operands[4] = CONSTM1_RTX (<VPRED>mode);
+  }
+  [(set_attr "movprfx" "*,yes")]
+)
+
+;; Unpredicated bitwise select with inverted second operand, alternative form.
+;; (bsl_dup ? bsl_mov : ~op3) == ((bsl_dup & bsl_mov) | (~bsl_dup & ~op3))
+(define_insn_and_rewrite "*aarch64_sve2_bsl2n<mode>"
+  [(set (match_operand:SVE_I 0 "register_operand" "=w, ?&w")
+	(ior:SVE_I
+	  (and:SVE_I
+	    (match_operand:SVE_I 1 "register_operand" "<bsl_1st>, w")
+	    (match_operand:SVE_I 2 "register_operand" "<bsl_2nd>, w"))
+	  (unspec:SVE_I
+	    [(match_operand 4)
+	     (and:SVE_I
+	       (not:SVE_I
+		 (match_dup BSL_DUP))
+	       (not:SVE_I
+		 (match_operand:SVE_I 3 "register_operand" "w, w")))]
+	    UNSPEC_PRED_X)))]
+  "TARGET_SVE2"
+  "@
+  bsl2n\t%0.d, %0.d, %3.d, %<bsl_dup>.d
+  movprfx\t%0, %<bsl_mov>\;bsl2n\t%0.d, %0.d, %3.d, %<bsl_dup>.d"
+  "&& !CONSTANT_P (operands[4])"
+  {
+    operands[4] = CONSTM1_RTX (<VPRED>mode);
+  }
+  [(set_attr "movprfx" "*,yes")]
+)
diff --git a/gcc/config/aarch64/iterators.md b/gcc/config/aarch64/iterators.md
index 1e321af710b..f879fadb007 100644
--- a/gcc/config/aarch64/iterators.md
+++ b/gcc/config/aarch64/iterators.md
@@ -1611,6 +1611,8 @@
 
 (define_int_iterator SHRNT [UNSPEC_SHRNT UNSPEC_RSHRNT])
 
+(define_int_iterator BSL_DUP [1 2])
+
 (define_int_iterator DOTPROD [UNSPEC_SDOT UNSPEC_UDOT])
 
 (define_int_iterator ADDSUBHN [UNSPEC_ADDHN UNSPEC_RADDHN
@@ -1976,6 +1978,18 @@
 			 (UNSPEC_RADDHN2 "add")
 			 (UNSPEC_RSUBHN2 "sub")])
 
+;; BSL variants: first commutative operand.
+(define_int_attr bsl_1st [(1 "w") (2 "0")])
+
+;; BSL variants: second commutative operand.
+(define_int_attr bsl_2nd [(1 "0") (2 "w")])
+
+;; BSL variants: duplicated input operand.
+(define_int_attr bsl_dup [(1 "1") (2 "2")])
+
+;; BSL variants: operand which requires preserving via movprfx.
+(define_int_attr bsl_mov [(1 "2") (2 "1")])
+
 (define_int_attr offsetlr [(UNSPEC_SSLI "") (UNSPEC_USLI "")
 			   (UNSPEC_SSRI "offset_")
 			   (UNSPEC_USRI "offset_")])
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index abd9dd1c2dd..5ea4a731243 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,13 @@
+2019-10-17  Yuliang Wang  <yuliang.wang@arm.com>
+
+	* gcc.target/aarch64/sve2/eor3_1.c: New test.
+	* gcc.target/aarch64/sve2/nlogic_1.c: As above.
+	* gcc.target/aarch64/sve2/nlogic_2.c: As above.
+	* gcc.target/aarch64/sve2/bitsel_1.c: As above.
+	* gcc.target/aarch64/sve2/bitsel_2.c: As above.
+	* gcc.target/aarch64/sve2/bitsel_3.c: As above.
+	* gcc.target/aarch64/sve2/bitsel_4.c: As above.
+
 2019-10-17  Aldy Hernandez  <aldyh@redhat.com>
 
 	* gcc.dg/tree-ssa/evrp4.c: Check for +INF instead of -1.
diff --git a/gcc/testsuite/gcc.target/aarch64/sve2/bitsel_1.c b/gcc/testsuite/gcc.target/aarch64/sve2/bitsel_1.c
new file mode 100644
index 00000000000..629f74167a0
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/sve2/bitsel_1.c
@@ -0,0 +1,32 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -ftree-vectorize -fdump-tree-vect-details --save-temps" } */
+
+#include <stdint.h>
+
+#ifndef OP
+#define OP(x,y,z) (((x) & (z)) | ((y) & ~(z)))
+#endif
+
+#define TYPE(N) int##N##_t
+
+#define TEMPLATE(SIZE)						\
+void __attribute__ ((noinline, noclone))			\
+f_##SIZE##_##OP							\
+  (TYPE(SIZE) *restrict a, TYPE(SIZE) *restrict b,		\
+   TYPE(SIZE) *restrict c, TYPE(SIZE) *restrict d, int n)	\
+{								\
+  for (int i = 0; i < n; i++)					\
+    a[i] = OP (b[i], c[i], d[i]);				\
+}
+
+TEMPLATE (8);
+TEMPLATE (16);
+TEMPLATE (32);
+TEMPLATE (64);
+
+/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
+
+/* { dg-final { scan-assembler-not {\teor\tz[0-9]+\.[bhsd]} } } */
+/* { dg-final { scan-assembler-not {\tand\tz[0-9]+\.[bhsd]} } } */
+
+/* { dg-final { scan-assembler-times {\tbsl\tz[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d\n} 4 } } */
diff --git a/gcc/testsuite/gcc.target/aarch64/sve2/bitsel_2.c b/gcc/testsuite/gcc.target/aarch64/sve2/bitsel_2.c
new file mode 100644
index 00000000000..ee2d4a35a1d
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/sve2/bitsel_2.c
@@ -0,0 +1,14 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -ftree-vectorize -fdump-tree-vect-details --save-temps" } */
+
+#define OP(x,y,z) (~(((x) & (z)) | ((y) & ~(z))))
+
+#include "bitsel_1.c"
+
+/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
+
+/* { dg-final { scan-assembler-not {\teor\tz[0-9]+\.[bhsd]} } } */
+/* { dg-final { scan-assembler-not {\tand\tz[0-9]+\.[bhsd]} } } */
+/* { dg-final { scan-assembler-not {\tnot\tz[0-9]+\.[bhsd]} } } */
+
+/* { dg-final { scan-assembler-times {\tnbsl\tz[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d\n} 4 } } */
diff --git a/gcc/testsuite/gcc.target/aarch64/sve2/bitsel_3.c b/gcc/testsuite/gcc.target/aarch64/sve2/bitsel_3.c
new file mode 100644
index 00000000000..d0dc713d92c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/sve2/bitsel_3.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -ftree-vectorize -fdump-tree-vect-details --save-temps" } */
+
+#define OP(x,y,z) ((~(x) & (z)) | ((y) & ~(z)))
+
+#include "bitsel_1.c"
+
+/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
+
+/* { dg-final { scan-assembler-not {\teor\tz[0-9]+\.[bhsd]} } } */
+/* { dg-final { scan-assembler-not {\tbic\tz[0-9]+\.[bhsd]} } } */
+
+/* { dg-final { scan-assembler-times {\tbsl1n\tz[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d\n} 4 } } */
diff --git a/gcc/testsuite/gcc.target/aarch64/sve2/bitsel_4.c b/gcc/testsuite/gcc.target/aarch64/sve2/bitsel_4.c
new file mode 100644
index 00000000000..5eb71c93ae7
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/sve2/bitsel_4.c
@@ -0,0 +1,14 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -ftree-vectorize -fdump-tree-vect-details --save-temps" } */
+
+#define OP(x,y,z) (((x) & (z)) | (~(y) & ~(z)))
+
+#include "bitsel_1.c"
+
+/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
+
+/* { dg-final { scan-assembler-not {\torr\tz[0-9]+\.[bhsd]} } } */
+/* { dg-final { scan-assembler-not {\tand\tz[0-9]+\.[bhsd]} } } */
+/* { dg-final { scan-assembler-not {\tnot\tz[0-9]+\.[bhsd]} } } */
+
+/* { dg-final { scan-assembler-times {\tbsl2n\tz[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d\n} 4 } } */
diff --git a/gcc/testsuite/gcc.target/aarch64/sve2/eor3_1.c b/gcc/testsuite/gcc.target/aarch64/sve2/eor3_1.c
new file mode 100644
index 00000000000..13df93e56b1
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/sve2/eor3_1.c
@@ -0,0 +1,12 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -ftree-vectorize -fdump-tree-vect-details --save-temps" } */
+
+#define OP(x,y,z) ((x) ^ (y) ^ (z))
+
+#include "bitsel_1.c"
+
+/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
+
+/* { dg-final { scan-assembler-not {\teor\tz[0-9]+\.[bhsd]} } } */
+
+/* { dg-final { scan-assembler-times {\teor3\tz[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d\n} 4 } } */
diff --git a/gcc/testsuite/gcc.target/aarch64/sve2/nlogic_1.c b/gcc/testsuite/gcc.target/aarch64/sve2/nlogic_1.c
new file mode 100644
index 00000000000..de34b6d817a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/sve2/nlogic_1.c
@@ -0,0 +1,33 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -ftree-vectorize -fdump-tree-vect-details --save-temps" } */
+
+#include <stdint.h>
+
+#ifndef OP
+#define OP(x,y) (~((x) | (y)))
+#endif
+
+#define TYPE(N) int##N##_t
+
+#define TEMPLATE(SIZE)					\
+void __attribute__ ((noinline, noclone))		\
+f_##SIZE##_##OP						\
+  (TYPE(SIZE) *restrict a, TYPE(SIZE) *restrict b,	\
+   TYPE(SIZE) *restrict c, int n)			\
+{							\
+  for (int i = 0; i < n; i++)				\
+    a[i] = OP (b[i], c[i]);				\
+}
+
+TEMPLATE (8);
+TEMPLATE (16);
+TEMPLATE (32);
+TEMPLATE (64);
+
+/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
+
+/* { dg-final { scan-assembler-not {\tand\tz[0-9]+\.[bhsd]} } } */
+/* { dg-final { scan-assembler-not {\torr\tz[0-9]+\.[bhsd]} } } */
+/* { dg-final { scan-assembler-not {\tnot\tz[0-9]+\.[bhsd]} } } */
+
+/* { dg-final { scan-assembler-times {\tnbsl\tz[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d\n} 4 } } */
diff --git a/gcc/testsuite/gcc.target/aarch64/sve2/nlogic_2.c b/gcc/testsuite/gcc.target/aarch64/sve2/nlogic_2.c
new file mode 100644
index 00000000000..14400b5713e
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/sve2/nlogic_2.c
@@ -0,0 +1,14 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -ftree-vectorize -fdump-tree-vect-details --save-temps" } */
+
+#define OP(x,y) (~((x) & (y)))
+
+#include "nlogic_1.c"
+
+/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
+
+/* { dg-final { scan-assembler-not {\tand\tz[0-9]+\.[bhsd]} } } */
+/* { dg-final { scan-assembler-not {\torr\tz[0-9]+\.[bhsd]} } } */
+/* { dg-final { scan-assembler-not {\tnot\tz[0-9]+\.[bhsd]} } } */
+
+/* { dg-final { scan-assembler-times {\tnbsl\tz[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d\n} 4 } } */