ZhaoQi f9b080cf5f
[LoongArch] Custom legalizing ConstantFP to avoid float loads (#158050)
This commit custom legalize `ConstantFP` using code sequence
rather than simpily loading the fp values from constant pool.

A new option (`-loongarch-materialize-float-imm=<enum>`) is
added to set the maximum number of instructions (including
code sequence to generate the value and moving the value to
FPR) alllowed to be used when materializing floating-point
immediates.

The default value of the option is set to `3` on both LA32 and
LA64. Which means:

- For `f32` on both LA32 and LA64: `2 insts + movgr2fr.w`;
  (will cover all `f32` values)
- For `f64` on LA64: `2 insts + movgr2fr.d`;
- For `f64` on LA32: `1 inst + movgr2fr.w + movgr2frh.w`.
  (same inst latency as using constant pool)

The option can be set in range `0,2-6`. (6 behaves same
as 5 on LA64.)
2025-09-22 18:05:16 +08:00

155 lines
5.3 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc --mtriple=loongarch32 --mattr=+f,-d -loongarch-materialize-float-imm=0 < %s | FileCheck %s --check-prefixes=LA32,LA32-0
; RUN: llc --mtriple=loongarch32 --mattr=+f,-d -loongarch-materialize-float-imm=2 < %s | FileCheck %s --check-prefixes=LA32,LA32-2
; RUN: llc --mtriple=loongarch32 --mattr=+f,-d -loongarch-materialize-float-imm=3 < %s | FileCheck %s --check-prefixes=LA32,LA32-3
; RUN: llc --mtriple=loongarch32 --mattr=+f,-d -loongarch-materialize-float-imm=4 < %s | FileCheck %s --check-prefixes=LA32,LA32-3
; RUN: llc --mtriple=loongarch32 --mattr=+f,-d -loongarch-materialize-float-imm=5 < %s | FileCheck %s --check-prefixes=LA32,LA32-3
; RUN: llc --mtriple=loongarch32 --mattr=+f,-d -loongarch-materialize-float-imm=6 < %s | FileCheck %s --check-prefixes=LA32,LA32-3
; RUN: llc --mtriple=loongarch64 --mattr=+f,-d -loongarch-materialize-float-imm=0 < %s | FileCheck %s --check-prefixes=LA64,LA64-0
; RUN: llc --mtriple=loongarch64 --mattr=+f,-d -loongarch-materialize-float-imm=2 < %s | FileCheck %s --check-prefixes=LA64,LA64-2
; RUN: llc --mtriple=loongarch64 --mattr=+f,-d -loongarch-materialize-float-imm=3 < %s | FileCheck %s --check-prefixes=LA64,LA64-3
; RUN: llc --mtriple=loongarch64 --mattr=+f,-d -loongarch-materialize-float-imm=4 < %s | FileCheck %s --check-prefixes=LA64,LA64-3
; RUN: llc --mtriple=loongarch64 --mattr=+f,-d -loongarch-materialize-float-imm=5 < %s | FileCheck %s --check-prefixes=LA64,LA64-3
; RUN: llc --mtriple=loongarch64 --mattr=+f,-d -loongarch-materialize-float-imm=6 < %s | FileCheck %s --check-prefixes=LA64,LA64-3
define float @f32_positive_zero() nounwind {
; LA32-LABEL: f32_positive_zero:
; LA32: # %bb.0:
; LA32-NEXT: movgr2fr.w $fa0, $zero
; LA32-NEXT: ret
;
; LA64-LABEL: f32_positive_zero:
; LA64: # %bb.0:
; LA64-NEXT: movgr2fr.w $fa0, $zero
; LA64-NEXT: ret
ret float 0.0
}
define float @f32_negative_zero() nounwind {
; LA32-LABEL: f32_negative_zero:
; LA32: # %bb.0:
; LA32-NEXT: movgr2fr.w $fa0, $zero
; LA32-NEXT: fneg.s $fa0, $fa0
; LA32-NEXT: ret
;
; LA64-LABEL: f32_negative_zero:
; LA64: # %bb.0:
; LA64-NEXT: movgr2fr.w $fa0, $zero
; LA64-NEXT: fneg.s $fa0, $fa0
; LA64-NEXT: ret
ret float -0.0
}
define float @f32_constant_ins1() nounwind {
; LA32-0-LABEL: f32_constant_ins1:
; LA32-0: # %bb.0:
; LA32-0-NEXT: pcalau12i $a0, %pc_hi20(.LCPI2_0)
; LA32-0-NEXT: fld.s $fa0, $a0, %pc_lo12(.LCPI2_0)
; LA32-0-NEXT: ret
;
; LA32-2-LABEL: f32_constant_ins1:
; LA32-2: # %bb.0:
; LA32-2-NEXT: lu12i.w $a0, 270336
; LA32-2-NEXT: movgr2fr.w $fa0, $a0
; LA32-2-NEXT: ret
;
; LA32-3-LABEL: f32_constant_ins1:
; LA32-3: # %bb.0:
; LA32-3-NEXT: lu12i.w $a0, 270336
; LA32-3-NEXT: movgr2fr.w $fa0, $a0
; LA32-3-NEXT: ret
;
; LA64-0-LABEL: f32_constant_ins1:
; LA64-0: # %bb.0:
; LA64-0-NEXT: pcalau12i $a0, %pc_hi20(.LCPI2_0)
; LA64-0-NEXT: fld.s $fa0, $a0, %pc_lo12(.LCPI2_0)
; LA64-0-NEXT: ret
;
; LA64-2-LABEL: f32_constant_ins1:
; LA64-2: # %bb.0:
; LA64-2-NEXT: lu12i.w $a0, 270336
; LA64-2-NEXT: movgr2fr.w $fa0, $a0
; LA64-2-NEXT: ret
;
; LA64-3-LABEL: f32_constant_ins1:
; LA64-3: # %bb.0:
; LA64-3-NEXT: lu12i.w $a0, 270336
; LA64-3-NEXT: movgr2fr.w $fa0, $a0
; LA64-3-NEXT: ret
ret float 32.0
}
define float @f32_constant_pi() nounwind {
; LA32-0-LABEL: f32_constant_pi:
; LA32-0: # %bb.0:
; LA32-0-NEXT: pcalau12i $a0, %pc_hi20(.LCPI3_0)
; LA32-0-NEXT: fld.s $fa0, $a0, %pc_lo12(.LCPI3_0)
; LA32-0-NEXT: ret
;
; LA32-2-LABEL: f32_constant_pi:
; LA32-2: # %bb.0:
; LA32-2-NEXT: pcalau12i $a0, %pc_hi20(.LCPI3_0)
; LA32-2-NEXT: fld.s $fa0, $a0, %pc_lo12(.LCPI3_0)
; LA32-2-NEXT: ret
;
; LA32-3-LABEL: f32_constant_pi:
; LA32-3: # %bb.0:
; LA32-3-NEXT: lu12i.w $a0, 263312
; LA32-3-NEXT: ori $a0, $a0, 4059
; LA32-3-NEXT: movgr2fr.w $fa0, $a0
; LA32-3-NEXT: ret
;
; LA64-0-LABEL: f32_constant_pi:
; LA64-0: # %bb.0:
; LA64-0-NEXT: pcalau12i $a0, %pc_hi20(.LCPI3_0)
; LA64-0-NEXT: fld.s $fa0, $a0, %pc_lo12(.LCPI3_0)
; LA64-0-NEXT: ret
;
; LA64-2-LABEL: f32_constant_pi:
; LA64-2: # %bb.0:
; LA64-2-NEXT: pcalau12i $a0, %pc_hi20(.LCPI3_0)
; LA64-2-NEXT: fld.s $fa0, $a0, %pc_lo12(.LCPI3_0)
; LA64-2-NEXT: ret
;
; LA64-3-LABEL: f32_constant_pi:
; LA64-3: # %bb.0:
; LA64-3-NEXT: lu12i.w $a0, 263312
; LA64-3-NEXT: ori $a0, $a0, 4059
; LA64-3-NEXT: movgr2fr.w $fa0, $a0
; LA64-3-NEXT: ret
ret float 3.14159274101257324218750
}
define float @f32_add_fimm1(float %a) nounwind {
; LA32-LABEL: f32_add_fimm1:
; LA32: # %bb.0:
; LA32-NEXT: lu12i.w $a0, 260096
; LA32-NEXT: movgr2fr.w $fa1, $a0
; LA32-NEXT: fadd.s $fa0, $fa0, $fa1
; LA32-NEXT: ret
;
; LA64-LABEL: f32_add_fimm1:
; LA64: # %bb.0:
; LA64-NEXT: lu12i.w $a0, 260096
; LA64-NEXT: movgr2fr.w $fa1, $a0
; LA64-NEXT: fadd.s $fa0, $fa0, $fa1
; LA64-NEXT: ret
%1 = fadd float %a, 1.0
ret float %1
}
define float @f32_positive_fimm1() nounwind {
; LA32-LABEL: f32_positive_fimm1:
; LA32: # %bb.0:
; LA32-NEXT: lu12i.w $a0, 260096
; LA32-NEXT: movgr2fr.w $fa0, $a0
; LA32-NEXT: ret
;
; LA64-LABEL: f32_positive_fimm1:
; LA64: # %bb.0:
; LA64-NEXT: lu12i.w $a0, 260096
; LA64-NEXT: movgr2fr.w $fa0, $a0
; LA64-NEXT: ret
ret float 1.0
}