Illustrate for an issue related to custom CC and tail-call elimination

CMakeLists.txt

cmake_minimum_required(VERSION 3.10)
project(testllvm C)

set(CMAKE_C_STANDARD 99)

add_executable(testllvm test.c)

test.c

typedef struct _Context {
	int data1;
	int data2;
} Context;

typedef int CallFunc(Context*, int);
CallFunc handlerFunc0, handlerFunc1, terminatorFunc;

CallFunc *funcTable[] = {
		handlerFunc0,
		handlerFunc1,
		terminatorFunc
};

int starterFunc(Context *ctx)
{
	int startValue = ctx->data1 > ctx->data2 ? 1 : 0;
	CallFunc *func = funcTable[startValue];
	int res = func(ctx, startValue);
	if (res) {
		ctx->data2 = res;
	} else {
		ctx->data1 = res;
	}
	return res;
}

int terminatorFunc(Context* ctx, int arg0)
{
	return arg0;
}

int handlerFunc0(Context* ctx, int arg0)
{
	ctx->data1++;
	printf("Hello handlerFunc0!\n");
	return funcTable[++arg0](ctx, arg0);
}

int handlerFunc1(Context* ctx, int arg0)
{
	ctx->data2++;
	printf("Hello handlerFunc1!\n");
	return funcTable[++arg0](ctx, arg0);
}

int main()
{
	Context ctx = {5, 6};
	return starterFunc(&ctx);
}

test.ll

; ModuleID = 'test.c'
source_filename = "test.c"
target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
target triple = "armv7-unknown-linux-gnueabihf"
; target triple = "x86_64-unknown-linux"

%struct._Context = type { i32, i32 }

@funcTable = dso_local local_unnamed_addr global [3 x i32 (%struct._Context*, i32)*] [i32 (%struct._Context*, i32)* @handlerFunc0, i32 (%struct._Context*, i32)* @handlerFunc1, i32 (%struct._Context*, i32)* @terminatorFunc], align 4
@str = private unnamed_addr constant [20 x i8] c"Hello handlerFunc0!\00"
@str.2 = private unnamed_addr constant [20 x i8] c"Hello handlerFunc1!\00"

define internal cc10 i32 @handlerFunc0(%struct._Context*, i32) #0 {
  %3 = getelementptr inbounds %struct._Context, %struct._Context* %0, i32 0, i32 0
  %4 = load i32, i32* %3, align 4
  %5 = add nsw i32 %4, 1
  store i32 %5, i32* %3, align 4
  %6 = tail call i32 @puts(i8* getelementptr inbounds ([20 x i8], [20 x i8]* @str, i32 0, i32 0))
  %7 = add nsw i32 %1, 1
  %8 = getelementptr inbounds [3 x i32 (%struct._Context*, i32)*], [3 x i32 (%struct._Context*, i32)*]* @funcTable, i32 0, i32 %7
  %9 = load i32 (%struct._Context*, i32)*, i32 (%struct._Context*, i32)** %8, align 4
  %10 = tail call i32 %9(%struct._Context* %0, i32 %7) #3
  ret i32 %10
}

define internal cc10 i32 @handlerFunc1(%struct._Context*, i32) #0 {
  %3 = getelementptr inbounds %struct._Context, %struct._Context* %0, i32 0, i32 1
  %4 = load i32, i32* %3, align 4
  %5 = add nsw i32 %4, 1
  store i32 %5, i32* %3, align 4
  %6 = tail call i32 @puts(i8* getelementptr inbounds ([20 x i8], [20 x i8]* @str.2, i32 0, i32 0))
  %7 = add nsw i32 %1, 1
  %8 = getelementptr inbounds [3 x i32 (%struct._Context*, i32)*], [3 x i32 (%struct._Context*, i32)*]* @funcTable, i32 0, i32 %7
  %9 = load i32 (%struct._Context*, i32)*, i32 (%struct._Context*, i32)** %8, align 4
  %10 = tail call i32 %9(%struct._Context* %0, i32 %7) #3
  ret i32 %10
}

define internal cc10 i32 @terminatorFunc(%struct._Context* nocapture readnone, i32 returned) #0 {
  ret i32 %1
}

define internal i32 @starterFunc(%struct._Context*) local_unnamed_addr #0 {
  %2 = getelementptr inbounds %struct._Context, %struct._Context* %0, i32 0, i32 0
  %3 = load i32, i32* %2, align 4
  %4 = getelementptr inbounds %struct._Context, %struct._Context* %0, i32 0, i32 1
  %5 = load i32, i32* %4, align 4
  %6 = icmp sgt i32 %3, %5
  %7 = zext i1 %6 to i32
  %8 = getelementptr inbounds [3 x i32 (%struct._Context*, i32)*], [3 x i32 (%struct._Context*, i32)*]* @funcTable, i32 0, i32 %7
  %9 = load i32 (%struct._Context*, i32)*, i32 (%struct._Context*, i32)** %8, align 4
  %10 = call i32 %9(%struct._Context* %0, i32 %7) #3
  %11 = icmp eq i32 %10, 0
  br i1 %11, label %13, label %12

; <label>:12:                                     ; preds = %1
  store i32 %10, i32* %4, align 4
  br label %14

; <label>:13:                                     ; preds = %1
  store i32 0, i32* %2, align 4
  br label %14

; <label>:14:                                     ; preds = %13, %12
  ret i32 %10
}

declare i32 @puts(i8* nocapture readonly) local_unnamed_addr #3

attributes #0 = { nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="generic" "target-features"="+armv7-a,+dsp,+neon,+vfp3,-thumb-mode" "unsafe-fp-math"="false" "use-soft-float"="false" }

test.s

	.text
	.syntax unified
	.eabi_attribute	67, "2.09"	@ Tag_conformance
	.eabi_attribute	6, 10	@ Tag_CPU_arch
	.eabi_attribute	7, 65	@ Tag_CPU_arch_profile
	.eabi_attribute	8, 1	@ Tag_ARM_ISA_use
	.eabi_attribute	9, 2	@ Tag_THUMB_ISA_use
	.fpu	neon
	.eabi_attribute	34, 1	@ Tag_CPU_unaligned_access
	.eabi_attribute	17, 1	@ Tag_ABI_PCS_GOT_use
	.eabi_attribute	20, 1	@ Tag_ABI_FP_denormal
	.eabi_attribute	21, 1	@ Tag_ABI_FP_exceptions
	.eabi_attribute	23, 3	@ Tag_ABI_FP_number_model
	.eabi_attribute	24, 1	@ Tag_ABI_align_needed
	.eabi_attribute	25, 1	@ Tag_ABI_align_preserved
	.eabi_attribute	28, 1	@ Tag_ABI_VFP_args
	.eabi_attribute	38, 1	@ Tag_ABI_FP_16bit_format
	.eabi_attribute	14, 0	@ Tag_ABI_PCS_R9_use
	.file	"test.c"
	.p2align	2               @ -- Begin function handlerFunc0
	.type	handlerFunc0,%function
	.code	32                      @ @handlerFunc0
handlerFunc0:
	.fnstart
@ %bb.0:
	ldr	r7, [r4]
	movw	r0, :lower16:.Lstr
	movt	r0, :upper16:.Lstr
	add	r7, r7, #1
	str	r7, [r4]
	bl	puts
	movw	r7, :lower16:funcTable
	add	r1, r5, #1
	movt	r7, :upper16:funcTable
	mov	r0, r4
	ldr	r2, [r7, r1, lsl #2]
	bx	r2
.Lfunc_end0:
	.size	handlerFunc0, .Lfunc_end0-handlerFunc0
	.cantunwind
	.fnend
                                        @ -- End function
	.p2align	2               @ -- Begin function handlerFunc1
	.type	handlerFunc1,%function
	.code	32                      @ @handlerFunc1
handlerFunc1:
	.fnstart
@ %bb.0:
	ldr	r7, [r4, #4]
	movw	r0, :lower16:.Lstr.2
	movt	r0, :upper16:.Lstr.2
	add	r7, r7, #1
	str	r7, [r4, #4]
	bl	puts
	movw	r7, :lower16:funcTable
	add	r1, r5, #1
	movt	r7, :upper16:funcTable
	mov	r0, r4
	ldr	r2, [r7, r1, lsl #2]
	bx	r2
.Lfunc_end1:
	.size	handlerFunc1, .Lfunc_end1-handlerFunc1
	.cantunwind
	.fnend
                                        @ -- End function
	.p2align	2               @ -- Begin function terminatorFunc
	.type	terminatorFunc,%function
	.code	32                      @ @terminatorFunc
terminatorFunc:
	.fnstart
@ %bb.0:
	mov	r0, r5
	bx	lr
.Lfunc_end2:
	.size	terminatorFunc, .Lfunc_end2-terminatorFunc
	.cantunwind
	.fnend
                                        @ -- End function
	.p2align	2               @ -- Begin function starterFunc
	.type	starterFunc,%function
	.code	32                      @ @starterFunc
starterFunc:
	.fnstart
@ %bb.0:
	.save	{r4, r5, r11, lr}
	push	{r4, r5, r11, lr}
	mov	r4, r0
	ldr	r0, [r0]
	ldr	r1, [r4, #4]
	mov	r5, #0
	cmp	r0, r1
	mov	r1, #0
	movw	r0, :lower16:funcTable
	movwgt	r1, #1
	movt	r0, :upper16:funcTable
	ldr	r2, [r0, r1, lsl #2]
	mov	r0, r4
	blx	r2
	cmp	r0, #0
	streq	r5, [r4]
	strne	r0, [r4, #4]
	pop	{r4, r5, r11, pc}
.Lfunc_end3:
	.size	starterFunc, .Lfunc_end3-starterFunc
	.cantunwind
	.fnend
                                        @ -- End function
	.type	funcTable,%object       @ @funcTable
	.data
	.globl	funcTable
	.p2align	2
funcTable:
	.long	handlerFunc0
	.long	handlerFunc1
	.long	terminatorFunc
	.size	funcTable, 12

	.type	.Lstr,%object           @ @str
	.section	.rodata.str1.16,"aMS",%progbits,1
	.p2align	4
.Lstr:
	.asciz	"Hello handlerFunc0!"
	.size	.Lstr, 20

	.type	.Lstr.2,%object         @ @str.2
	.p2align	4
.Lstr.2:
	.asciz	"Hello handlerFunc1!"
	.size	.Lstr.2, 20


	.section	".note.GNU-stack","",%progbits
	.eabi_attribute	30, 1	@ Tag_ABI_optimization_goals