shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
llvm-project/llvm/test/CodeGen/RISCV/shrinkwrap.ll
Philip Reames 06246b2952 [RISCV] Add shrinkwrap test cases showing gaps in current impl 
This covers multiple interactions reduced from larger workloads:
1) Rematerializing addi s0, x0, <imm> to avoid the need to spill a CSR, with
   the common user being a branch.  (i.e. branch on immediate idioms)
2) Rematerializing addi s0, a0, <imm> to avoid the need to spill a CSR, with
   the common user being a vector load or store.  (i.e. because we don't
   have (reg+imm) addressing on vector.)
3) Independent of the previous, we could still shrink wrap these by locally
   using a non-CSR, and deferring the move into csr into the non-shrink
   wrapped path.
4) Weirdly, MachineCSE is producing a different result when an edge is
   manually split.  This edge split should be irrelevant to the CSE?
2024-11-27 11:14:51 -08:00

625 lines
22 KiB
LLVM
Raw Permalink Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple riscv32 -enable-shrink-wrap=false < %s \
; RUN: | FileCheck %s -check-prefix=RV32I-SW-NO
; RUN: llc -mtriple riscv32 < %s \
; RUN: | FileCheck %s -check-prefix=RV32I-SW
; RUN: llc -mtriple riscv32 -mattr=+save-restore < %s \
; RUN: | FileCheck %s -check-prefix=RV32I-SW-SR
; RUN: llc -mtriple riscv64 < %s \
; RUN: | FileCheck %s -check-prefix=RV64I-SW
declare void @abort()
define void @eliminate_restore(i32 %n) nounwind {
; RV32I-SW-NO-LABEL: eliminate_restore:
; RV32I-SW-NO: # %bb.0:
; RV32I-SW-NO-NEXT: addi sp, sp, -16
; RV32I-SW-NO-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
; RV32I-SW-NO-NEXT: li a1, 32
; RV32I-SW-NO-NEXT: bgeu a1, a0, .LBB0_2
; RV32I-SW-NO-NEXT: # %bb.1: # %if.end
; RV32I-SW-NO-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
; RV32I-SW-NO-NEXT: addi sp, sp, 16
; RV32I-SW-NO-NEXT: ret
; RV32I-SW-NO-NEXT: .LBB0_2: # %if.then
; RV32I-SW-NO-NEXT: call abort
;
; RV32I-SW-LABEL: eliminate_restore:
; RV32I-SW: # %bb.0:
; RV32I-SW-NEXT: li a1, 32
; RV32I-SW-NEXT: bgeu a1, a0, .LBB0_2
; RV32I-SW-NEXT: # %bb.1: # %if.end
; RV32I-SW-NEXT: ret
; RV32I-SW-NEXT: .LBB0_2: # %if.then
; RV32I-SW-NEXT: addi sp, sp, -16
; RV32I-SW-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
; RV32I-SW-NEXT: call abort
;
; RV32I-SW-SR-LABEL: eliminate_restore:
; RV32I-SW-SR: # %bb.0:
; RV32I-SW-SR-NEXT: li a1, 32
; RV32I-SW-SR-NEXT: bgeu a1, a0, .LBB0_2
; RV32I-SW-SR-NEXT: # %bb.1: # %if.end
; RV32I-SW-SR-NEXT: ret
; RV32I-SW-SR-NEXT: .LBB0_2: # %if.then
; RV32I-SW-SR-NEXT: call t0, __riscv_save_0
; RV32I-SW-SR-NEXT: call abort
;
; RV64I-SW-LABEL: eliminate_restore:
; RV64I-SW: # %bb.0:
; RV64I-SW-NEXT: sext.w a0, a0
; RV64I-SW-NEXT: li a1, 32
; RV64I-SW-NEXT: bgeu a1, a0, .LBB0_2
; RV64I-SW-NEXT: # %bb.1: # %if.end
; RV64I-SW-NEXT: ret
; RV64I-SW-NEXT: .LBB0_2: # %if.then
; RV64I-SW-NEXT: addi sp, sp, -16
; RV64I-SW-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
; RV64I-SW-NEXT: call abort
%cmp = icmp ule i32 %n, 32
br i1 %cmp, label %if.then, label %if.end
if.then:
call void @abort()
unreachable
if.end:
ret void
}
declare void @notdead(ptr)
define void @conditional_alloca(i32 %n) nounwind {
; RV32I-SW-NO-LABEL: conditional_alloca:
; RV32I-SW-NO: # %bb.0:
; RV32I-SW-NO-NEXT: addi sp, sp, -16
; RV32I-SW-NO-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
; RV32I-SW-NO-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
; RV32I-SW-NO-NEXT: addi s0, sp, 16
; RV32I-SW-NO-NEXT: li a1, 32
; RV32I-SW-NO-NEXT: bltu a1, a0, .LBB1_2
; RV32I-SW-NO-NEXT: # %bb.1: # %if.then
; RV32I-SW-NO-NEXT: addi a0, a0, 15
; RV32I-SW-NO-NEXT: andi a0, a0, -16
; RV32I-SW-NO-NEXT: sub a0, sp, a0
; RV32I-SW-NO-NEXT: mv sp, a0
; RV32I-SW-NO-NEXT: call notdead
; RV32I-SW-NO-NEXT: .LBB1_2: # %if.end
; RV32I-SW-NO-NEXT: addi sp, s0, -16
; RV32I-SW-NO-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
; RV32I-SW-NO-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
; RV32I-SW-NO-NEXT: addi sp, sp, 16
; RV32I-SW-NO-NEXT: ret
;
; RV32I-SW-LABEL: conditional_alloca:
; RV32I-SW: # %bb.0:
; RV32I-SW-NEXT: li a1, 32
; RV32I-SW-NEXT: bltu a1, a0, .LBB1_2
; RV32I-SW-NEXT: # %bb.1: # %if.then
; RV32I-SW-NEXT: addi sp, sp, -16
; RV32I-SW-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
; RV32I-SW-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
; RV32I-SW-NEXT: addi s0, sp, 16
; RV32I-SW-NEXT: addi a0, a0, 15
; RV32I-SW-NEXT: andi a0, a0, -16
; RV32I-SW-NEXT: sub a0, sp, a0
; RV32I-SW-NEXT: mv sp, a0
; RV32I-SW-NEXT: call notdead
; RV32I-SW-NEXT: addi sp, s0, -16
; RV32I-SW-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
; RV32I-SW-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
; RV32I-SW-NEXT: addi sp, sp, 16
; RV32I-SW-NEXT: .LBB1_2: # %if.end
; RV32I-SW-NEXT: ret
;
; RV32I-SW-SR-LABEL: conditional_alloca:
; RV32I-SW-SR: # %bb.0:
; RV32I-SW-SR-NEXT: li a1, 32
; RV32I-SW-SR-NEXT: bltu a1, a0, .LBB1_2
; RV32I-SW-SR-NEXT: # %bb.1: # %if.then
; RV32I-SW-SR-NEXT: call t0, __riscv_save_1
; RV32I-SW-SR-NEXT: addi s0, sp, 16
; RV32I-SW-SR-NEXT: addi a0, a0, 15
; RV32I-SW-SR-NEXT: andi a0, a0, -16
; RV32I-SW-SR-NEXT: sub a0, sp, a0
; RV32I-SW-SR-NEXT: mv sp, a0
; RV32I-SW-SR-NEXT: call notdead
; RV32I-SW-SR-NEXT: addi sp, s0, -16
; RV32I-SW-SR-NEXT: tail __riscv_restore_1
; RV32I-SW-SR-NEXT: .LBB1_2: # %if.end
; RV32I-SW-SR-NEXT: ret
;
; RV64I-SW-LABEL: conditional_alloca:
; RV64I-SW: # %bb.0:
; RV64I-SW-NEXT: sext.w a1, a0
; RV64I-SW-NEXT: li a2, 32
; RV64I-SW-NEXT: bltu a2, a1, .LBB1_2
; RV64I-SW-NEXT: # %bb.1: # %if.then
; RV64I-SW-NEXT: addi sp, sp, -16
; RV64I-SW-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
; RV64I-SW-NEXT: sd s0, 0(sp) # 8-byte Folded Spill
; RV64I-SW-NEXT: addi s0, sp, 16
; RV64I-SW-NEXT: slli a0, a0, 32
; RV64I-SW-NEXT: srli a0, a0, 32
; RV64I-SW-NEXT: addi a0, a0, 15
; RV64I-SW-NEXT: andi a0, a0, -16
; RV64I-SW-NEXT: sub a0, sp, a0
; RV64I-SW-NEXT: mv sp, a0
; RV64I-SW-NEXT: call notdead
; RV64I-SW-NEXT: addi sp, s0, -16
; RV64I-SW-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
; RV64I-SW-NEXT: ld s0, 0(sp) # 8-byte Folded Reload
; RV64I-SW-NEXT: addi sp, sp, 16
; RV64I-SW-NEXT: .LBB1_2: # %if.end
; RV64I-SW-NEXT: ret
%cmp = icmp ule i32 %n, 32
br i1 %cmp, label %if.then, label %if.end
if.then:
%addr = alloca i8, i32 %n
call void @notdead(ptr %addr)
br label %if.end
if.end:
ret void
}
; FIXME: Rematerialize "li s1, 57" for the second use instead of spilling s1,
; and then shrink wrap the result avoiding frame setup on the %bb.0 -> %exit
; edge.
define void @li_straightline_a(i32 zeroext %a, i32 zeroext %b) {
; RV32I-SW-NO-LABEL: li_straightline_a:
; RV32I-SW-NO: # %bb.0:
; RV32I-SW-NO-NEXT: addi sp, sp, -16
; RV32I-SW-NO-NEXT: .cfi_def_cfa_offset 16
; RV32I-SW-NO-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
; RV32I-SW-NO-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
; RV32I-SW-NO-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
; RV32I-SW-NO-NEXT: .cfi_offset ra, -4
; RV32I-SW-NO-NEXT: .cfi_offset s0, -8
; RV32I-SW-NO-NEXT: .cfi_offset s1, -12
; RV32I-SW-NO-NEXT: li s1, 57
; RV32I-SW-NO-NEXT: beq a0, s1, .LBB2_3
; RV32I-SW-NO-NEXT: # %bb.1: # %do_call
; RV32I-SW-NO-NEXT: mv s0, a1
; RV32I-SW-NO-NEXT: call foo
; RV32I-SW-NO-NEXT: beq s0, s1, .LBB2_3
; RV32I-SW-NO-NEXT: # %bb.2: # %do_call2
; RV32I-SW-NO-NEXT: call foo
; RV32I-SW-NO-NEXT: .LBB2_3: # %exit
; RV32I-SW-NO-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
; RV32I-SW-NO-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
; RV32I-SW-NO-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
; RV32I-SW-NO-NEXT: .cfi_restore ra
; RV32I-SW-NO-NEXT: .cfi_restore s0
; RV32I-SW-NO-NEXT: .cfi_restore s1
; RV32I-SW-NO-NEXT: addi sp, sp, 16
; RV32I-SW-NO-NEXT: .cfi_def_cfa_offset 0
; RV32I-SW-NO-NEXT: ret
;
; RV32I-SW-LABEL: li_straightline_a:
; RV32I-SW: # %bb.0:
; RV32I-SW-NEXT: addi sp, sp, -16
; RV32I-SW-NEXT: .cfi_def_cfa_offset 16
; RV32I-SW-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
; RV32I-SW-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
; RV32I-SW-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
; RV32I-SW-NEXT: .cfi_offset ra, -4
; RV32I-SW-NEXT: .cfi_offset s0, -8
; RV32I-SW-NEXT: .cfi_offset s1, -12
; RV32I-SW-NEXT: li s1, 57
; RV32I-SW-NEXT: beq a0, s1, .LBB2_3
; RV32I-SW-NEXT: # %bb.1: # %do_call
; RV32I-SW-NEXT: mv s0, a1
; RV32I-SW-NEXT: call foo
; RV32I-SW-NEXT: beq s0, s1, .LBB2_3
; RV32I-SW-NEXT: # %bb.2: # %do_call2
; RV32I-SW-NEXT: call foo
; RV32I-SW-NEXT: .LBB2_3: # %exit
; RV32I-SW-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
; RV32I-SW-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
; RV32I-SW-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
; RV32I-SW-NEXT: .cfi_restore ra
; RV32I-SW-NEXT: .cfi_restore s0
; RV32I-SW-NEXT: .cfi_restore s1
; RV32I-SW-NEXT: addi sp, sp, 16
; RV32I-SW-NEXT: .cfi_def_cfa_offset 0
; RV32I-SW-NEXT: ret
;
; RV32I-SW-SR-LABEL: li_straightline_a:
; RV32I-SW-SR: # %bb.0:
; RV32I-SW-SR-NEXT: call t0, __riscv_save_2
; RV32I-SW-SR-NEXT: .cfi_def_cfa_offset 16
; RV32I-SW-SR-NEXT: .cfi_offset ra, -4
; RV32I-SW-SR-NEXT: .cfi_offset s0, -8
; RV32I-SW-SR-NEXT: .cfi_offset s1, -12
; RV32I-SW-SR-NEXT: li s1, 57
; RV32I-SW-SR-NEXT: beq a0, s1, .LBB2_3
; RV32I-SW-SR-NEXT: # %bb.1: # %do_call
; RV32I-SW-SR-NEXT: mv s0, a1
; RV32I-SW-SR-NEXT: call foo
; RV32I-SW-SR-NEXT: beq s0, s1, .LBB2_3
; RV32I-SW-SR-NEXT: # %bb.2: # %do_call2
; RV32I-SW-SR-NEXT: call foo
; RV32I-SW-SR-NEXT: .LBB2_3: # %exit
; RV32I-SW-SR-NEXT: tail __riscv_restore_2
;
; RV64I-SW-LABEL: li_straightline_a:
; RV64I-SW: # %bb.0:
; RV64I-SW-NEXT: addi sp, sp, -32
; RV64I-SW-NEXT: .cfi_def_cfa_offset 32
; RV64I-SW-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
; RV64I-SW-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
; RV64I-SW-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
; RV64I-SW-NEXT: .cfi_offset ra, -8
; RV64I-SW-NEXT: .cfi_offset s0, -16
; RV64I-SW-NEXT: .cfi_offset s1, -24
; RV64I-SW-NEXT: li s1, 57
; RV64I-SW-NEXT: beq a0, s1, .LBB2_3
; RV64I-SW-NEXT: # %bb.1: # %do_call
; RV64I-SW-NEXT: mv s0, a1
; RV64I-SW-NEXT: call foo
; RV64I-SW-NEXT: beq s0, s1, .LBB2_3
; RV64I-SW-NEXT: # %bb.2: # %do_call2
; RV64I-SW-NEXT: call foo
; RV64I-SW-NEXT: .LBB2_3: # %exit
; RV64I-SW-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
; RV64I-SW-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
; RV64I-SW-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
; RV64I-SW-NEXT: .cfi_restore ra
; RV64I-SW-NEXT: .cfi_restore s0
; RV64I-SW-NEXT: .cfi_restore s1
; RV64I-SW-NEXT: addi sp, sp, 32
; RV64I-SW-NEXT: .cfi_def_cfa_offset 0
; RV64I-SW-NEXT: ret
%cmp0 = icmp eq i32 %a, 57
br i1 %cmp0, label %exit, label %do_call
do_call:
call i32 @foo()
%cmp1 = icmp eq i32 %b, 57
br i1 %cmp1, label %exit, label %do_call2
do_call2:
call i32 @foo()
br label %exit
exit:
ret void
}
; FIXME: The "mv s0, a1" is only required along the do_call path, and can
; be sunk there. We can also shrink wrap to avoid the frame setup along
; the %bb.0 -> %exit edge.
define void @li_straightline_b(i32 zeroext %a, i32 zeroext %b) {
; RV32I-SW-NO-LABEL: li_straightline_b:
; RV32I-SW-NO: # %bb.0:
; RV32I-SW-NO-NEXT: addi sp, sp, -16
; RV32I-SW-NO-NEXT: .cfi_def_cfa_offset 16
; RV32I-SW-NO-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
; RV32I-SW-NO-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
; RV32I-SW-NO-NEXT: .cfi_offset ra, -4
; RV32I-SW-NO-NEXT: .cfi_offset s0, -8
; RV32I-SW-NO-NEXT: mv s0, a1
; RV32I-SW-NO-NEXT: li a1, 57
; RV32I-SW-NO-NEXT: beq a0, a1, .LBB3_3
; RV32I-SW-NO-NEXT: # %bb.1: # %do_call
; RV32I-SW-NO-NEXT: call foo
; RV32I-SW-NO-NEXT: li a0, 57
; RV32I-SW-NO-NEXT: beq s0, a0, .LBB3_3
; RV32I-SW-NO-NEXT: # %bb.2: # %do_call2
; RV32I-SW-NO-NEXT: call foo
; RV32I-SW-NO-NEXT: .LBB3_3: # %exit
; RV32I-SW-NO-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
; RV32I-SW-NO-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
; RV32I-SW-NO-NEXT: .cfi_restore ra
; RV32I-SW-NO-NEXT: .cfi_restore s0
; RV32I-SW-NO-NEXT: addi sp, sp, 16
; RV32I-SW-NO-NEXT: .cfi_def_cfa_offset 0
; RV32I-SW-NO-NEXT: ret
;
; RV32I-SW-LABEL: li_straightline_b:
; RV32I-SW: # %bb.0:
; RV32I-SW-NEXT: addi sp, sp, -16
; RV32I-SW-NEXT: .cfi_def_cfa_offset 16
; RV32I-SW-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
; RV32I-SW-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
; RV32I-SW-NEXT: .cfi_offset ra, -4
; RV32I-SW-NEXT: .cfi_offset s0, -8
; RV32I-SW-NEXT: mv s0, a1
; RV32I-SW-NEXT: li a1, 57
; RV32I-SW-NEXT: beq a0, a1, .LBB3_3
; RV32I-SW-NEXT: # %bb.1: # %do_call
; RV32I-SW-NEXT: call foo
; RV32I-SW-NEXT: li a0, 57
; RV32I-SW-NEXT: beq s0, a0, .LBB3_3
; RV32I-SW-NEXT: # %bb.2: # %do_call2
; RV32I-SW-NEXT: call foo
; RV32I-SW-NEXT: .LBB3_3: # %exit
; RV32I-SW-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
; RV32I-SW-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
; RV32I-SW-NEXT: .cfi_restore ra
; RV32I-SW-NEXT: .cfi_restore s0
; RV32I-SW-NEXT: addi sp, sp, 16
; RV32I-SW-NEXT: .cfi_def_cfa_offset 0
; RV32I-SW-NEXT: ret
;
; RV32I-SW-SR-LABEL: li_straightline_b:
; RV32I-SW-SR: # %bb.0:
; RV32I-SW-SR-NEXT: call t0, __riscv_save_1
; RV32I-SW-SR-NEXT: .cfi_def_cfa_offset 16
; RV32I-SW-SR-NEXT: .cfi_offset ra, -4
; RV32I-SW-SR-NEXT: .cfi_offset s0, -8
; RV32I-SW-SR-NEXT: mv s0, a1
; RV32I-SW-SR-NEXT: li a1, 57
; RV32I-SW-SR-NEXT: beq a0, a1, .LBB3_3
; RV32I-SW-SR-NEXT: # %bb.1: # %do_call
; RV32I-SW-SR-NEXT: call foo
; RV32I-SW-SR-NEXT: li a0, 57
; RV32I-SW-SR-NEXT: beq s0, a0, .LBB3_3
; RV32I-SW-SR-NEXT: # %bb.2: # %do_call2
; RV32I-SW-SR-NEXT: call foo
; RV32I-SW-SR-NEXT: .LBB3_3: # %exit
; RV32I-SW-SR-NEXT: tail __riscv_restore_1
;
; RV64I-SW-LABEL: li_straightline_b:
; RV64I-SW: # %bb.0:
; RV64I-SW-NEXT: addi sp, sp, -16
; RV64I-SW-NEXT: .cfi_def_cfa_offset 16
; RV64I-SW-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
; RV64I-SW-NEXT: sd s0, 0(sp) # 8-byte Folded Spill
; RV64I-SW-NEXT: .cfi_offset ra, -8
; RV64I-SW-NEXT: .cfi_offset s0, -16
; RV64I-SW-NEXT: mv s0, a1
; RV64I-SW-NEXT: li a1, 57
; RV64I-SW-NEXT: beq a0, a1, .LBB3_3
; RV64I-SW-NEXT: # %bb.1: # %do_call
; RV64I-SW-NEXT: call foo
; RV64I-SW-NEXT: li a0, 57
; RV64I-SW-NEXT: beq s0, a0, .LBB3_3
; RV64I-SW-NEXT: # %bb.2: # %do_call2
; RV64I-SW-NEXT: call foo
; RV64I-SW-NEXT: .LBB3_3: # %exit
; RV64I-SW-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
; RV64I-SW-NEXT: ld s0, 0(sp) # 8-byte Folded Reload
; RV64I-SW-NEXT: .cfi_restore ra
; RV64I-SW-NEXT: .cfi_restore s0
; RV64I-SW-NEXT: addi sp, sp, 16
; RV64I-SW-NEXT: .cfi_def_cfa_offset 0
; RV64I-SW-NEXT: ret
%cmp0 = icmp eq i32 %a, 57
br i1 %cmp0, label %exit, label %do_call
do_call:
call i32 @foo()
br label %next
next:
%cmp1 = icmp eq i32 %b, 57
br i1 %cmp1, label %exit, label %do_call2
do_call2:
call i32 @foo()
br label %exit
exit:
ret void
}
; A further variant of "b" to show that we can shrink wrap this case
; if the second use of 57 is remat, and we eliminate the second live
; argument.
define void @li_straightline_c(i32 zeroext %a) {
; RV32I-SW-NO-LABEL: li_straightline_c:
; RV32I-SW-NO: # %bb.0:
; RV32I-SW-NO-NEXT: addi sp, sp, -16
; RV32I-SW-NO-NEXT: .cfi_def_cfa_offset 16
; RV32I-SW-NO-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
; RV32I-SW-NO-NEXT: .cfi_offset ra, -4
; RV32I-SW-NO-NEXT: li a1, 57
; RV32I-SW-NO-NEXT: beq a0, a1, .LBB4_3
; RV32I-SW-NO-NEXT: # %bb.1: # %do_call
; RV32I-SW-NO-NEXT: call foo
; RV32I-SW-NO-NEXT: li a1, 57
; RV32I-SW-NO-NEXT: beq a0, a1, .LBB4_3
; RV32I-SW-NO-NEXT: # %bb.2: # %do_call2
; RV32I-SW-NO-NEXT: call foo
; RV32I-SW-NO-NEXT: .LBB4_3: # %exit
; RV32I-SW-NO-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
; RV32I-SW-NO-NEXT: .cfi_restore ra
; RV32I-SW-NO-NEXT: addi sp, sp, 16
; RV32I-SW-NO-NEXT: .cfi_def_cfa_offset 0
; RV32I-SW-NO-NEXT: ret
;
; RV32I-SW-LABEL: li_straightline_c:
; RV32I-SW: # %bb.0:
; RV32I-SW-NEXT: li a1, 57
; RV32I-SW-NEXT: beq a0, a1, .LBB4_4
; RV32I-SW-NEXT: # %bb.1: # %do_call
; RV32I-SW-NEXT: addi sp, sp, -16
; RV32I-SW-NEXT: .cfi_def_cfa_offset 16
; RV32I-SW-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
; RV32I-SW-NEXT: .cfi_offset ra, -4
; RV32I-SW-NEXT: call foo
; RV32I-SW-NEXT: li a1, 57
; RV32I-SW-NEXT: beq a0, a1, .LBB4_3
; RV32I-SW-NEXT: # %bb.2: # %do_call2
; RV32I-SW-NEXT: call foo
; RV32I-SW-NEXT: .LBB4_3:
; RV32I-SW-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
; RV32I-SW-NEXT: .cfi_restore ra
; RV32I-SW-NEXT: addi sp, sp, 16
; RV32I-SW-NEXT: .cfi_def_cfa_offset 0
; RV32I-SW-NEXT: .LBB4_4: # %exit
; RV32I-SW-NEXT: ret
;
; RV32I-SW-SR-LABEL: li_straightline_c:
; RV32I-SW-SR: # %bb.0:
; RV32I-SW-SR-NEXT: li a1, 57
; RV32I-SW-SR-NEXT: beq a0, a1, .LBB4_4
; RV32I-SW-SR-NEXT: # %bb.1: # %do_call
; RV32I-SW-SR-NEXT: call t0, __riscv_save_0
; RV32I-SW-SR-NEXT: .cfi_def_cfa_offset 16
; RV32I-SW-SR-NEXT: .cfi_offset ra, -4
; RV32I-SW-SR-NEXT: call foo
; RV32I-SW-SR-NEXT: li a1, 57
; RV32I-SW-SR-NEXT: beq a0, a1, .LBB4_3
; RV32I-SW-SR-NEXT: # %bb.2: # %do_call2
; RV32I-SW-SR-NEXT: call foo
; RV32I-SW-SR-NEXT: .LBB4_3:
; RV32I-SW-SR-NEXT: tail __riscv_restore_0
; RV32I-SW-SR-NEXT: j .LBB4_4
; RV32I-SW-SR-NEXT: .LBB4_4: # %exit
; RV32I-SW-SR-NEXT: ret
;
; RV64I-SW-LABEL: li_straightline_c:
; RV64I-SW: # %bb.0:
; RV64I-SW-NEXT: li a1, 57
; RV64I-SW-NEXT: beq a0, a1, .LBB4_4
; RV64I-SW-NEXT: # %bb.1: # %do_call
; RV64I-SW-NEXT: addi sp, sp, -16
; RV64I-SW-NEXT: .cfi_def_cfa_offset 16
; RV64I-SW-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
; RV64I-SW-NEXT: .cfi_offset ra, -8
; RV64I-SW-NEXT: call foo
; RV64I-SW-NEXT: li a1, 57
; RV64I-SW-NEXT: beq a0, a1, .LBB4_3
; RV64I-SW-NEXT: # %bb.2: # %do_call2
; RV64I-SW-NEXT: call foo
; RV64I-SW-NEXT: .LBB4_3:
; RV64I-SW-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
; RV64I-SW-NEXT: .cfi_restore ra
; RV64I-SW-NEXT: addi sp, sp, 16
; RV64I-SW-NEXT: .cfi_def_cfa_offset 0
; RV64I-SW-NEXT: .LBB4_4: # %exit
; RV64I-SW-NEXT: ret
%cmp0 = icmp eq i32 %a, 57
br i1 %cmp0, label %exit, label %do_call
do_call:
%b = call i32 @foo()
br label %next
next:
%cmp1 = icmp eq i32 %b, 57
br i1 %cmp1, label %exit, label %do_call2
do_call2:
call i32 @foo()
br label %exit
exit:
ret void
}
; In this case, the second use is in a loop, so using a callee
; saved register to avoid a remat is the profitable choice.
; FIXME: We can shrink wrap the frame setup around the loop
; and avoid it along the %bb.0 -> %exit edge
define void @li_loop(i32 zeroext %a, i32 zeroext %b) {
; RV32I-SW-NO-LABEL: li_loop:
; RV32I-SW-NO: # %bb.0:
; RV32I-SW-NO-NEXT: addi sp, sp, -16
; RV32I-SW-NO-NEXT: .cfi_def_cfa_offset 16
; RV32I-SW-NO-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
; RV32I-SW-NO-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
; RV32I-SW-NO-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
; RV32I-SW-NO-NEXT: .cfi_offset ra, -4
; RV32I-SW-NO-NEXT: .cfi_offset s0, -8
; RV32I-SW-NO-NEXT: .cfi_offset s1, -12
; RV32I-SW-NO-NEXT: li s1, 57
; RV32I-SW-NO-NEXT: beq a0, s1, .LBB5_3
; RV32I-SW-NO-NEXT: # %bb.1: # %do_call.preheader
; RV32I-SW-NO-NEXT: mv s0, a1
; RV32I-SW-NO-NEXT: .LBB5_2: # %do_call
; RV32I-SW-NO-NEXT: # =>This Inner Loop Header: Depth=1
; RV32I-SW-NO-NEXT: call foo
; RV32I-SW-NO-NEXT: bne s0, s1, .LBB5_2
; RV32I-SW-NO-NEXT: .LBB5_3: # %exit
; RV32I-SW-NO-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
; RV32I-SW-NO-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
; RV32I-SW-NO-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
; RV32I-SW-NO-NEXT: .cfi_restore ra
; RV32I-SW-NO-NEXT: .cfi_restore s0
; RV32I-SW-NO-NEXT: .cfi_restore s1
; RV32I-SW-NO-NEXT: addi sp, sp, 16
; RV32I-SW-NO-NEXT: .cfi_def_cfa_offset 0
; RV32I-SW-NO-NEXT: ret
;
; RV32I-SW-LABEL: li_loop:
; RV32I-SW: # %bb.0:
; RV32I-SW-NEXT: addi sp, sp, -16
; RV32I-SW-NEXT: .cfi_def_cfa_offset 16
; RV32I-SW-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
; RV32I-SW-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
; RV32I-SW-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
; RV32I-SW-NEXT: .cfi_offset ra, -4
; RV32I-SW-NEXT: .cfi_offset s0, -8
; RV32I-SW-NEXT: .cfi_offset s1, -12
; RV32I-SW-NEXT: li s1, 57
; RV32I-SW-NEXT: beq a0, s1, .LBB5_3
; RV32I-SW-NEXT: # %bb.1: # %do_call.preheader
; RV32I-SW-NEXT: mv s0, a1
; RV32I-SW-NEXT: .LBB5_2: # %do_call
; RV32I-SW-NEXT: # =>This Inner Loop Header: Depth=1
; RV32I-SW-NEXT: call foo
; RV32I-SW-NEXT: bne s0, s1, .LBB5_2
; RV32I-SW-NEXT: .LBB5_3: # %exit
; RV32I-SW-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
; RV32I-SW-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
; RV32I-SW-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
; RV32I-SW-NEXT: .cfi_restore ra
; RV32I-SW-NEXT: .cfi_restore s0
; RV32I-SW-NEXT: .cfi_restore s1
; RV32I-SW-NEXT: addi sp, sp, 16
; RV32I-SW-NEXT: .cfi_def_cfa_offset 0
; RV32I-SW-NEXT: ret
;
; RV32I-SW-SR-LABEL: li_loop:
; RV32I-SW-SR: # %bb.0:
; RV32I-SW-SR-NEXT: call t0, __riscv_save_2
; RV32I-SW-SR-NEXT: .cfi_def_cfa_offset 16
; RV32I-SW-SR-NEXT: .cfi_offset ra, -4
; RV32I-SW-SR-NEXT: .cfi_offset s0, -8
; RV32I-SW-SR-NEXT: .cfi_offset s1, -12
; RV32I-SW-SR-NEXT: li s1, 57
; RV32I-SW-SR-NEXT: beq a0, s1, .LBB5_3
; RV32I-SW-SR-NEXT: # %bb.1: # %do_call.preheader
; RV32I-SW-SR-NEXT: mv s0, a1
; RV32I-SW-SR-NEXT: .LBB5_2: # %do_call
; RV32I-SW-SR-NEXT: # =>This Inner Loop Header: Depth=1
; RV32I-SW-SR-NEXT: call foo
; RV32I-SW-SR-NEXT: bne s0, s1, .LBB5_2
; RV32I-SW-SR-NEXT: .LBB5_3: # %exit
; RV32I-SW-SR-NEXT: tail __riscv_restore_2
;
; RV64I-SW-LABEL: li_loop:
; RV64I-SW: # %bb.0:
; RV64I-SW-NEXT: addi sp, sp, -32
; RV64I-SW-NEXT: .cfi_def_cfa_offset 32
; RV64I-SW-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
; RV64I-SW-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
; RV64I-SW-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
; RV64I-SW-NEXT: .cfi_offset ra, -8
; RV64I-SW-NEXT: .cfi_offset s0, -16
; RV64I-SW-NEXT: .cfi_offset s1, -24
; RV64I-SW-NEXT: li s1, 57
; RV64I-SW-NEXT: beq a0, s1, .LBB5_3
; RV64I-SW-NEXT: # %bb.1: # %do_call.preheader
; RV64I-SW-NEXT: mv s0, a1
; RV64I-SW-NEXT: .LBB5_2: # %do_call
; RV64I-SW-NEXT: # =>This Inner Loop Header: Depth=1
; RV64I-SW-NEXT: call foo
; RV64I-SW-NEXT: bne s0, s1, .LBB5_2
; RV64I-SW-NEXT: .LBB5_3: # %exit
; RV64I-SW-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
; RV64I-SW-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
; RV64I-SW-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
; RV64I-SW-NEXT: .cfi_restore ra
; RV64I-SW-NEXT: .cfi_restore s0
; RV64I-SW-NEXT: .cfi_restore s1
; RV64I-SW-NEXT: addi sp, sp, 32
; RV64I-SW-NEXT: .cfi_def_cfa_offset 0
; RV64I-SW-NEXT: ret
%cmp0 = icmp eq i32 %a, 57
br i1 %cmp0, label %exit, label %do_call
do_call:
call i32 @foo()
%cmp1 = icmp eq i32 %b, 57
br i1 %cmp1, label %exit, label %do_call
exit:
ret void
}
declare zeroext i32 @foo()
Reference in New Issue View Git Blame Copy Permalink