On 3/27/19 9:24 AM, Jeff Law wrote:
> On 3/27/19 8:36 AM, Jakub Jelinek wrote:
>> On Sun, Mar 24, 2019 at 09:20:07AM -0600, Jeff Law wrote:
>>> However, I'm increasingly of the opinion that MIPS targets need to drop
>>> off the priority platform list. Given the trajectory I see for MIPS
>>> based processors in industry, it's really hard to justify spending this
>>> much time on them, particularly for low priority code quality issues.
>>
>> Besides what has been discussed on IRC for the PR89826 fix, that we really
>> need a df_analyze before processing the first block, because otherwise we
>> can't rely on the REG_UNUSED notes in the IL, I see some other issues, but I
>> admit I don't know much about df nor regcprop.
> RIght. I plan to commit that today along with the test reordering you
> pointed out.
And the actual patch committed after the usual bootstrapping &
regression test on x86_64 plus testing on various *-elf platforms,
mips64-linux-gnu, mips64el-linux-gnu and others :-)
Jeff
diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 07a1333cee0..f8a353d16c0 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,13 @@
+2019-03-27 Jeff Law <l...@redhat.com>
+
+
+ PR rtl-optimization/87761
+ PR rtl-optimization/89826
+ * regcprop.c (copyprop_hardreg_forward_1): Move may_trap_p test
+ slightly later.
+ (pass_cprop_hardreg::execute): Call df_analyze after adding the
+ note problem to get REG_DEAD/REG_UNUSED notes updated.
+
2019-03-27 Richard Biener <rguent...@suse.de>
PR tree-optimization/89463
diff --git a/gcc/regcprop.c b/gcc/regcprop.c
index 8ca523ffe23..3efe21f377c 100644
--- a/gcc/regcprop.c
+++ b/gcc/regcprop.c
@@ -800,9 +800,9 @@ copyprop_hardreg_forward_1 (basic_block bb, struct
value_data *vd)
/* Detect obviously dead sets (via REG_UNUSED notes) and remove them. */
if (set
- && !may_trap_p (set)
&& !RTX_FRAME_RELATED_P (insn)
&& INSN_P (insn)
+ && !may_trap_p (set)
&& find_reg_note (insn, REG_UNUSED, SET_DEST (set))
&& !side_effects_p (SET_SRC (set))
&& !side_effects_p (SET_DEST (set)))
@@ -1293,7 +1293,10 @@ pass_cprop_hardreg::execute (function *fun)
auto_sbitmap visited (last_basic_block_for_fn (fun));
bitmap_clear (visited);
+ /* We need accurate notes. Earlier passes such as if-conversion may
+ leave notes in an inconsistent state. */
df_note_add_problem ();
+ df_analyze ();
/* It is tempting to set DF_LR_RUN_DCE, but DCE may choose to delete
an insn and this pass would not have visibility into the removal.
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 0679cb72e52..b2c649cfb3e 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,9 @@
+2019-03-26 Jeff Law <l...@redhat.com>
+
+ PR rtl-optimization/87761
+ PR rtl-optimization/89826
+ * gcc.c-torture/execute/pr89826.c: New test.
+
2019-03-27 Richard Biener <rguent...@suse.de>
* gcc.dg/torture/20190327-1.c: New testcase.
diff --git a/gcc/testsuite/gcc.c-torture/execute/pr89826.c
b/gcc/testsuite/gcc.c-torture/execute/pr89826.c
new file mode 100644
index 00000000000..091644849d3
--- /dev/null
+++ b/gcc/testsuite/gcc.c-torture/execute/pr89826.c
@@ -0,0 +1,21 @@
+typedef unsigned int u32;
+typedef unsigned long long u64;
+u64 a;
+u32 b;
+
+u64
+foo (u32 d)
+{
+ a -= d ? 0 : ~a;
+ return a + b;
+}
+
+int
+main (void)
+{
+ u64 x = foo (2);
+ if (x != 0)
+ __builtin_abort();
+ return 0;
+}
+
