blob: 9d9ce34fcec49fa6b3296891182f85b7124b4c2c [file]
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=riscv64-unknown-linux-gnu -mattr=+m | FileCheck %s
; Test optimization of 32-bit unsigned division by constants with 33-bit magic
; constants (IsAdd=true) on RISC-V64. The optimization uses the mulhu instruction.
define i32 @udiv_by_7(i32 %x) nounwind {
; CHECK-LABEL: udiv_by_7:
; CHECK: # %bb.0:
; CHECK-NEXT: lui a1, %hi(.LCPI0_0)
; CHECK-NEXT: ld a1, %lo(.LCPI0_0)(a1)
; CHECK-NEXT: slli a0, a0, 32
; CHECK-NEXT: srli a0, a0, 32
; CHECK-NEXT: mulhu a0, a0, a1
; CHECK-NEXT: ret
%div = udiv i32 %x, 7
ret i32 %div
}
define i32 @udiv_by_19(i32 %x) nounwind {
; CHECK-LABEL: udiv_by_19:
; CHECK: # %bb.0:
; CHECK-NEXT: lui a1, 717447
; CHECK-NEXT: addi a1, a1, -1077
; CHECK-NEXT: slli a0, a0, 32
; CHECK-NEXT: slli a1, a1, 31
; CHECK-NEXT: srli a0, a0, 32
; CHECK-NEXT: srli a1, a1, 4
; CHECK-NEXT: mulhu a0, a0, a1
; CHECK-NEXT: ret
%div = udiv i32 %x, 19
ret i32 %div
}
define i32 @udiv_by_21(i32 %x) nounwind {
; CHECK-LABEL: udiv_by_21:
; CHECK: # %bb.0:
; CHECK-NEXT: lui a1, 549254
; CHECK-NEXT: addi a1, a1, 391
; CHECK-NEXT: slli a0, a0, 32
; CHECK-NEXT: slli a1, a1, 31
; CHECK-NEXT: srli a0, a0, 32
; CHECK-NEXT: srli a1, a1, 4
; CHECK-NEXT: mulhu a0, a0, a1
; CHECK-NEXT: ret
%div = udiv i32 %x, 21
ret i32 %div
}
; Test non-optimized case
define i32 @udiv_by_3(i32 %x) nounwind {
; CHECK-LABEL: udiv_by_3:
; CHECK: # %bb.0:
; CHECK-NEXT: lui a1, 699051
; CHECK-NEXT: addi a1, a1, -1365
; CHECK-NEXT: slli a0, a0, 32
; CHECK-NEXT: slli a1, a1, 32
; CHECK-NEXT: mulhu a0, a0, a1
; CHECK-NEXT: srli a0, a0, 33
; CHECK-NEXT: ret
%div = udiv i32 %x, 3
ret i32 %div
}
; Even divisor with 33-bit magic. On RISC-V the widen path is not used (i32->i64
; zero-extension is not free); the even-divisor rewrite is kept instead.
define i32 @udiv_by_14(i32 %x) nounwind {
; CHECK-LABEL: udiv_by_14:
; CHECK: # %bb.0:
; CHECK-NEXT: lui a1, 299593
; CHECK-NEXT: slli a1, a1, 1
; CHECK-NEXT: srliw a0, a0, 1
; CHECK-NEXT: addi a1, a1, 1171
; CHECK-NEXT: mul a0, a0, a1
; CHECK-NEXT: srli a0, a0, 34
; CHECK-NEXT: ret
%div = udiv i32 %x, 14
ret i32 %div
}