[clang] [llvm] [RISCV] Add partial support for -fzero-call-used-regs (PR #194883)
Lucas Chollet via cfe-commits
cfe-commits at lists.llvm.org
Fri Jun 5 06:04:51 PDT 2026
================
@@ -0,0 +1,297 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc < %s -verify-machineinstrs -mtriple=riscv64-unknown-unknown | FileCheck %s --check-prefixes=CHECK,64-BITS
+; RUN: llc < %s -verify-machineinstrs -mtriple=riscv32-unknown-unknown | FileCheck %s --check-prefixes=CHECK,32-BITS
+
+target triple = "riscv64-unknown-linux-gnu"
+
+define i32 @skip(i32 noundef %a, i32 noundef %b, i32 noundef %c) #0 "zero-call-used-regs"="skip" {
+; 64-BITS-LABEL: skip:
+; 64-BITS: # %bb.0: # %entry
+; 64-BITS-NEXT: mulw a0, a1, a0
+; 64-BITS-NEXT: or a0, a0, a2
+; 64-BITS-NEXT: ret
+;
+; 32-BITS-LABEL: skip:
+; 32-BITS: # %bb.0: # %entry
+; 32-BITS-NEXT: mul a0, a1, a0
+; 32-BITS-NEXT: or a0, a0, a2
+; 32-BITS-NEXT: ret
+
+entry:
+ %mul = mul nsw i32 %b, %a
+ %or = or i32 %mul, %c
+ ret i32 %or
+}
+
+define i32 @used_gpr_arg(i32 noundef %a, i32 noundef %b, i32 noundef %c) #0 "zero-call-used-regs"="used-gpr-arg" {
+; 64-BITS-LABEL: used_gpr_arg:
+; 64-BITS: # %bb.0: # %entry
+; 64-BITS-NEXT: mulw a0, a1, a0
+; 64-BITS-NEXT: or a0, a0, a2
+; 64-BITS-NEXT: li a1, 0
+; 64-BITS-NEXT: li a2, 0
+; 64-BITS-NEXT: ret
+;
+; 32-BITS-LABEL: used_gpr_arg:
+; 32-BITS: # %bb.0: # %entry
+; 32-BITS-NEXT: mul a0, a1, a0
+; 32-BITS-NEXT: or a0, a0, a2
+; 32-BITS-NEXT: li a1, 0
+; 32-BITS-NEXT: li a2, 0
+; 32-BITS-NEXT: ret
+
+entry:
+ %mul = mul nsw i32 %b, %a
+ %or = or i32 %mul, %c
+ ret i32 %or
+}
+
+define i32 @used_gpr(i32 noundef %a, i32 noundef %b, i32 noundef %c) #0 "zero-call-used-regs"="used-gpr" {
+; 64-BITS-LABEL: used_gpr:
+; 64-BITS: # %bb.0: # %entry
+; 64-BITS-NEXT: mulw a0, a1, a0
+; 64-BITS-NEXT: or a0, a0, a2
+; 64-BITS-NEXT: li a1, 0
+; 64-BITS-NEXT: li a2, 0
+; 64-BITS-NEXT: ret
+;
+; 32-BITS-LABEL: used_gpr:
+; 32-BITS: # %bb.0: # %entry
+; 32-BITS-NEXT: mul a0, a1, a0
+; 32-BITS-NEXT: or a0, a0, a2
+; 32-BITS-NEXT: li a1, 0
+; 32-BITS-NEXT: li a2, 0
+; 32-BITS-NEXT: ret
+
+entry:
+ %mul = mul nsw i32 %b, %a
+ %or = or i32 %mul, %c
+ ret i32 %or
+}
+
+define i32 @used_arg(i32 noundef %a, i32 noundef %b, i32 noundef %c) #0 "zero-call-used-regs"="used-arg" {
+; 64-BITS-LABEL: used_arg:
+; 64-BITS: # %bb.0: # %entry
+; 64-BITS-NEXT: mulw a0, a1, a0
+; 64-BITS-NEXT: or a0, a0, a2
+; 64-BITS-NEXT: li a1, 0
+; 64-BITS-NEXT: li a2, 0
+; 64-BITS-NEXT: ret
+;
+; 32-BITS-LABEL: used_arg:
+; 32-BITS: # %bb.0: # %entry
+; 32-BITS-NEXT: mul a0, a1, a0
+; 32-BITS-NEXT: or a0, a0, a2
+; 32-BITS-NEXT: li a1, 0
+; 32-BITS-NEXT: li a2, 0
+; 32-BITS-NEXT: ret
+
+entry:
+ %mul = mul nsw i32 %b, %a
+ %or = or i32 %mul, %c
+ ret i32 %or
+}
+
+define i32 @used(i32 noundef %a, i32 noundef %b, i32 noundef %c) #0 "zero-call-used-regs"="used" {
+; 64-BITS-LABEL: used:
+; 64-BITS: # %bb.0: # %entry
+; 64-BITS-NEXT: mulw a0, a1, a0
+; 64-BITS-NEXT: or a0, a0, a2
+; 64-BITS-NEXT: li a1, 0
+; 64-BITS-NEXT: li a2, 0
+; 64-BITS-NEXT: ret
+;
+; 32-BITS-LABEL: used:
+; 32-BITS: # %bb.0: # %entry
+; 32-BITS-NEXT: mul a0, a1, a0
+; 32-BITS-NEXT: or a0, a0, a2
+; 32-BITS-NEXT: li a1, 0
+; 32-BITS-NEXT: li a2, 0
+; 32-BITS-NEXT: ret
+
+entry:
+ %mul = mul nsw i32 %b, %a
+ %or = or i32 %mul, %c
+ ret i32 %or
+}
+
+define i32 @all_gpr_arg(i32 noundef %a, i32 noundef %b, i32 noundef %c) #0 "zero-call-used-regs"="all-gpr-arg" {
+; 64-BITS-LABEL: all_gpr_arg:
+; 64-BITS: # %bb.0: # %entry
+; 64-BITS-NEXT: mulw a0, a1, a0
+; 64-BITS-NEXT: or a0, a0, a2
+; 64-BITS-NEXT: li a1, 0
+; 64-BITS-NEXT: li a2, 0
+; 64-BITS-NEXT: li a3, 0
+; 64-BITS-NEXT: li a4, 0
+; 64-BITS-NEXT: li a5, 0
+; 64-BITS-NEXT: li a6, 0
+; 64-BITS-NEXT: li a7, 0
+; 64-BITS-NEXT: ret
+;
+; 32-BITS-LABEL: all_gpr_arg:
+; 32-BITS: # %bb.0: # %entry
+; 32-BITS-NEXT: mul a0, a1, a0
+; 32-BITS-NEXT: or a0, a0, a2
+; 32-BITS-NEXT: li a1, 0
+; 32-BITS-NEXT: li a2, 0
+; 32-BITS-NEXT: li a3, 0
+; 32-BITS-NEXT: li a4, 0
+; 32-BITS-NEXT: li a5, 0
+; 32-BITS-NEXT: li a6, 0
+; 32-BITS-NEXT: li a7, 0
+; 32-BITS-NEXT: ret
+
+entry:
+ %mul = mul nsw i32 %b, %a
+ %or = or i32 %mul, %c
+ ret i32 %or
+}
+
+define i32 @all_gpr(i32 noundef %a, i32 noundef %b, i32 noundef %c) #0 "zero-call-used-regs"="all-gpr" {
+; 64-BITS-LABEL: all_gpr:
+; 64-BITS: # %bb.0: # %entry
+; 64-BITS-NEXT: mulw a0, a1, a0
+; 64-BITS-NEXT: or a0, a0, a2
+; 64-BITS-NEXT: li t0, 0
+; 64-BITS-NEXT: li t1, 0
+; 64-BITS-NEXT: li t2, 0
+; 64-BITS-NEXT: li a1, 0
+; 64-BITS-NEXT: li a2, 0
+; 64-BITS-NEXT: li a3, 0
+; 64-BITS-NEXT: li a4, 0
+; 64-BITS-NEXT: li a5, 0
+; 64-BITS-NEXT: li a6, 0
+; 64-BITS-NEXT: li a7, 0
+; 64-BITS-NEXT: li t3, 0
+; 64-BITS-NEXT: li t4, 0
+; 64-BITS-NEXT: li t5, 0
+; 64-BITS-NEXT: li t6, 0
+; 64-BITS-NEXT: ret
+;
+; 32-BITS-LABEL: all_gpr:
+; 32-BITS: # %bb.0: # %entry
+; 32-BITS-NEXT: mul a0, a1, a0
+; 32-BITS-NEXT: or a0, a0, a2
+; 32-BITS-NEXT: li t0, 0
+; 32-BITS-NEXT: li t1, 0
+; 32-BITS-NEXT: li t2, 0
+; 32-BITS-NEXT: li a1, 0
+; 32-BITS-NEXT: li a2, 0
+; 32-BITS-NEXT: li a3, 0
+; 32-BITS-NEXT: li a4, 0
+; 32-BITS-NEXT: li a5, 0
+; 32-BITS-NEXT: li a6, 0
+; 32-BITS-NEXT: li a7, 0
+; 32-BITS-NEXT: li t3, 0
+; 32-BITS-NEXT: li t4, 0
+; 32-BITS-NEXT: li t5, 0
+; 32-BITS-NEXT: li t6, 0
+; 32-BITS-NEXT: ret
+
+entry:
+ %mul = mul nsw i32 %b, %a
+ %or = or i32 %mul, %c
+ ret i32 %or
+}
+
+define double @skip_float(double noundef %a, float noundef %b) #0 "zero-call-used-regs"="skip" {
+; CHECK-LABEL: skip_float:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: fcvt.d.s fa5, fa1
+; CHECK-NEXT: fmul.d fa0, fa5, fa0
+; CHECK-NEXT: ret
+
+entry:
+ %conv = fpext float %b to double
+ %mul = fmul double %conv, %a
+ ret double %mul
+}
+
+define double @used_gpr_arg_float(double noundef %a, float noundef %b) #0 "zero-call-used-regs"="used-gpr-arg" {
+; CHECK-LABEL: used_gpr_arg_float:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: fcvt.d.s fa5, fa1
+; CHECK-NEXT: fmul.d fa0, fa5, fa0
+; CHECK-NEXT: ret
+
+entry:
+ %conv = fpext float %b to double
+ %mul = fmul double %conv, %a
+ ret double %mul
+}
+
+define double @used_gpr_float(double noundef %a, float noundef %b) #0 "zero-call-used-regs"="used-gpr" {
+; CHECK-LABEL: used_gpr_float:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: fcvt.d.s fa5, fa1
+; CHECK-NEXT: fmul.d fa0, fa5, fa0
+; CHECK-NEXT: ret
+
+entry:
+ %conv = fpext float %b to double
+ %mul = fmul double %conv, %a
+ ret double %mul
+}
+
+define double @all_gpr_arg_float(double noundef %a, float noundef %b) #0 "zero-call-used-regs"="all-gpr-arg" {
+; CHECK-LABEL: all_gpr_arg_float:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: fcvt.d.s fa5, fa1
+; CHECK-NEXT: fmul.d fa0, fa5, fa0
+; CHECK-NEXT: li a0, 0
+; CHECK-NEXT: li a1, 0
+; CHECK-NEXT: li a2, 0
+; CHECK-NEXT: li a3, 0
+; CHECK-NEXT: li a4, 0
+; CHECK-NEXT: li a5, 0
+; CHECK-NEXT: li a6, 0
+; CHECK-NEXT: li a7, 0
+; CHECK-NEXT: ret
+
+entry:
+ %conv = fpext float %b to double
+ %mul = fmul double %conv, %a
+ ret double %mul
+}
+
+define double @all_gpr_float(double noundef %a, float noundef %b) #0 "zero-call-used-regs"="all-gpr" {
+; CHECK-LABEL: all_gpr_float:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: fcvt.d.s fa5, fa1
+; CHECK-NEXT: fmul.d fa0, fa5, fa0
+; CHECK-NEXT: li t0, 0
+; CHECK-NEXT: li t1, 0
+; CHECK-NEXT: li t2, 0
+; CHECK-NEXT: li a0, 0
+; CHECK-NEXT: li a1, 0
+; CHECK-NEXT: li a2, 0
+; CHECK-NEXT: li a3, 0
+; CHECK-NEXT: li a4, 0
+; CHECK-NEXT: li a5, 0
+; CHECK-NEXT: li a6, 0
+; CHECK-NEXT: li a7, 0
+; CHECK-NEXT: li t3, 0
+; CHECK-NEXT: li t4, 0
+; CHECK-NEXT: li t5, 0
+; CHECK-NEXT: li t6, 0
+; CHECK-NEXT: ret
+
+entry:
+ %conv = fpext float %b to double
+ %mul = fmul double %conv, %a
+ ret double %mul
+}
+
+; Don't emit zeroing registers in "main" function.
----------------
LucasChollet wrote:
Not sure if I mentioned that before, but this test file is heavily inspired from`test/CodeGen/AArch64/zero-call-used-regs.ll` (I just removed irrelevant tests and updated the assembly).
There is a special case for `main` in the clang frontend, but I fail to see how that test exercises it. From what I understand from the initial [review](https://reviews.llvm.org/D110869) of the implementation of -fzero-call-used-regs, the special case used to be in the backend, and it was moved to the frontend, I guess that the test was just not updated accordingly, and you're the first one to catch that.
I removed the test from this PR and opened #201834
https://github.com/llvm/llvm-project/pull/194883
More information about the cfe-commits
mailing list