[AArch64] Predicated SVE comparison folds

This patch adds SVE patterns that combine a PTRUE-predicated
comparison with a separate AND.  The main benefit is for
optimising ANDs with the loop predicate, as in the testcase.
However, one of the potential drawbacks is that it triggers
even for cases in which two naturally-parallel comparisons
are ANDed together.  Whether that's a win or a less will
depend on the schedule, but it has the potential to be a win
more often than a loss.

The combine patterns are undeniably ugly.  One way of getting
around them would be to allow 1->1 "splits" when combining
2 instructions, as well as 1->2 splits when combining more
than 2 instructions (although that wouldn't really be a split).
Another would be to have a way of defining target-specific
rtx simplifications.  branches/ARM/sve-branch has a prototype
implementation of that, but it would need some clean-up before being
ready to submit.  It would also be good to make it closer to the
match.pd style.

Until then, I think what the combine patterns are doing is the
"correct" implementation given the current infrastructure.

2018-05-08  Richard Sandiford  <richard.sandiford@linaro.org>
	    Alan Hayward  <alan.hayward@arm.com>
	    David Sherwood  <david.sherwood@arm.com>

gcc/
	* config/aarch64/aarch64-sve.md (*pred_cmp<cmp_op><mode>_combine)
	(*pred_cmp<cmp_op><mode>, *fcm<cmp_op><mode>_and_combine)
	(*fcmuo<mode>_and_combine, *fcm<cmp_op><mode>_and)
	(*fcmuo<mode>_and): New patterns.

gcc/testsuite/
	* gcc.target/aarch64/sve/vcond_6.c: Do not expect any ANDs.
	XFAIL the BIC test.
	* gcc.target/aarch64/sve/vcond_7.c: New test.
	* gcc.target/aarch64/sve/vcond_7_run.c: Likewise.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>

From-SVN: r260031

6 files changed, 402 insertions, 2 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index cf85cc8..748cf93 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,4 +1,13 @@
 2018-05-08  Richard Sandiford  <richard.sandiford@linaro.org>
+	    Alan Hayward  <alan.hayward@arm.com>
+	    David Sherwood  <david.sherwood@arm.com>
+
+	* config/aarch64/aarch64-sve.md (*pred_cmp<cmp_op><mode>_combine)
+	(*pred_cmp<cmp_op><mode>, *fcm<cmp_op><mode>_and_combine)
+	(*fcmuo<mode>_and_combine, *fcm<cmp_op><mode>_and)
+	(*fcmuo<mode>_and): New patterns.
+
+2018-05-08  Richard Sandiford  <richard.sandiford@linaro.org>
 
 	* config/aarch64/iterators.md (UNSPEC_COND_LO, UNSPEC_COND_LS)
 	(UNSPEC_COND_HI, UNSPEC_COND_HS, UNSPEC_COND_UO): Delete.
diff --git a/gcc/config/aarch64/aarch64-sve.md b/gcc/config/aarch64/aarch64-sve.md
index 7a1f58a..5c1427d 100644
--- a/gcc/config/aarch64/aarch64-sve.md
+++ b/gcc/config/aarch64/aarch64-sve.md
@@ -1358,6 +1358,49 @@
    cmp<cmp_op>\t%0.<Vetype>, %1/z, %2.<Vetype>, %3.<Vetype>"
 )
 
+;; Predicated integer comparisons, formed by combining a PTRUE-predicated
+;; comparison with an AND.  Split the instruction into its preferred form
+;; (below) at the earliest opportunity, in order to get rid of the
+;; redundant operand 1.
+(define_insn_and_split "*pred_cmp<cmp_op><mode>_combine"
+  [(set (match_operand:<VPRED> 0 "register_operand" "=Upa, Upa")
+       (and:<VPRED>
+         (unspec:<VPRED>
+           [(match_operand:<VPRED> 1)
+            (SVE_INT_CMP:<VPRED>
+              (match_operand:SVE_I 2 "register_operand" "w, w")
+              (match_operand:SVE_I 3 "aarch64_sve_cmp_<sve_imm_con>_operand" "<sve_imm_con>, w"))]
+           UNSPEC_MERGE_PTRUE)
+         (match_operand:<VPRED> 4 "register_operand" "Upl, Upl")))
+   (clobber (reg:CC CC_REGNUM))]
+  "TARGET_SVE"
+  "#"
+  "&& 1"
+  [(parallel
+     [(set (match_dup 0)
+          (and:<VPRED>
+            (SVE_INT_CMP:<VPRED>
+              (match_dup 2)
+              (match_dup 3))
+            (match_dup 4)))
+      (clobber (reg:CC CC_REGNUM))])]
+)
+
+;; Predicated integer comparisons.
+(define_insn "*pred_cmp<cmp_op><mode>"
+  [(set (match_operand:<VPRED> 0 "register_operand" "=Upa, Upa")
+	(and:<VPRED>
+	  (SVE_INT_CMP:<VPRED>
+	    (match_operand:SVE_I 2 "register_operand" "w, w")
+	    (match_operand:SVE_I 3 "aarch64_sve_cmp_<sve_imm_con>_operand" "<sve_imm_con>, w"))
+	  (match_operand:<VPRED> 1 "register_operand" "Upl, Upl")))
+   (clobber (reg:CC CC_REGNUM))]
+  "TARGET_SVE"
+  "@
+   cmp<cmp_op>\t%0.<Vetype>, %1/z, %2.<Vetype>, #%3
+   cmp<cmp_op>\t%0.<Vetype>, %1/z, %2.<Vetype>, %3.<Vetype>"
+)
+
 ;; Floating-point comparisons predicated with a PTRUE.
 (define_insn "*fcm<cmp_op><mode>"
   [(set (match_operand:<VPRED> 0 "register_operand" "=Upa, Upa")
@@ -1385,6 +1428,83 @@
   "fcmuo\t%0.<Vetype>, %1/z, %2.<Vetype>, %3.<Vetype>"
 )
 
+;; Floating-point comparisons predicated on a PTRUE, with the results ANDed
+;; with another predicate P.  This does not have the same trapping behavior
+;; as predicating the comparison itself on P, but it's a legitimate fold,
+;; since we can drop any potentially-trapping operations whose results
+;; are not needed.
+;;
+;; Split the instruction into its preferred form (below) at the earliest
+;; opportunity, in order to get rid of the redundant operand 1.
+(define_insn_and_split "*fcm<cmp_op><mode>_and_combine"
+  [(set (match_operand:<VPRED> 0 "register_operand" "=Upa, Upa")
+	(and:<VPRED>
+	  (unspec:<VPRED>
+	    [(match_operand:<VPRED> 1)
+	     (SVE_FP_CMP
+	       (match_operand:SVE_F 2 "register_operand" "w, w")
+	       (match_operand:SVE_F 3 "aarch64_simd_reg_or_zero" "Dz, w"))]
+	    UNSPEC_MERGE_PTRUE)
+	  (match_operand:<VPRED> 4 "register_operand" "Upl, Upl")))]
+  "TARGET_SVE"
+  "#"
+  "&& 1"
+  [(set (match_dup 0)
+	(and:<VPRED>
+	  (SVE_FP_CMP:<VPRED>
+	    (match_dup 2)
+	    (match_dup 3))
+	  (match_dup 4)))]
+)
+
+(define_insn_and_split "*fcmuo<mode>_and_combine"
+  [(set (match_operand:<VPRED> 0 "register_operand" "=Upa")
+	(and:<VPRED>
+	  (unspec:<VPRED>
+	    [(match_operand:<VPRED> 1)
+	     (unordered
+	       (match_operand:SVE_F 2 "register_operand" "w")
+	       (match_operand:SVE_F 3 "register_operand" "w"))]
+	    UNSPEC_MERGE_PTRUE)
+	  (match_operand:<VPRED> 4 "register_operand" "Upl")))]
+  "TARGET_SVE"
+  "#"
+  "&& 1"
+  [(set (match_dup 0)
+	(and:<VPRED>
+	  (unordered:<VPRED>
+	    (match_dup 2)
+	    (match_dup 3))
+	  (match_dup 4)))]
+)
+
+;; Unpredicated floating-point comparisons, with the results ANDed
+;; with another predicate.  This is a valid fold for the same reasons
+;; as above.
+(define_insn "*fcm<cmp_op><mode>_and"
+  [(set (match_operand:<VPRED> 0 "register_operand" "=Upa, Upa")
+	(and:<VPRED>
+	  (SVE_FP_CMP:<VPRED>
+	    (match_operand:SVE_F 2 "register_operand" "w, w")
+	    (match_operand:SVE_F 3 "aarch64_simd_reg_or_zero" "Dz, w"))
+	  (match_operand:<VPRED> 1 "register_operand" "Upl, Upl")))]
+  "TARGET_SVE"
+  "@
+   fcm<cmp_op>\t%0.<Vetype>, %1/z, %2.<Vetype>, #0.0
+   fcm<cmp_op>\t%0.<Vetype>, %1/z, %2.<Vetype>, %3.<Vetype>"
+)
+
+(define_insn "*fcmuo<mode>_and"
+  [(set (match_operand:<VPRED> 0 "register_operand" "=Upa")
+	(and:<VPRED>
+	  (unordered:<VPRED>
+	    (match_operand:SVE_F 2 "register_operand" "w")
+	    (match_operand:SVE_F 3 "register_operand" "w"))
+	  (match_operand:<VPRED> 1 "register_operand" "Upl")))]
+  "TARGET_SVE"
+  "fcmuo\t%0.<Vetype>, %1/z, %2.<Vetype>, %3.<Vetype>"
+)
+
 ;; Predicated floating-point comparisons.  We don't need a version
 ;; of this for unordered comparisons.
 (define_insn "*pred_fcm<cmp_op><mode>"
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index d5126b3..a292ea0 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,12 @@
+2018-05-08  Richard Sandiford  <richard.sandiford@linaro.org>
+	    Alan Hayward  <alan.hayward@arm.com>
+	    David Sherwood  <david.sherwood@arm.com>
+
+	* gcc.target/aarch64/sve/vcond_6.c: Do not expect any ANDs.
+	XFAIL the BIC test.
+	* gcc.target/aarch64/sve/vcond_7.c: New test.
+	* gcc.target/aarch64/sve/vcond_7_run.c: Likewise.
+
 2018-05-08  Paolo Carlini  <paolo.carlini@oracle.com>
 
 	PR c++/70563
diff --git a/gcc/testsuite/gcc.target/aarch64/sve/vcond_6.c b/gcc/testsuite/gcc.target/aarch64/sve/vcond_6.c
index 718afae..a59f08d 100644
--- a/gcc/testsuite/gcc.target/aarch64/sve/vcond_6.c
+++ b/gcc/testsuite/gcc.target/aarch64/sve/vcond_6.c
@@ -43,10 +43,16 @@
 
 TEST_ALL (LOOP)
 
-/* { dg-final { scan-assembler-times {\tand\tp[0-9]+\.b, p[0-9]+/z, p[0-9]+\.b, p[0-9]+\.b} 3 } } */
+/* ??? We predicate one of the comparisons on the result of the other,
+   but whether that's a win or a loss will depend on the schedule.  */
+/* { dg-final { scan-assembler-not {\tand\t} } } */
 /* { dg-final { scan-assembler-times {\torr\tp[0-9]+\.b, p[0-9]+/z, p[0-9]+\.b, p[0-9]+\.b} 3 } } */
 /* { dg-final { scan-assembler-times {\teor\tp[0-9]+\.b, p[0-9]+/z, p[0-9]+\.b, p[0-9]+\.b} 3 } } */
 /* { dg-final { scan-assembler-times {\tnand\tp[0-9]+\.b, p[0-9]+/z, p[0-9]+\.b, p[0-9]+\.b} 3 } } */
 /* { dg-final { scan-assembler-times {\tnor\tp[0-9]+\.b, p[0-9]+/z, p[0-9]+\.b, p[0-9]+\.b} 3 } } */
-/* { dg-final { scan-assembler-times {\tbic\tp[0-9]+\.b, p[0-9]+/z, p[0-9]+\.b, p[0-9]+\.b} 3 } } */
+/* Currently we predicate one of the comparisons on the result of the other
+   and then use NOT, but the original BIC sequence is better.  It's a fairly
+   niche failure though.  We'd handle most other types of comparison by
+   using the inverse operation instead of a separate NOT.  */
+/* { dg-final { scan-assembler-times {\tbic\tp[0-9]+\.b, p[0-9]+/z, p[0-9]+\.b, p[0-9]+\.b} 3 { xfail *-*-* } } */
 /* { dg-final { scan-assembler-times {\torn\tp[0-9]+\.b, p[0-9]+/z, p[0-9]+\.b, p[0-9]+\.b} 3 } } */
diff --git a/gcc/testsuite/gcc.target/aarch64/sve/vcond_7.c b/gcc/testsuite/gcc.target/aarch64/sve/vcond_7.c
new file mode 100644
index 0000000..d2cdbdc
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/sve/vcond_7.c
@@ -0,0 +1,216 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -ftree-vectorize" } */
+
+#include <stdint.h>
+
+#define N 100
+
+#define eq(A, B) ((A) == (B))
+#define ne(A, B) ((A) != (B))
+#define lt(A, B) ((A) < (B))
+#define le(A, B) ((A) <= (B))
+#define ge(A, B) ((A) >= (B))
+#define gt(A, B) ((A) > (B))
+#define unordered(A, B) (__builtin_isunordered (A, B))
+
+#define DEF_CONST_LOOP(NAME, SUFFIX, TYPE, CONST)			\
+  void __attribute__ ((noipa))						\
+  NAME##_##SUFFIX##_##TYPE (TYPE *restrict dst, TYPE *restrict src)	\
+  {									\
+    for (int i = 0; i < N; ++i)						\
+      if (NAME (src[i], CONST))						\
+	dst[i] = 1;							\
+  }
+
+#define DEF_LOOP(NAME, TYPE, CONST1, CONST2)				\
+  void __attribute__ ((noipa))						\
+  NAME##_var_##TYPE (TYPE *restrict dst, TYPE *restrict src, TYPE x)	\
+  {									\
+    for (int i = 0; i < N; ++i)						\
+      if (NAME (src[i], x))						\
+	dst[i] = x;							\
+  }									\
+  DEF_CONST_LOOP (NAME, const1, TYPE, CONST1)				\
+  DEF_CONST_LOOP (NAME, const2, TYPE, CONST2)
+
+#define FOR_EACH_INT_OPERATOR(T, TYPE, CONST1, CONST2) \
+  T (eq, TYPE, CONST1, CONST2) \
+  T (ne, TYPE, CONST1, CONST2) \
+  T (le, TYPE, CONST1, CONST2) \
+  T (lt, TYPE, CONST1, CONST2) \
+  T (gt, TYPE, CONST1, CONST2) \
+  T (ge, TYPE, CONST1, CONST2)
+
+#define FOR_EACH_FLOAT_OPERATOR(T, TYPE, CONST1, CONST2) \
+  FOR_EACH_INT_OPERATOR(T, TYPE, CONST1, CONST2) \
+  T (unordered, TYPE, CONST1, CONST2)
+
+#define FOR_EACH_TYPE(T) \
+  FOR_EACH_INT_OPERATOR (T, int8_t, 2, 100) \
+  FOR_EACH_INT_OPERATOR (T, int16_t, 3, 1000) \
+  FOR_EACH_INT_OPERATOR (T, int32_t, 4, 2000) \
+  FOR_EACH_INT_OPERATOR (T, int64_t, 5, 3000) \
+  FOR_EACH_INT_OPERATOR (T, uint8_t, 2, 160) \
+  FOR_EACH_INT_OPERATOR (T, uint16_t, 3, 500) \
+  FOR_EACH_INT_OPERATOR (T, uint32_t, 4, 1500) \
+  FOR_EACH_INT_OPERATOR (T, uint64_t, 5, 2500) \
+  FOR_EACH_FLOAT_OPERATOR (T, _Float16, 0, 1) \
+  FOR_EACH_FLOAT_OPERATOR (T, float, 0, 1) \
+  FOR_EACH_FLOAT_OPERATOR (T, double, 0, 1)
+
+FOR_EACH_TYPE (DEF_LOOP)
+
+/* { dg-final { scan-assembler-not {\tand\t} } } */
+
+/* { dg-final { scan-assembler-times {\tcmpeq\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, z[0-9]+\.b\n} 4 } } */
+/* { dg-final { scan-assembler-times {\tcmpeq\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 4 } } */
+/* { dg-final { scan-assembler-times {\tcmpeq\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 4 } } */
+/* { dg-final { scan-assembler-times {\tcmpeq\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 4 } } */
+
+/* { dg-final { scan-assembler-times {\tcmpeq\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, #2\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tcmpeq\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #3\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tcmpeq\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #4\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tcmpeq\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #5\n} 2 } } */
+
+/* { dg-final { scan-assembler-times {\tcmpne\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, z[0-9]+\.b\n} 4 } } */
+/* { dg-final { scan-assembler-times {\tcmpne\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 4 } } */
+/* { dg-final { scan-assembler-times {\tcmpne\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 4 } } */
+/* { dg-final { scan-assembler-times {\tcmpne\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 4 } } */
+
+/* { dg-final { scan-assembler-times {\tcmpne\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, #2\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tcmpne\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #3\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tcmpne\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #4\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tcmpne\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #5\n} 2 } } */
+
+/* { dg-final { scan-assembler-times {\tcmplt\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, z[0-9]+\.b\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmplt\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmplt\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmplt\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tcmple\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, #1\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmple\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #2\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmple\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #3\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmple\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #4\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tcmple\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, z[0-9]+\.b\n} 3 } } */
+/* { dg-final { scan-assembler-times {\tcmple\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 3 } } */
+/* { dg-final { scan-assembler-times {\tcmple\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 3 } } */
+/* { dg-final { scan-assembler-times {\tcmple\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 3 } } */
+
+/* { dg-final { scan-assembler-times {\tcmple\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, #2\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmple\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #3\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmple\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #4\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmple\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #5\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tcmpgt\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, z[0-9]+\.b\n} 3 } } */
+/* { dg-final { scan-assembler-times {\tcmpgt\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 3 } } */
+/* { dg-final { scan-assembler-times {\tcmpgt\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 3 } } */
+/* { dg-final { scan-assembler-times {\tcmpgt\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 3 } } */
+
+/* { dg-final { scan-assembler-times {\tcmpgt\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, #2\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpgt\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #3\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpgt\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #4\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpgt\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #5\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tcmpge\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, z[0-9]+\.b\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpge\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpge\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpge\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tcmpgt\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, #1\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpgt\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #2\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpgt\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #3\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpgt\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #4\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tcmplo\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, z[0-9]+\.b\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmplo\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmplo\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmplo\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tcmpls\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, #1\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpls\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #2\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpls\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #3\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpls\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #4\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tcmpls\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, z[0-9]+\.b\n} 3 } } */
+/* { dg-final { scan-assembler-times {\tcmpls\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 3 } } */
+/* { dg-final { scan-assembler-times {\tcmpls\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 3 } } */
+/* { dg-final { scan-assembler-times {\tcmpls\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 3 } } */
+
+/* { dg-final { scan-assembler-times {\tcmpls\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, #2\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpls\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #3\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpls\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #4\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmpls\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #5\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tcmphi\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, z[0-9]+\.b\n} 3 } } */
+/* { dg-final { scan-assembler-times {\tcmphi\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 3 } } */
+/* { dg-final { scan-assembler-times {\tcmphi\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 3 } } */
+/* { dg-final { scan-assembler-times {\tcmphi\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 3 } } */
+
+/* { dg-final { scan-assembler-times {\tcmphi\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, #2\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmphi\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #3\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmphi\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #4\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmphi\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #5\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tcmphs\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, z[0-9]+\.b\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmphs\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmphs\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmphs\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tcmphi\tp[0-7]\.b, p[0-7]/z, z[0-9]+\.b, #1\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmphi\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #2\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmphi\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #3\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tcmphi\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #4\n} 1 } } */
+
+
+/* { dg-final { scan-assembler-times {\tfcmeq\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tfcmeq\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tfcmeq\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 2 } } */
+
+/* { dg-final { scan-assembler-times {\tfcmeq\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #0\.0\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tfcmeq\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #0\.0\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tfcmeq\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #0\.0\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tfcmne\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tfcmne\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tfcmne\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 2 } } */
+
+/* { dg-final { scan-assembler-times {\tfcmne\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #0\.0\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tfcmne\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #0\.0\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tfcmne\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #0\.0\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tfcmle\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tfcmle\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tfcmle\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 2 } } */
+
+/* { dg-final { scan-assembler-times {\tfcmle\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #0\.0\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tfcmle\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #0\.0\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tfcmle\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #0\.0\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tfcmlt\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tfcmlt\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tfcmlt\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 2 } } */
+
+/* { dg-final { scan-assembler-times {\tfcmlt\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #0\.0\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tfcmlt\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #0\.0\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tfcmlt\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #0\.0\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tfcmge\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tfcmge\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tfcmge\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 2 } } */
+
+/* { dg-final { scan-assembler-times {\tfcmge\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #0\.0\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tfcmge\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #0\.0\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tfcmge\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #0\.0\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tfcmgt\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tfcmgt\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tfcmgt\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 2 } } */
+
+/* { dg-final { scan-assembler-times {\tfcmgt\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, #0\.0\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tfcmgt\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, #0\.0\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tfcmgt\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, #0\.0\n} 1 } } */
+
+/* { dg-final { scan-assembler-times {\tfcmuo\tp[0-7]\.h, p[0-7]/z, z[0-9]+\.h, z[0-9]+\.h\n} 3 } } */
+/* { dg-final { scan-assembler-times {\tfcmuo\tp[0-7]\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 3 } } */
+/* { dg-final { scan-assembler-times {\tfcmuo\tp[0-7]\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 3 } } */
diff --git a/gcc/testsuite/gcc.target/aarch64/sve/vcond_7_run.c b/gcc/testsuite/gcc.target/aarch64/sve/vcond_7_run.c
new file mode 100644
index 0000000..e440f3a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/sve/vcond_7_run.c
@@ -0,0 +1,40 @@
+/* { dg-do run { target aarch64_sve_hw } } */
+/* { dg-options "-O2 -ftree-vectorize" } */
+
+#include "vcond_7.c"
+
+#define TEST_CONST_LOOP(NAME, SUFFIX, TYPE, CONST)			\
+  {									\
+    for (int i = 0; i < N; ++i)						\
+      {									\
+	dst[i] = i * 3;							\
+	src[i] = i % (CONST + 3);					\
+      }									\
+    NAME##_##SUFFIX##_##TYPE (dst, src);				\
+    for (int i = 0; i < N; ++i)						\
+      if (dst[i] != (NAME (src[i], CONST) ? (TYPE) 1 : (TYPE) (i * 3)))	\
+	__builtin_abort ();						\
+  }
+
+#define TEST_LOOPS(NAME, TYPE, CONST1, CONST2)				\
+  {									\
+    TYPE dst[N], src[N];						\
+    for (int i = 0; i < N; ++i)						\
+      {									\
+	dst[i] = i * 2;							\
+	src[i] = i % 5;							\
+      }									\
+    NAME##_var_##TYPE (dst, src, 3);					\
+    for (int i = 0; i < N; ++i)						\
+      if (dst[i] != (NAME (src[i], 3) ? (TYPE) 3 : (TYPE) (i * 2)))	\
+	__builtin_abort ();						\
+    TEST_CONST_LOOP (NAME, const1, TYPE, CONST1)			\
+    TEST_CONST_LOOP (NAME, const2, TYPE, CONST2)			\
+  }
+
+int __attribute__ ((noipa))
+main (void)
+{
+  FOR_EACH_TYPE (TEST_LOOPS);
+  return 0;
+}