From: juewang <[email protected]>
Some targets only provide a valid issue rate when instruction scheduling
is enabled. Fall back to serial costing when flag_schedule_insns is
false.
Assisted-by: Codex (OpenAI)
gcc/ChangeLog:
* ifcvt.cc (noce_parallel_seq_cost): Fall back to serial cost when
instruction scheduling is disabled.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/ifcvt-parallel-cost-1.c: New test.
* gcc.target/riscv/ifcvt-parallel-cost-2.c: Likewise.
* gcc.target/riscv/ifcvt-parallel-cost-3.c: Likewise.
---
gcc/ifcvt.cc | 4 ++-
.../gcc.target/riscv/ifcvt-parallel-cost-1.c | 26 +++++++++++++++++
.../gcc.target/riscv/ifcvt-parallel-cost-2.c | 29 +++++++++++++++++++
.../gcc.target/riscv/ifcvt-parallel-cost-3.c | 28 ++++++++++++++++++
4 files changed, 86 insertions(+), 1 deletion(-)
create mode 100644 gcc/testsuite/gcc.target/riscv/ifcvt-parallel-cost-1.c
create mode 100644 gcc/testsuite/gcc.target/riscv/ifcvt-parallel-cost-2.c
create mode 100644 gcc/testsuite/gcc.target/riscv/ifcvt-parallel-cost-3.c
diff --git a/gcc/ifcvt.cc b/gcc/ifcvt.cc
index 8df6c3a6527..404185301b4 100644
--- a/gcc/ifcvt.cc
+++ b/gcc/ifcvt.cc
@@ -867,7 +867,9 @@ noce_parallel_seq_cost (rtx_insn *seq, bool speed_p)
{
unsigned int serial_cost = seq_cost (seq, speed_p);
- if (!speed_p || !targetm.sched.issue_rate)
+ if (!speed_p
+ || !flag_schedule_insns
+ || !targetm.sched.issue_rate)
return serial_cost;
unsigned int issue_rate = MAX (targetm.sched.issue_rate (), 1);
diff --git a/gcc/testsuite/gcc.target/riscv/ifcvt-parallel-cost-1.c
b/gcc/testsuite/gcc.target/riscv/ifcvt-parallel-cost-1.c
new file mode 100644
index 00000000000..1bec3f77dcd
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/ifcvt-parallel-cost-1.c
@@ -0,0 +1,26 @@
+/* { dg-do compile } */
+/* { dg-skip-if "" { *-*-* } { "-O0" "-Og" "-Os" "-Oz" } } */
+/* { dg-options "-O2 -march=rv64gc_zicond -mabi=lp64d -mtune=xt-c9501fdvt
-mbranch-cost=3" } */
+
+extern void consume (void *, void *, void *);
+
+__attribute__ ((noinline, noclone))
+void
+select_three_pointers (long x, long y,
+ void *a0, void *a1,
+ void *b0, void *b1,
+ void *c0, void *c1)
+{
+ if (x > y)
+ {
+ a0 = a1;
+ b0 = b1;
+ c0 = c1;
+ }
+
+ consume (a0, b0, c0);
+}
+
+/* { dg-final { scan-assembler-times "\\sczero\\.eqz\\s" 3 } } */
+/* { dg-final { scan-assembler-times "\\sczero\\.nez\\s" 3 } } */
+/* { dg-final { scan-assembler-not "\\s(?:bge|bgt|ble|blt)\\s" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/ifcvt-parallel-cost-2.c
b/gcc/testsuite/gcc.target/riscv/ifcvt-parallel-cost-2.c
new file mode 100644
index 00000000000..6e1f9ed7849
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/ifcvt-parallel-cost-2.c
@@ -0,0 +1,29 @@
+/* { dg-do compile } */
+/* { dg-skip-if "" { *-*-* } { "-O0" "-Og" "-Os" "-Oz" } } */
+/* { dg-options "-O2 -march=rv64gc_zicond -mabi=lp64d -mtune=generic
-mbranch-cost=3" } */
+
+extern void consume (void *, void *, void *);
+
+__attribute__ ((noinline, noclone))
+void
+select_three_pointers_generic (long x, long y,
+ void *a0, void *a1,
+ void *b0, void *b1,
+ void *c0, void *c1)
+{
+ if (x >= y)
+ {
+ a0 = a1;
+ b0 = b1;
+ c0 = c1;
+ }
+
+ consume (a0, b0, c0);
+}
+
+/* The generic tuning model has an issue rate of one, so the three pointer
+ selections are not profitable at a branch cost of three. */
+
+/* { dg-final { scan-assembler-times "\\sbge\\s" 1 } } */
+/* { dg-final { scan-assembler-not "\\sczero\\.eqz\\s" } } */
+/* { dg-final { scan-assembler-not "\\sczero\\.nez\\s" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/ifcvt-parallel-cost-3.c
b/gcc/testsuite/gcc.target/riscv/ifcvt-parallel-cost-3.c
new file mode 100644
index 00000000000..c499b2fe089
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/ifcvt-parallel-cost-3.c
@@ -0,0 +1,28 @@
+/* { dg-do compile } */
+/* { dg-skip-if "" { *-*-* } { "-O0" "-Og" "-Os" "-Oz" } } */
+/* { dg-options "-O2 -march=rv64gc_zicond -mabi=lp64d -mtune=xt-c9501fdvt
-mbranch-cost=3 -fno-schedule-insns" } */
+
+extern void consume (void *, void *, void *);
+
+__attribute__ ((noinline, noclone))
+void
+select_three_pointers_no_sched (long x, long y,
+ void *a0, void *a1,
+ void *b0, void *b1,
+ void *c0, void *c1)
+{
+ if (x >= y)
+ {
+ a0 = a1;
+ b0 = b1;
+ c0 = c1;
+ }
+
+ consume (a0, b0, c0);
+}
+
+/* Do not use the issue rate when instruction scheduling is disabled. */
+
+/* { dg-final { scan-assembler-times "\\sbge\\s" 1 } } */
+/* { dg-final { scan-assembler-not "\\sczero\\.eqz\\s" } } */
+/* { dg-final { scan-assembler-not "\\sczero\\.nez\\s" } } */
--
2.34.1
