1 ; RUN: llc %s -o - -enable-shrink-wrap=true | FileCheck %s --check-prefix=CHECK --check-prefix=ENABLE
2 ; RUN: llc %s -o - -enable-shrink-wrap=false | FileCheck %s --check-prefix=CHECK --check-prefix=DISABLE
4 ; Note: Lots of tests use inline asm instead of regular calls.
5 ; This allows to have a better control on what the allocation will do.
6 ; Otherwise, we may have spill right in the entry block, defeating
7 ; shrink-wrapping. Moreover, some of the inline asm statement (nop)
8 ; are here to ensure that the related paths do not end up as critical
10 target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
11 target triple = "x86_64-apple-macosx"
14 ; Initial motivating example: Simple diamond with a call just on one side.
17 ; Compare the arguments and jump to exit.
19 ; ENABLE: movl %edi, [[ARG0CPY:%e[a-z]+]]
20 ; ENABLE-NEXT: cmpl %esi, [[ARG0CPY]]
21 ; ENABLE-NEXT: jge [[EXIT_LABEL:LBB[0-9_]+]]
24 ; (What we push does not matter. It should be some random sratch register.)
27 ; Compare the arguments and jump to exit.
28 ; After the prologue is set.
29 ; DISABLE: movl %edi, [[ARG0CPY:%e[a-z]+]]
30 ; DISABLE-NEXT: cmpl %esi, [[ARG0CPY]]
31 ; DISABLE-NEXT: jge [[EXIT_LABEL:LBB[0-9_]+]]
33 ; Store %a in the alloca.
34 ; CHECK: movl [[ARG0CPY]], 4(%rsp)
35 ; Set the alloca address in the second argument.
36 ; CHECK-NEXT: leaq 4(%rsp), %rsi
37 ; Set the first argument to zero.
38 ; CHECK-NEXT: xorl %edi, %edi
39 ; CHECK-NEXT: callq _doSomething
41 ; With shrink-wrapping, epilogue is just after the call.
42 ; ENABLE-NEXT: addq $8, %rsp
44 ; CHECK: [[EXIT_LABEL]]:
46 ; Without shrink-wrapping, epilogue is in the exit block.
47 ; Epilogue code. (What we pop does not matter.)
51 define i32 @foo(i32 %a, i32 %b) {
52 %tmp = alloca i32, align 4
53 %tmp2 = icmp slt i32 %a, %b
54 br i1 %tmp2, label %true, label %false
57 store i32 %a, i32* %tmp, align 4
58 %tmp4 = call i32 @doSomething(i32 0, i32* %tmp)
62 %tmp.0 = phi i32 [ %tmp4, %true ], [ %a, %0 ]
66 ; Function Attrs: optsize
67 declare i32 @doSomething(i32, i32*)
70 ; Check that we do not perform the restore inside the loop whereas the save
72 ; CHECK-LABEL: freqSaveAndRestoreOutsideLoop:
74 ; Shrink-wrapping allows to skip the prologue in the else case.
75 ; ENABLE: testl %edi, %edi
76 ; ENABLE: je [[ELSE_LABEL:LBB[0-9_]+]]
79 ; Make sure we save the CSR used in the inline asm: rbx.
82 ; DISABLE: testl %edi, %edi
83 ; DISABLE: je [[ELSE_LABEL:LBB[0-9_]+]]
85 ; SUM is in %esi because it is coalesced with the second
86 ; argument on the else path.
87 ; CHECK: xorl [[SUM:%esi]], [[SUM]]
88 ; CHECK-NEXT: movl $10, [[IV:%e[a-z]+]]
91 ; CHECK: [[LOOP:LBB[0-9_]+]]: ## %for.body
92 ; CHECK: movl $1, [[TMP:%e[a-z]+]]
93 ; CHECK: addl [[TMP]], [[SUM]]
94 ; CHECK-NEXT: decl [[IV]]
95 ; CHECK-NEXT: jne [[LOOP]]
99 ; CHECK: shll $3, [[SUM]]
102 ; DISABLE: jmp [[EPILOG_BB:LBB[0-9_]+]]
104 ; DISABLE: [[ELSE_LABEL]]: ## %if.else
105 ; Shift second argument by one and store into returned register.
106 ; DISABLE: addl %esi, %esi
107 ; DISABLE: [[EPILOG_BB]]: ## %if.end
110 ; CHECK-DAG: popq %rbx
111 ; CHECK-DAG: movl %esi, %eax
114 ; ENABLE: [[ELSE_LABEL]]: ## %if.else
115 ; Shift second argument by one and store into returned register.
116 ; ENABLE: addl %esi, %esi
117 ; ENABLE-NEXT: movl %esi, %eax
119 define i32 @freqSaveAndRestoreOutsideLoop(i32 %cond, i32 %N) {
121 %tobool = icmp eq i32 %cond, 0
122 br i1 %tobool, label %if.else, label %for.preheader
125 tail call void asm "nop", ""()
128 for.body: ; preds = %entry, %for.body
129 %i.05 = phi i32 [ %inc, %for.body ], [ 0, %for.preheader ]
130 %sum.04 = phi i32 [ %add, %for.body ], [ 0, %for.preheader ]
131 %call = tail call i32 asm "movl $$1, $0", "=r,~{ebx}"()
132 %add = add nsw i32 %call, %sum.04
133 %inc = add nuw nsw i32 %i.05, 1
134 %exitcond = icmp eq i32 %inc, 10
135 br i1 %exitcond, label %for.end, label %for.body
137 for.end: ; preds = %for.body
138 %shl = shl i32 %add, 3
141 if.else: ; preds = %entry
142 %mul = shl nsw i32 %N, 1
145 if.end: ; preds = %if.else, %for.end
146 %sum.1 = phi i32 [ %shl, %for.end ], [ %mul, %if.else ]
150 declare i32 @something(...)
152 ; Check that we do not perform the shrink-wrapping inside the loop even
153 ; though that would be legal. The cost model must prevent that.
154 ; CHECK-LABEL: freqSaveAndRestoreOutsideLoop2:
156 ; Make sure we save the CSR used in the inline asm: rbx.
159 ; CHECK: xorl [[SUM:%e[a-z]+]], [[SUM]]
160 ; CHECK-NEXT: movl $10, [[IV:%e[a-z]+]]
162 ; CHECK: [[LOOP_LABEL:LBB[0-9_]+]]: ## %for.body
163 ; CHECK: movl $1, [[TMP:%e[a-z]+]]
164 ; CHECK: addl [[TMP]], [[SUM]]
165 ; CHECK-NEXT: decl [[IV]]
166 ; CHECK-NEXT: jne [[LOOP_LABEL]]
168 ; CHECK: ## %for.exit
172 define i32 @freqSaveAndRestoreOutsideLoop2(i32 %cond) {
174 br label %for.preheader
177 tail call void asm "nop", ""()
180 for.body: ; preds = %for.body, %entry
181 %i.04 = phi i32 [ 0, %for.preheader ], [ %inc, %for.body ]
182 %sum.03 = phi i32 [ 0, %for.preheader ], [ %add, %for.body ]
183 %call = tail call i32 asm "movl $$1, $0", "=r,~{ebx}"()
184 %add = add nsw i32 %call, %sum.03
185 %inc = add nuw nsw i32 %i.04, 1
186 %exitcond = icmp eq i32 %inc, 10
187 br i1 %exitcond, label %for.exit, label %for.body
190 tail call void asm "nop", ""()
193 for.end: ; preds = %for.body
197 ; Check with a more complex case that we do not have save within the loop and
199 ; CHECK-LABEL: loopInfoSaveOutsideLoop:
201 ; ENABLE: testl %edi, %edi
202 ; ENABLE-NEXT: je [[ELSE_LABEL:LBB[0-9_]+]]
205 ; Make sure we save the CSR used in the inline asm: rbx.
208 ; DISABLE: testl %edi, %edi
209 ; DISABLE-NEXT: je [[ELSE_LABEL:LBB[0-9_]+]]
212 ; CHECK: xorl [[SUM:%esi]], [[SUM]]
213 ; CHECK-NEXT: movl $10, [[IV:%e[a-z]+]]
215 ; CHECK: [[LOOP_LABEL:LBB[0-9_]+]]: ## %for.body
216 ; CHECK: movl $1, [[TMP:%e[a-z]+]]
217 ; CHECK: addl [[TMP]], [[SUM]]
218 ; CHECK-NEXT: decl [[IV]]
219 ; CHECK-NEXT: jne [[LOOP_LABEL]]
222 ; CHECK: shll $3, [[SUM]]
224 ; DISABLE: jmp [[EPILOG_BB:LBB[0-9_]+]]
226 ; DISABLE: [[ELSE_LABEL]]: ## %if.else
227 ; Shift second argument by one and store into returned register.
228 ; DISABLE: addl %esi, %esi
229 ; DISABLE: [[EPILOG_BB]]: ## %if.end
232 ; CHECK-DAG: popq %rbx
233 ; CHECK-DAG: movl %esi, %eax
236 ; ENABLE: [[ELSE_LABEL]]: ## %if.else
237 ; Shift second argument by one and store into returned register.
238 ; ENABLE: addl %esi, %esi
239 ; ENABLE-NEXT: movl %esi, %eax
241 define i32 @loopInfoSaveOutsideLoop(i32 %cond, i32 %N) {
243 %tobool = icmp eq i32 %cond, 0
244 br i1 %tobool, label %if.else, label %for.preheader
247 tail call void asm "nop", ""()
250 for.body: ; preds = %entry, %for.body
251 %i.05 = phi i32 [ %inc, %for.body ], [ 0, %for.preheader ]
252 %sum.04 = phi i32 [ %add, %for.body ], [ 0, %for.preheader ]
253 %call = tail call i32 asm "movl $$1, $0", "=r,~{ebx}"()
254 %add = add nsw i32 %call, %sum.04
255 %inc = add nuw nsw i32 %i.05, 1
256 %exitcond = icmp eq i32 %inc, 10
257 br i1 %exitcond, label %for.end, label %for.body
259 for.end: ; preds = %for.body
260 tail call void asm "nop", "~{ebx}"()
261 %shl = shl i32 %add, 3
264 if.else: ; preds = %entry
265 %mul = shl nsw i32 %N, 1
268 if.end: ; preds = %if.else, %for.end
269 %sum.1 = phi i32 [ %shl, %for.end ], [ %mul, %if.else ]
273 ; Check with a more complex case that we do not have restore within the loop and
275 ; CHECK-LABEL: loopInfoRestoreOutsideLoop:
277 ; ENABLE: testl %edi, %edi
278 ; ENABLE-NEXT: je [[ELSE_LABEL:LBB[0-9_]+]]
281 ; Make sure we save the CSR used in the inline asm: rbx.
284 ; DISABLE: testl %edi, %edi
285 ; DISABLE-NEXT: je [[ELSE_LABEL:LBB[0-9_]+]]
288 ; CHECK: xorl [[SUM:%esi]], [[SUM]]
289 ; CHECK-NEXT: movl $10, [[IV:%e[a-z]+]]
291 ; CHECK: [[LOOP_LABEL:LBB[0-9_]+]]: ## %for.body
292 ; CHECK: movl $1, [[TMP:%e[a-z]+]]
293 ; CHECK: addl [[TMP]], [[SUM]]
294 ; CHECK-NEXT: decl [[IV]]
295 ; CHECK-NEXT: jne [[LOOP_LABEL]]
297 ; CHECK: shll $3, [[SUM]]
299 ; DISABLE: jmp [[EPILOG_BB:LBB[0-9_]+]]
301 ; DISABLE: [[ELSE_LABEL]]: ## %if.else
303 ; Shift second argument by one and store into returned register.
304 ; DISABLE: addl %esi, %esi
305 ; DISABLE: [[EPILOG_BB]]: ## %if.end
308 ; CHECK-DAG: popq %rbx
309 ; CHECK-DAG: movl %esi, %eax
312 ; ENABLE: [[ELSE_LABEL]]: ## %if.else
313 ; Shift second argument by one and store into returned register.
314 ; ENABLE: addl %esi, %esi
315 ; ENABLE-NEXT: movl %esi, %eax
317 define i32 @loopInfoRestoreOutsideLoop(i32 %cond, i32 %N) #0 {
319 %tobool = icmp eq i32 %cond, 0
320 br i1 %tobool, label %if.else, label %if.then
322 if.then: ; preds = %entry
323 tail call void asm "nop", "~{ebx}"()
326 for.body: ; preds = %for.body, %if.then
327 %i.05 = phi i32 [ 0, %if.then ], [ %inc, %for.body ]
328 %sum.04 = phi i32 [ 0, %if.then ], [ %add, %for.body ]
329 %call = tail call i32 asm "movl $$1, $0", "=r,~{ebx}"()
330 %add = add nsw i32 %call, %sum.04
331 %inc = add nuw nsw i32 %i.05, 1
332 %exitcond = icmp eq i32 %inc, 10
333 br i1 %exitcond, label %for.end, label %for.body
335 for.end: ; preds = %for.body
336 %shl = shl i32 %add, 3
339 if.else: ; preds = %entry
340 %mul = shl nsw i32 %N, 1
343 if.end: ; preds = %if.else, %for.end
344 %sum.1 = phi i32 [ %shl, %for.end ], [ %mul, %if.else ]
348 ; Check that we handle function with no frame information correctly.
349 ; CHECK-LABEL: emptyFrame:
351 ; CHECK-NEXT: xorl %eax, %eax
353 define i32 @emptyFrame() {
358 ; Check that we handle inline asm correctly.
359 ; CHECK-LABEL: inlineAsm:
361 ; ENABLE: testl %edi, %edi
362 ; ENABLE-NEXT: je [[ELSE_LABEL:LBB[0-9_]+]]
365 ; Make sure we save the CSR used in the inline asm: rbx.
368 ; DISABLE: testl %edi, %edi
369 ; DISABLE-NEXT: je [[ELSE_LABEL:LBB[0-9_]+]]
372 ; CHECK: movl $10, [[IV:%e[a-z]+]]
374 ; CHECK: [[LOOP_LABEL:LBB[0-9_]+]]: ## %for.body
375 ; Inline asm statement.
376 ; CHECK: addl $1, %ebx
378 ; CHECK-NEXT: jne [[LOOP_LABEL]]
381 ; CHECK: xorl %esi, %esi
383 ; DISABLE: jmp [[EPILOG_BB:LBB[0-9_]+]]
385 ; DISABLE: [[ELSE_LABEL]]: ## %if.else
386 ; Shift second argument by one and store into returned register.
387 ; DISABLE: addl %esi, %esi
388 ; DISABLE: [[EPILOG_BB]]: ## %if.end
391 ; CHECK-DAG: popq %rbx
392 ; CHECK-DAG: movl %esi, %eax
395 ; ENABLE: [[ELSE_LABEL]]: ## %if.else
396 ; Shift second argument by one and store into returned register.
397 ; ENABLE: addl %esi, %esi
398 ; ENABLE-NEXT: movl %esi, %eax
400 define i32 @inlineAsm(i32 %cond, i32 %N) {
402 %tobool = icmp eq i32 %cond, 0
403 br i1 %tobool, label %if.else, label %for.preheader
406 tail call void asm "nop", ""()
409 for.body: ; preds = %entry, %for.body
410 %i.03 = phi i32 [ %inc, %for.body ], [ 0, %for.preheader ]
411 tail call void asm "addl $$1, %ebx", "~{ebx}"()
412 %inc = add nuw nsw i32 %i.03, 1
413 %exitcond = icmp eq i32 %inc, 10
414 br i1 %exitcond, label %for.exit, label %for.body
417 tail call void asm "nop", ""()
420 if.else: ; preds = %entry
421 %mul = shl nsw i32 %N, 1
424 if.end: ; preds = %for.body, %if.else
425 %sum.0 = phi i32 [ %mul, %if.else ], [ 0, %for.exit ]
429 ; Check that we handle calls to variadic functions correctly.
430 ; CHECK-LABEL: callVariadicFunc:
432 ; ENABLE: testl %edi, %edi
433 ; ENABLE-NEXT: je [[ELSE_LABEL:LBB[0-9_]+]]
438 ; DISABLE: testl %edi, %edi
439 ; DISABLE-NEXT: je [[ELSE_LABEL:LBB[0-9_]+]]
441 ; Setup of the varags.
442 ; CHECK: movl %esi, (%rsp)
443 ; CHECK-NEXT: xorl %eax, %eax
444 ; CHECK-NEXT: %esi, %edi
445 ; CHECK-NEXT: %esi, %edx
446 ; CHECK-NEXT: %esi, %ecx
447 ; CHECK-NEXT: %esi, %r8d
448 ; CHECK-NEXT: %esi, %r9d
449 ; CHECK-NEXT: callq _someVariadicFunc
450 ; CHECK-NEXT: movl %eax, %esi
451 ; CHECK-NEXT: shll $3, %esi
453 ; ENABLE-NEXT: addq $8, %rsp
454 ; ENABLE-NEXT: movl %esi, %eax
457 ; DISABLE: jmp [[IFEND_LABEL:LBB[0-9_]+]]
459 ; CHECK: [[ELSE_LABEL]]: ## %if.else
460 ; Shift second argument by one and store into returned register.
461 ; CHECK: addl %esi, %esi
463 ; DISABLE: [[IFEND_LABEL]]: ## %if.end
466 ; CHECK-NEXT: movl %esi, %eax
469 define i32 @callVariadicFunc(i32 %cond, i32 %N) {
471 %tobool = icmp eq i32 %cond, 0
472 br i1 %tobool, label %if.else, label %if.then
474 if.then: ; preds = %entry
475 %call = tail call i32 (i32, ...) @someVariadicFunc(i32 %N, i32 %N, i32 %N, i32 %N, i32 %N, i32 %N, i32 %N)
476 %shl = shl i32 %call, 3
479 if.else: ; preds = %entry
480 %mul = shl nsw i32 %N, 1
483 if.end: ; preds = %if.else, %if.then
484 %sum.0 = phi i32 [ %shl, %if.then ], [ %mul, %if.else ]
488 declare i32 @someVariadicFunc(i32, ...)
490 ; Check that we use LEA not to clobber EFLAGS.
491 %struct.temp_slot = type { %struct.temp_slot*, %struct.rtx_def*, %struct.rtx_def*, i32, i64, %union.tree_node*, %union.tree_node*, i8, i8, i32, i32, i64, i64 }
492 %union.tree_node = type { %struct.tree_decl }
493 %struct.tree_decl = type { %struct.tree_common, i8*, i32, i32, %union.tree_node*, i48, %union.anon, %union.tree_node*, %union.tree_node*, %union.tree_node*, %union.tree_node*, %union.tree_node*, %union.tree_node*, %union.tree_node*, %union.tree_node*, %union.tree_node*, %union.tree_node*, %struct.rtx_def*, %struct.rtx_def*, %union.anon.1, %union.tree_node*, %union.tree_node*, %union.tree_node*, i64, %struct.lang_decl* }
494 %struct.tree_common = type { %union.tree_node*, %union.tree_node*, i32 }
495 %union.anon = type { i64 }
496 %union.anon.1 = type { %struct.function* }
497 %struct.function = type { %struct.eh_status*, %struct.stmt_status*, %struct.expr_status*, %struct.emit_status*, %struct.varasm_status*, i8*, %union.tree_node*, %struct.function*, i32, i32, i32, i32, %struct.rtx_def*, %struct.ix86_args, %struct.rtx_def*, %struct.rtx_def*, i8*, %struct.initial_value_struct*, i32, %union.tree_node*, %struct.rtx_def*, %struct.rtx_def*, %struct.rtx_def*, %struct.rtx_def*, %struct.rtx_def*, %struct.rtx_def*, %struct.rtx_def*, %union.tree_node*, %struct.rtx_def*, %struct.rtx_def*, %struct.rtx_def*, %struct.rtx_def*, i64, %union.tree_node*, %union.tree_node*, %struct.rtx_def*, %struct.rtx_def*, i32, %struct.rtx_def**, %struct.temp_slot*, i32, i32, i32, %struct.var_refs_queue*, i32, i32, i8*, %union.tree_node*, %struct.rtx_def*, i32, i32, %struct.machine_function*, i32, i32, %struct.language_function*, %struct.rtx_def*, i24 }
498 %struct.eh_status = type opaque
499 %struct.stmt_status = type opaque
500 %struct.expr_status = type { i32, i32, i32, %struct.rtx_def*, %struct.rtx_def*, %struct.rtx_def*, %struct.rtx_def* }
501 %struct.emit_status = type { i32, i32, %struct.rtx_def*, %struct.rtx_def*, %union.tree_node*, %struct.sequence_stack*, i32, i32, i8*, i32, i8*, %union.tree_node**, %struct.rtx_def** }
502 %struct.sequence_stack = type { %struct.rtx_def*, %struct.rtx_def*, %union.tree_node*, %struct.sequence_stack* }
503 %struct.varasm_status = type opaque
504 %struct.ix86_args = type { i32, i32, i32, i32, i32, i32, i32 }
505 %struct.initial_value_struct = type opaque
506 %struct.var_refs_queue = type { %struct.rtx_def*, i32, i32, %struct.var_refs_queue* }
507 %struct.machine_function = type opaque
508 %struct.language_function = type opaque
509 %struct.lang_decl = type opaque
510 %struct.rtx_def = type { i32, [1 x %union.rtunion_def] }
511 %union.rtunion_def = type { i64 }
513 declare hidden fastcc %struct.temp_slot* @find_temp_slot_from_address(%struct.rtx_def* readonly)
515 ; CHECK-LABEL: useLEA:
518 ; CHECK: testq %rdi, %rdi
519 ; CHECK-NEXT: je [[CLEANUP:LBB[0-9_]+]]
521 ; CHECK: movzwl (%rdi), [[BF_LOAD:%e[a-z]+]]
522 ; CHECK-NEXT: cmpl $66, [[BF_LOAD]]
523 ; CHECK-NEXT: jne [[CLEANUP]]
525 ; CHECK: movq 8(%rdi), %rdi
526 ; CHECK-NEXT: movzwl (%rdi), %e[[BF_LOAD2:[a-z]+]]
527 ; CHECK-NEXT: leal -54(%r[[BF_LOAD2]]), [[TMP:%e[a-z]+]]
528 ; CHECK-NEXT: cmpl $14, [[TMP]]
529 ; CHECK-NEXT: ja [[LOR_LHS_FALSE:LBB[0-9_]+]]
531 ; CHECK: movl $24599, [[TMP2:%e[a-z]+]]
532 ; CHECK-NEXT: btl [[TMP]], [[TMP2]]
533 ; CHECK-NEXT: jae [[LOR_LHS_FALSE:LBB[0-9_]+]]
535 ; CHECK: [[CLEANUP]]: ## %cleanup
539 ; CHECK: [[LOR_LHS_FALSE]]: ## %lor.lhs.false
540 ; CHECK: cmpl $134, %e[[BF_LOAD2]]
541 ; CHECK-NEXT: je [[CLEANUP]]
543 ; CHECK: cmpl $140, %e[[BF_LOAD2]]
544 ; CHECK-NEXT: je [[CLEANUP]]
547 ; CHECK: callq _find_temp_slot_from_address
548 ; CHECK-NEXT: testq %rax, %rax
550 ; The adjustment must use LEA here (or be moved above the test).
551 ; ENABLE-NEXT: leaq 8(%rsp), %rsp
553 ; CHECK-NEXT: je [[CLEANUP]]
555 ; CHECK: movb $1, 57(%rax)
556 define void @useLEA(%struct.rtx_def* readonly %x) {
558 %cmp = icmp eq %struct.rtx_def* %x, null
559 br i1 %cmp, label %cleanup, label %if.end
561 if.end: ; preds = %entry
562 %tmp = getelementptr inbounds %struct.rtx_def, %struct.rtx_def* %x, i64 0, i32 0
563 %bf.load = load i32, i32* %tmp, align 8
564 %bf.clear = and i32 %bf.load, 65535
565 %cmp1 = icmp eq i32 %bf.clear, 66
566 br i1 %cmp1, label %lor.lhs.false, label %cleanup
568 lor.lhs.false: ; preds = %if.end
569 %arrayidx = getelementptr inbounds %struct.rtx_def, %struct.rtx_def* %x, i64 0, i32 1, i64 0
570 %rtx = bitcast %union.rtunion_def* %arrayidx to %struct.rtx_def**
571 %tmp1 = load %struct.rtx_def*, %struct.rtx_def** %rtx, align 8
572 %tmp2 = getelementptr inbounds %struct.rtx_def, %struct.rtx_def* %tmp1, i64 0, i32 0
573 %bf.load2 = load i32, i32* %tmp2, align 8
574 %bf.clear3 = and i32 %bf.load2, 65535
575 switch i32 %bf.clear3, label %if.end.55 [
576 i32 67, label %cleanup
577 i32 68, label %cleanup
578 i32 54, label %cleanup
579 i32 55, label %cleanup
580 i32 58, label %cleanup
581 i32 134, label %cleanup
582 i32 56, label %cleanup
583 i32 140, label %cleanup
586 if.end.55: ; preds = %lor.lhs.false
587 %call = tail call fastcc %struct.temp_slot* @find_temp_slot_from_address(%struct.rtx_def* %tmp1) #2
588 %cmp59 = icmp eq %struct.temp_slot* %call, null
589 br i1 %cmp59, label %cleanup, label %if.then.60
591 if.then.60: ; preds = %if.end.55
592 %addr_taken = getelementptr inbounds %struct.temp_slot, %struct.temp_slot* %call, i64 0, i32 8
593 store i8 1, i8* %addr_taken, align 1
596 cleanup: ; preds = %if.then.60, %if.end.55, %lor.lhs.false, %lor.lhs.false, %lor.lhs.false, %lor.lhs.false, %lor.lhs.false, %lor.lhs.false, %lor.lhs.false, %lor.lhs.false, %if.end, %entry
600 ; Make sure we do not insert unreachable code after noreturn function.
601 ; Although this is not incorrect to insert such code, it is useless
602 ; and it hurts the binary size.
604 ; CHECK-LABEL: noreturn:
607 ; CHECK: testb %dil, %dil
608 ; CHECK-NEXT: jne [[ABORT:LBB[0-9_]+]]
610 ; CHECK: movl $42, %eax
616 ; CHECK: [[ABORT]]: ## %if.abort
620 ; CHECK: callq _abort
622 define i32 @noreturn(i8 signext %bad_thing) {
624 %tobool = icmp eq i8 %bad_thing, 0
625 br i1 %tobool, label %if.end, label %if.abort
628 tail call void @abort() #0
635 declare void @abort() #0
637 attributes #0 = { noreturn nounwind }
640 ; Make sure that we handle infinite loops properly When checking that the Save
641 ; and Restore blocks are control flow equivalent, the loop searches for the
642 ; immediate (post) dominator for the (restore) save blocks. When either the Save
643 ; or Restore block is located in an infinite loop the only immediate (post)
644 ; dominator is itself. In this case, we cannot perform shrink wrapping, but we
645 ; should return gracefully and continue compilation.
646 ; The only condition for this test is the compilation finishes correctly.
648 ; CHECK-LABEL: infiniteloop
650 define void @infiniteloop() {
652 br i1 undef, label %if.then, label %if.end
655 %ptr = alloca i32, i32 4
658 for.body: ; preds = %for.body, %entry
659 %sum.03 = phi i32 [ 0, %if.then ], [ %add, %for.body ]
660 %call = tail call i32 asm "movl $$1, $0", "=r,~{ebx}"()
661 %add = add nsw i32 %call, %sum.03
662 store i32 %add, i32* %ptr
669 ; Another infinite loop test this time with a body bigger than just one block.
670 ; CHECK-LABEL: infiniteloop2
672 define void @infiniteloop2() {
674 br i1 undef, label %if.then, label %if.end
677 %ptr = alloca i32, i32 4
680 for.body: ; preds = %for.body, %entry
681 %sum.03 = phi i32 [ 0, %if.then ], [ %add, %body1 ], [ 1, %body2]
682 %call = tail call i32 asm "movl $$1, $0", "=r,~{ebx}"()
683 %add = add nsw i32 %call, %sum.03
684 store i32 %add, i32* %ptr
685 br i1 undef, label %body1, label %body2
688 tail call void asm sideeffect "nop", "~{ebx}"()
692 tail call void asm sideeffect "nop", "~{ebx}"()
699 ; Another infinite loop test this time with two nested infinite loop.
700 ; CHECK-LABEL: infiniteloop3
702 define void @infiniteloop3() {
704 br i1 undef, label %loop2a, label %body
706 body: ; preds = %entry
707 br i1 undef, label %loop2a, label %end
709 loop1: ; preds = %loop2a, %loop2b
710 %var.phi = phi i32* [ %next.phi, %loop2b ], [ %var, %loop2a ]
711 %next.phi = phi i32* [ %next.load, %loop2b ], [ %next.var, %loop2a ]
712 %0 = icmp eq i32* %var, null
713 %next.load = load i32*, i32** undef
714 br i1 %0, label %loop2a, label %loop2b
716 loop2a: ; preds = %loop1, %body, %entry
717 %var = phi i32* [ null, %body ], [ null, %entry ], [ %next.phi, %loop1 ]
718 %next.var = phi i32* [ undef, %body ], [ null, %entry ], [ %next.load, %loop1 ]
721 loop2b: ; preds = %loop1
722 %gep1 = bitcast i32* %var.phi to i32*
723 %next.ptr = bitcast i32* %gep1 to i32**
724 store i32* %next.phi, i32** %next.ptr
731 ; Check that we just don't bail out on RegMask.
732 ; In this case, the RegMask does not touch a CSR so we are good to go!
733 ; CHECK-LABEL: regmask:
735 ; Compare the arguments and jump to exit.
736 ; No prologue needed.
737 ; ENABLE: cmpl %esi, %edi
738 ; ENABLE-NEXT: jge [[EXIT_LABEL:LBB[0-9_]+]]
741 ; (What we push does not matter. It should be some random sratch register.)
744 ; Compare the arguments and jump to exit.
745 ; After the prologue is set.
746 ; DISABLE: cmpl %esi, %edi
747 ; DISABLE-NEXT: jge [[EXIT_LABEL:LBB[0-9_]+]]
750 ; Set the first argument to zero.
751 ; CHECK: xorl %edi, %edi
752 ; Set the second argument to addr.
753 ; CHECK-NEXT: movq %rdx, %rsi
754 ; CHECK-NEXT: callq _doSomething
758 ; CHECK: [[EXIT_LABEL]]:
759 ; Set the first argument to 6.
760 ; CHECK-NEXT: movl $6, %edi
761 ; Set the second argument to addr.
762 ; CHECK-NEXT: movq %rdx, %rsi
764 ; Without shrink-wrapping, we need to restore the stack before
765 ; making the tail call.
769 ; CHECK-NEXT: jmp _doSomething
770 define i32 @regmask(i32 %a, i32 %b, i32* %addr) {
771 %tmp2 = icmp slt i32 %a, %b
772 br i1 %tmp2, label %true, label %false
775 ; Clobber a CSR so that we check something on the regmask
777 tail call void asm sideeffect "nop", "~{ebx}"()
778 %tmp4 = call i32 @doSomething(i32 0, i32* %addr)
782 %tmp5 = tail call i32 @doSomething(i32 6, i32* %addr)
786 %tmp.0 = phi i32 [ %tmp4, %true ], [ %tmp5, %false ]
790 @b = internal unnamed_addr global i1 false
791 @c = internal unnamed_addr global i8 0, align 1
792 @a = common global i32 0, align 4
794 ; Make sure the prologue does not clobber the EFLAGS when
795 ; it is live accross.
797 ; Note: The registers may change in the following patterns, but
798 ; because they imply register hierarchy (e.g., eax, al) this is
799 ; tricky to write robust patterns.
801 ; CHECK-LABEL: useLEAForPrologue:
803 ; Prologue is at the beginning of the function when shrink-wrapping
806 ; The stack adjustment can use SUB instr because we do not need to
807 ; preserve the EFLAGS at this point.
808 ; DISABLE-NEXT: subq $16, %rsp
810 ; Load the value of b.
811 ; CHECK: movb _b(%rip), [[BOOL:%cl]]
812 ; Create the zero value for the select assignment.
813 ; CHECK-NEXT: xorl [[CMOVE_VAL:%eax]], [[CMOVE_VAL]]
814 ; CHECK-NEXT: testb [[BOOL]], [[BOOL]]
815 ; CHECK-NEXT: jne [[STOREC_LABEL:LBB[0-9_]+]]
817 ; CHECK: movb $48, [[CMOVE_VAL:%al]]
819 ; CHECK: [[STOREC_LABEL]]:
822 ; For the stack adjustment, we need to preserve the EFLAGS.
823 ; ENABLE-NEXT: leaq -16(%rsp), %rsp
825 ; Technically, we should use CMOVE_VAL here or its subregister.
826 ; CHECK-NEXT: movb %al, _c(%rip)
827 ; testb set the EFLAGS read here.
828 ; CHECK-NEXT: je [[VARFUNC_CALL:LBB[0-9_]+]]
830 ; The code of the loop is not interesting.
833 ; CHECK: [[VARFUNC_CALL]]:
834 ; Set the null parameter.
835 ; CHECK-NEXT: xorl %edi, %edi
836 ; CHECK-NEXT: callq _varfunc
838 ; Set the return value.
839 ; CHECK-NEXT: xorl %eax, %eax
842 ; CHECK-NEXT: addq $16, %rsp
845 define i32 @useLEAForPrologue(i32 %d, i32 %a, i8 %c) #3 {
848 %.b = load i1, i1* @b, align 1
849 %bool = select i1 %.b, i8 0, i8 48
850 store i8 %bool, i8* @c, align 1
851 br i1 %.b, label %for.body.lr.ph, label %for.end
853 for.body.lr.ph: ; preds = %entry
854 tail call void asm sideeffect "nop", "~{ebx}"()
857 for.body: ; preds = %for.body.lr.ph, %for.body
858 %inc6 = phi i8 [ %c, %for.body.lr.ph ], [ %inc, %for.body ]
859 %cond5 = phi i32 [ %a, %for.body.lr.ph ], [ %conv3, %for.body ]
860 %cmp2 = icmp slt i32 %d, %cond5
861 %conv3 = zext i1 %cmp2 to i32
862 %inc = add i8 %inc6, 1
863 %cmp = icmp slt i8 %inc, 45
864 br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge
866 for.cond.for.end_crit_edge: ; preds = %for.body
867 store i32 %conv3, i32* @a, align 4
870 for.end: ; preds = %for.cond.for.end_crit_edge, %entry
871 %call = tail call i32 (i8*) @varfunc(i8* null)
875 declare i32 @varfunc(i8* nocapture readonly)
877 @sum1 = external hidden thread_local global i32, align 4
880 ; Function Attrs: nounwind
881 ; Make sure the TLS call used to access @sum1 happens after the prologue
882 ; and before the epilogue.
883 ; TLS calls used to be wrongly model and shrink-wrapping would have inserted
884 ; the prologue and epilogue just around the call to doSomething.
887 ; CHECK-LABEL: tlsCall:
889 ; CHECK: testb $1, %dil
890 ; CHECK: je [[ELSE_LABEL:LBB[0-9_]+]]
893 ; CHECK: movq _sum1@TLVP(%rip), %rdi
894 ; CHECK-NEXT: callq *(%rdi)
895 ; CHECK: jmp [[EXIT_LABEL:LBB[0-9_]+]]
898 ; CHECK: callq _doSomething
903 define i32 @tlsCall(i1 %bool1, i32 %arg, i32* readonly dereferenceable(4) %sum1) #3 {
905 br i1 %bool1, label %master, label %else
908 %tmp1 = load i32, i32* %sum1, align 4
909 store i32 %tmp1, i32* @sum1, align 4
913 %call = call i32 @doSomething(i32 0, i32* null)
917 %res = phi i32 [ %arg, %master], [ %call, %else ]
921 attributes #3 = { nounwind }
923 @irreducibleCFGa = common global i32 0, align 4
924 @irreducibleCFGf = common global i8 0, align 1
925 @irreducibleCFGb = common global i32 0, align 4
927 ; Check that we do not run shrink-wrapping on irreducible CFGs until
928 ; it is actually supported.
929 ; At the moment, on those CFGs the loop information may be incorrect
930 ; and since we use that information to do the placement, we may end up
931 ; inserting the prologue/epilogue at incorrect places.
934 ; CHECK-LABEL: irreducibleCFG:
936 ; Make sure the prologue happens in the entry block.
939 ; Make sure the epilogue happens in the exit block.
944 define i32 @irreducibleCFG() #4 {
946 %i0 = load i32, i32* @irreducibleCFGa, align 4
947 %.pr = load i8, i8* @irreducibleCFGf, align 1
948 %bool = icmp eq i8 %.pr, 0
949 br i1 %bool, label %split, label %preheader
955 %i1 = load i32, i32* @irreducibleCFGb, align 4
956 %tobool1.i = icmp ne i32 %i1, 0
957 br i1 %tobool1.i, label %for.body4.i, label %for.cond8.i.preheader
960 %call.i = tail call i32 (...) @something(i32 %i0)
964 %p1 = phi i32 [ %inc18.i, %for.inc ], [ 0, %for.body4.i ]
965 %.pr1.pr = load i32, i32* @irreducibleCFGb, align 4
966 br label %for.cond8.i.preheader
968 for.cond8.i.preheader:
969 %.pr1 = phi i32 [ %.pr1.pr, %for.cond8 ], [ %i1, %split ]
970 %p13 = phi i32 [ %p1, %for.cond8 ], [ 0, %split ]
977 %inc18.i = add nuw nsw i32 %p13, 1
978 %cmp = icmp slt i32 %inc18.i, 7
979 br i1 %cmp, label %for.cond8, label %fn1.exit
982 attributes #4 = { "no-frame-pointer-elim"="true" }
984 @x = external global i32, align 4
985 @y = external global i32, align 4
987 ; The post-dominator tree does not include the branch containing the infinite
988 ; loop, which can occur into a misplacement of the restore block, if we're
989 ; looking for the nearest common post-dominator of an "unreachable" block.
991 ; CHECK-LABEL: infiniteLoopNoSuccessor:
993 ; Make sure the prologue happens in the entry block.
994 ; CHECK-NEXT: pushq %rbp
996 ; Make sure we don't shrink-wrap.
998 ; CHECK-NOT: pushq %rbp
1000 ; Make sure the epilogue happens in the exit block.
1004 define void @infiniteLoopNoSuccessor() #5 {
1005 %1 = load i32, i32* @x, align 4
1006 %2 = icmp ne i32 %1, 0
1007 br i1 %2, label %3, label %4
1010 store i32 0, i32* @x, align 4
1014 call void (...) @somethingElse()
1015 %5 = load i32, i32* @y, align 4
1016 %6 = icmp ne i32 %5, 0
1017 br i1 %6, label %10, label %7
1020 %8 = call i32 (...) @something()
1024 call void (...) @somethingElse()
1031 declare void @somethingElse(...)
1033 attributes #5 = { nounwind "no-frame-pointer-elim-non-leaf" }