make @boolToInt always return a u1 #15701
Conversation
Signed-off-by: tison <[email protected]>
|
There seems some more type inconsistent. I'm handling it: @Vexu could you avoid force push before we are pending for merging? Otherwise it's hard for me to fetch your changes. |
Signed-off-by: tison <[email protected]>
My bad, lol. |
Signed-off-by: tison <[email protected]>
A new issue without source code location. Let me try to reproduce it locally. |
|
There are plenty of other error you might want to start with if you skip the 3000 lines of noise. |
Signed-off-by: tison <[email protected]>
|
Fix the test result issue. For translate C issue, locally it logs: May I ask where this translate logic stay? |
|
That is generated with |
|
@Vexu Thanks for your information! I'll appreciate it if you can help in updating the related logics. I took a look at |
|
It seems now all aarch64 tests passed but x86_64 ones failed. How can our changes be related to arch? Related logs: |
Note that not all runners run the same tests, and that Specifically, diff --git a/src/arch/x86_64/CodeGen.zig b/src/arch/x86_64/CodeGen.zig
index a258f732f..594a1381a 100644
--- a/src/arch/x86_64/CodeGen.zig
+++ b/src/arch/x86_64/CodeGen.zig
@@ -10105,18 +10105,22 @@ fn airBitCast(self: *Self, inst: Air.Inst.Index) !void {
const dst_rc = regClassForType(dst_ty);
const src_rc = regClassForType(src_ty);
const src_mcv = try self.resolveInst(ty_op.operand);
- if (dst_rc.supersetOf(src_rc) and self.reuseOperand(inst, ty_op.operand, 0, src_mcv))
- break :result src_mcv;
const src_lock = if (src_mcv.getReg()) |reg| self.register_manager.lockReg(reg) else null;
defer if (src_lock) |lock| self.register_manager.unlockReg(lock);
- const dst_mcv = try self.allocRegOrMem(inst, true);
- try self.genCopy(
- if (!dst_mcv.isMemory() or src_mcv.isMemory()) dst_ty else src_ty,
- dst_mcv,
- src_mcv,
- );
+ const dst_mcv = if (dst_rc.supersetOf(src_rc) and
+ self.reuseOperand(inst, ty_op.operand, 0, src_mcv))
+ src_mcv
+ else dst: {
+ const dst_mcv = try self.allocRegOrMem(inst, true);
+ try self.genCopy(
+ if (!dst_mcv.isMemory() or src_mcv.isMemory()) dst_ty else src_ty,
+ dst_mcv,
+ src_mcv,
+ );
+ break :dst dst_mcv;
+ };
const dst_signedness =
if (dst_ty.isAbiInt()) dst_ty.intInfo(self.target.*).signedness else .unsigned; |
|
@jacobly0 Thank you! I've merged the latest master branch. Could you approve the CI workflow here? |
|
The langref should be updated: {#header_open|@boolToInt#}
<pre>{#syntax#}@boolToInt(value: bool) u1{#endsyntax#}</pre>
<p>
Converts {#syntax#}true{#endsyntax#} to {#syntax#}@as(u1, 1){#endsyntax#} and {#syntax#}false{#endsyntax#} to
{#syntax#}@as(u1, 0){#endsyntax#}.
</p>
- <p>
- If the value is known at compile-time, the return type is {#syntax#}comptime_int{#endsyntax#}
- instead of {#syntax#}u1{#endsyntax#}.
- </p>
{#header_close#}Anyway, wouldn't it be better if we just remove
|
|
@r00ster91 I'm updating the doc... For drop Since it has the similar effect of #7950 and #7950 has been accepted, I'll proceed this patch and wait for ideas on dropping |
Signed-off-by: tison <[email protected]>
|
Docs updated. |
|
I'd like to hear Andrew's opinion on #15794 first. |
@Vexu That sounds reasonable. Although, as #7950 was accepted and #7950 doesn't conflict #15794, this pull request can even be a prerequisite of #15794 since it fixes all recipients of |
This closes #7950.
cc @andrewrk @mlugg in case this patch helps.