@@ -1166,7 +1166,7 @@ Common Joined UInteger Var(param_vect_induction_float) Init(1) IntegerRage(0, 1)
Enable loop vectorization of floating point inductions.
-param=vrp1-mode=
-Common Joined Var(param_vrp1_mode) Enum(vrp_mode) Init(VRP_MODE_VRP) Param Optimization
+Common Joined Var(param_vrp1_mode) Enum(vrp_mode) Init(VRP_MODE_RANGER) Param Optimization
--param=vrp1-mode=[vrp|ranger] Specifies the mode VRP1 should operate in.
-param=vrp2-mode=
@@ -10,4 +10,4 @@ int foo (void)
return 0;
}
-/* { dg-final { scan-tree-dump "\\\[-INF, 0\\\]" "vrp1" } } */
+/* { dg-final { scan-tree-dump "\\\[-INF, -INF\\\]\\\[0, 0\\\]" "vrp1" } } */
new file mode 100644
@@ -0,0 +1,42 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-tree-vrp1-alias -fdump-tree-vrp2-alias" } */
+
+void dead (unsigned n);
+void alive (unsigned n);
+
+void func (unsigned n, unsigned m)
+{
+ if (n == 0)
+ __builtin_unreachable();
+ if (n == 1)
+ __builtin_unreachable();
+ if (n & 0x1)
+ __builtin_unreachable();
+ if (n == 2)
+ __builtin_unreachable();
+ if (n == 3)
+ __builtin_unreachable();
+ if (n & 0x2)
+ __builtin_unreachable();
+ if (n == 4)
+ __builtin_unreachable();
+ if (n == 5)
+ __builtin_unreachable();
+ if (n & 0x4)
+ __builtin_unreachable();
+ if (n == 6)
+ __builtin_unreachable();
+ if (n == 7)
+ __builtin_unreachable();
+ if (n <8)
+ dead (n);
+ if (n != m)
+ __builtin_unreachable();
+ alive (n);
+ alive (m);
+}
+
+/* { dg-final { scan-tree-dump-not "dead" "vrp1" } } */
+/* { dg-final { scan-tree-dump-times "builtin_unreachable" 1 "vrp1" } } */
+/* { dg-final { scan-tree-dump-not "builtin_unreachable" "vrp2" } } */
+/* { dg-final { scan-tree-dump-times "fff8" 4 "vrp2" } } */
This patch turns ranger on by default for the VRP1 pass. I needed to adjust gcc.dg/pr68217.c to scan for a better range ([-INF, -INF][0, 0]) than the original [-INF, 0] it was looking for. This also triggers the new __builtin_unreachable code in the ranger VRP pass, so I added a new testcase to show accumulated unreachables are combined properly. tree-ssa/pr107009.c also verifies that a non-dominated unreachable call doesn't incorrectly affect the global range. Bootstrapped on x86_64-pc-linux-gnu with no regressions*. Pushed. * OK. no regressions is debatable. I've been seeing the following spurious failure for the past few weeks. Ranger made them pass for a while, then made them fail, so Ive been ignoring them. I took a quick look. Basically, we optimize away an unnecessary statement feeding a condition based on a combination of what ranger calculates and results from loop analysis. (The statement is not referenced in any debug statement). That later causes a PHI to no longer been needed, and DCE2 removed the PHI and some other stuff, which then causes the debug_stmt to lose its reference. /blame dce2 :-) < FAIL: gcc.dg/guality/pr54693-2.c -O2 -flto -fuse-linker-plugin -fno-fat-lto-objects -DPREVENT_OPTIMIZATION line 21 x == 10 - i < FAIL: gcc.dg/guality/pr54693-2.c -O2 -flto -fuse-linker-plugin -fno-fat-lto-objects -DPREVENT_OPTIMIZATION line 21 y == 20 - 2 * i < FAIL: gcc.dg/guality/pr54693-2.c -O2 -flto -fuse-linker-plugin -fno-fat-lto-objects -DPREVENT_OPTIMIZATION line 21 z == 30 - 3 * i I choose to ignore this for now... I want to get VRP1 turned to ranger by default and see if any issues show up. I believe this removes the last remaining use of legacy vrp. Andrew commit e7310e24b1c0ca67b1bb507c1330b2bf39e59e32 Author: Andrew MacLeod <amacleod@redhat.com> Date: Tue Oct 25 16:42:41 2022 -0400 Make ranger vrp1 default. Turn on ranger as the default vrp1 pass and adjust testcases. gcc/ * params.opt (param_vrp1_mode): Make ranger default. gcc/testsuite/ * gcc.dg/pr68217.c: Test [-INF, -INF][0, 0] instead of [-INF, 0]. * gcc.dg/tree-ssa/vrp-unreachable.c: New. Test unreachable removal.