caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
dda73336ad22bd0b5ecda17040c50fb10fcbe5fb
clang-p2996/llvm/test/Transforms/TailCallElim/accum_recursion.ll
Joshua Cao 72ffaa9156 [IR][TRE] Support associative intrinsics (#74226) 
There is support for intrinsics in Instruction::isCommunative, but there
is no equivalent implementation for isAssociative. This patch builds
support for associative intrinsics with TRE as an application. TRE can
now have associative intrinsics as an accumulator. For example:
```
struct Node {
  Node *next;
  unsigned val;
}

unsigned maxval(struct Node *n) {
  if (!n) return 0;
  return std::max(n->val, maxval(n->next));
}
```
Can be transformed into:
```
unsigned maxval(struct Node *n) {
  struct Node *head = n;
  unsigned max = 0; // Identity of unsigned std::max
  while (true) {
    if (!head) return max;
    max = std::max(max, head->val);
    head = head->next;
  }
  return max;
}
```
This example results in about 5x speedup in local runs.

We conservatively only consider min/max and as associative for this
patch to limit testing scope. There are probably other intrinsics that
could be considered associative. There are a few consumers of
isAssociative() that could be impacted. Testing has only required to
Reassociate pass be updated.
2023-12-04 22:35:59 -08:00

331 lines
13 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
; RUN: opt < %s -passes=tailcallelim -verify-dom-info -S | FileCheck %s
define i32 @test1_factorial(i32 %x) {
; CHECK-LABEL: define i32 @test1_factorial(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[TAILRECURSE:%.*]]
; CHECK: tailrecurse:
; CHECK-NEXT: [[ACCUMULATOR_TR:%.*]] = phi i32 [ 1, [[ENTRY:%.*]] ], [ [[ACCUMULATE:%.*]], [[THEN:%.*]] ]
; CHECK-NEXT: [[X_TR:%.*]] = phi i32 [ [[X]], [[ENTRY]] ], [ [[TMP_6:%.*]], [[THEN]] ]
; CHECK-NEXT: [[TMP_1:%.*]] = icmp sgt i32 [[X_TR]], 0
; CHECK-NEXT: br i1 [[TMP_1]], label [[THEN]], label [[ELSE:%.*]]
; CHECK: then:
; CHECK-NEXT: [[TMP_6]] = add i32 [[X_TR]], -1
; CHECK-NEXT: [[ACCUMULATE]] = mul i32 [[ACCUMULATOR_TR]], [[X_TR]]
; CHECK-NEXT: br label [[TAILRECURSE]]
; CHECK: else:
; CHECK-NEXT: [[ACCUMULATOR_RET_TR:%.*]] = mul i32 [[ACCUMULATOR_TR]], 1
; CHECK-NEXT: ret i32 [[ACCUMULATOR_RET_TR]]
;
entry:
%tmp.1 = icmp sgt i32 %x, 0
br i1 %tmp.1, label %then, label %else
then:
%tmp.6 = add i32 %x, -1
%recurse = call i32 @test1_factorial( i32 %tmp.6 )
%accumulate = mul i32 %recurse, %x
ret i32 %accumulate
else:
ret i32 1
}
; This is a more aggressive form of accumulator recursion insertion, which
; requires noticing that X doesn't change as we perform the tailcall.
define i32 @test2_mul(i32 %x, i32 %y) {
; CHECK-LABEL: define i32 @test2_mul(
; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]]) {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[TAILRECURSE:%.*]]
; CHECK: tailrecurse:
; CHECK-NEXT: [[ACCUMULATOR_TR:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[ACCUMULATE:%.*]], [[ENDIF:%.*]] ]
; CHECK-NEXT: [[Y_TR:%.*]] = phi i32 [ [[Y]], [[ENTRY]] ], [ [[TMP_8:%.*]], [[ENDIF]] ]
; CHECK-NEXT: [[TMP_1:%.*]] = icmp eq i32 [[Y_TR]], 0
; CHECK-NEXT: br i1 [[TMP_1]], label [[RETURN:%.*]], label [[ENDIF]]
; CHECK: endif:
; CHECK-NEXT: [[TMP_8]] = add i32 [[Y_TR]], -1
; CHECK-NEXT: [[ACCUMULATE]] = add i32 [[ACCUMULATOR_TR]], [[X]]
; CHECK-NEXT: br label [[TAILRECURSE]]
; CHECK: return:
; CHECK-NEXT: [[ACCUMULATOR_RET_TR:%.*]] = add i32 [[ACCUMULATOR_TR]], [[X]]
; CHECK-NEXT: ret i32 [[ACCUMULATOR_RET_TR]]
;
entry:
%tmp.1 = icmp eq i32 %y, 0
br i1 %tmp.1, label %return, label %endif
endif:
%tmp.8 = add i32 %y, -1
%recurse = call i32 @test2_mul( i32 %x, i32 %tmp.8 )
%accumulate = add i32 %recurse, %x
ret i32 %accumulate
return:
ret i32 %x
}
define i64 @test3_fib(i64 %n) nounwind readnone {
; CHECK-LABEL: define i64 @test3_fib(
; CHECK-SAME: i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[TAILRECURSE:%.*]]
; CHECK: tailrecurse:
; CHECK-NEXT: [[ACCUMULATOR_TR:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[ACCUMULATE:%.*]], [[BB1:%.*]] ]
; CHECK-NEXT: [[N_TR:%.*]] = phi i64 [ [[N]], [[ENTRY]] ], [ [[TMP1:%.*]], [[BB1]] ]
; CHECK-NEXT: switch i64 [[N_TR]], label [[BB1]] [
; CHECK-NEXT: i64 0, label [[BB2:%.*]]
; CHECK-NEXT: i64 1, label [[BB2]]
; CHECK-NEXT: ]
; CHECK: bb1:
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N_TR]], -1
; CHECK-NEXT: [[RECURSE1:%.*]] = tail call i64 @test3_fib(i64 [[TMP0]]) #[[ATTR2:[0-9]+]]
; CHECK-NEXT: [[TMP1]] = add i64 [[N_TR]], -2
; CHECK-NEXT: [[ACCUMULATE]] = add nsw i64 [[ACCUMULATOR_TR]], [[RECURSE1]]
; CHECK-NEXT: br label [[TAILRECURSE]]
; CHECK: bb2:
; CHECK-NEXT: [[ACCUMULATOR_RET_TR:%.*]] = add nsw i64 [[ACCUMULATOR_TR]], [[N_TR]]
; CHECK-NEXT: ret i64 [[ACCUMULATOR_RET_TR]]
;
entry:
switch i64 %n, label %bb1 [
i64 0, label %bb2
i64 1, label %bb2
]
bb1:
%0 = add i64 %n, -1
%recurse1 = tail call i64 @test3_fib(i64 %0) nounwind
%1 = add i64 %n, -2
%recurse2 = tail call i64 @test3_fib(i64 %1) nounwind
%accumulate = add nsw i64 %recurse2, %recurse1
ret i64 %accumulate
bb2:
ret i64 %n
}
define i32 @test4_base_case_call() local_unnamed_addr {
; CHECK-LABEL: define i32 @test4_base_case_call() local_unnamed_addr {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[TAILRECURSE:%.*]]
; CHECK: tailrecurse:
; CHECK-NEXT: [[ACCUMULATOR_TR:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[ACCUMULATE:%.*]], [[SW_DEFAULT:%.*]] ]
; CHECK-NEXT: [[BASE:%.*]] = tail call i32 @test4_helper()
; CHECK-NEXT: switch i32 [[BASE]], label [[SW_DEFAULT]] [
; CHECK-NEXT: i32 1, label [[CLEANUP:%.*]]
; CHECK-NEXT: i32 5, label [[CLEANUP]]
; CHECK-NEXT: i32 7, label [[CLEANUP]]
; CHECK-NEXT: ]
; CHECK: sw.default:
; CHECK-NEXT: [[ACCUMULATE]] = add nsw i32 [[ACCUMULATOR_TR]], 1
; CHECK-NEXT: br label [[TAILRECURSE]]
; CHECK: cleanup:
; CHECK-NEXT: [[ACCUMULATOR_RET_TR:%.*]] = add nsw i32 [[ACCUMULATOR_TR]], [[BASE]]
; CHECK-NEXT: ret i32 [[ACCUMULATOR_RET_TR]]
;
entry:
%base = call i32 @test4_helper()
switch i32 %base, label %sw.default [
i32 1, label %cleanup
i32 5, label %cleanup
i32 7, label %cleanup
]
sw.default:
%recurse = call i32 @test4_base_case_call()
%accumulate = add nsw i32 %recurse, 1
br label %cleanup
cleanup:
%retval.0 = phi i32 [ %accumulate, %sw.default ], [ %base, %entry ], [ %base, %entry ], [ %base, %entry ]
ret i32 %retval.0
}
declare i32 @test4_helper()
define i32 @test5_base_case_load(ptr nocapture %A, i32 %n) local_unnamed_addr {
; CHECK-LABEL: define i32 @test5_base_case_load(
; CHECK-SAME: ptr nocapture [[A:%.*]], i32 [[N:%.*]]) local_unnamed_addr {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[TAILRECURSE:%.*]]
; CHECK: tailrecurse:
; CHECK-NEXT: [[ACCUMULATOR_TR:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[ACCUMULATE:%.*]], [[IF_END:%.*]] ]
; CHECK-NEXT: [[N_TR:%.*]] = phi i32 [ [[N]], [[ENTRY]] ], [ [[SUB:%.*]], [[IF_END]] ]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[N_TR]], 0
; CHECK-NEXT: br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END]]
; CHECK: if.then:
; CHECK-NEXT: [[BASE:%.*]] = load i32, ptr [[A]], align 4
; CHECK-NEXT: [[ACCUMULATOR_RET_TR:%.*]] = add i32 [[ACCUMULATOR_TR]], [[BASE]]
; CHECK-NEXT: ret i32 [[ACCUMULATOR_RET_TR]]
; CHECK: if.end:
; CHECK-NEXT: [[IDXPROM:%.*]] = zext i32 [[N_TR]] to i64
; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IDXPROM]]
; CHECK-NEXT: [[LOAD:%.*]] = load i32, ptr [[ARRAYIDX1]], align 4
; CHECK-NEXT: [[SUB]] = add i32 [[N_TR]], -1
; CHECK-NEXT: [[ACCUMULATE]] = add i32 [[ACCUMULATOR_TR]], [[LOAD]]
; CHECK-NEXT: br label [[TAILRECURSE]]
;
entry:
%cmp = icmp eq i32 %n, 0
br i1 %cmp, label %if.then, label %if.end
if.then:
%base = load i32, ptr %A, align 4
ret i32 %base
if.end:
%idxprom = zext i32 %n to i64
%arrayidx1 = getelementptr inbounds i32, ptr %A, i64 %idxprom
%load = load i32, ptr %arrayidx1, align 4
%sub = add i32 %n, -1
%recurse = tail call i32 @test5_base_case_load(ptr %A, i32 %sub)
%accumulate = add i32 %recurse, %load
ret i32 %accumulate
}
define i32 @test6_multiple_returns(i32 %x, i32 %y) local_unnamed_addr {
; CHECK-LABEL: define i32 @test6_multiple_returns(
; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]]) local_unnamed_addr {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[TAILRECURSE:%.*]]
; CHECK: tailrecurse:
; CHECK-NEXT: [[ACCUMULATOR_TR:%.*]] = phi i32 [ [[ACCUMULATOR_TR]], [[CASE99:%.*]] ], [ 0, [[ENTRY:%.*]] ], [ [[ACCUMULATE:%.*]], [[DEFAULT:%.*]] ]
; CHECK-NEXT: [[X_TR:%.*]] = phi i32 [ [[X]], [[ENTRY]] ], [ [[SUB1:%.*]], [[CASE99]] ], [ [[SUB2:%.*]], [[DEFAULT]] ]
; CHECK-NEXT: [[RET_TR:%.*]] = phi i32 [ poison, [[ENTRY]] ], [ [[CURRENT_RET_TR:%.*]], [[CASE99]] ], [ [[RET_TR]], [[DEFAULT]] ]
; CHECK-NEXT: [[RET_KNOWN_TR:%.*]] = phi i1 [ false, [[ENTRY]] ], [ true, [[CASE99]] ], [ [[RET_KNOWN_TR]], [[DEFAULT]] ]
; CHECK-NEXT: switch i32 [[X_TR]], label [[DEFAULT]] [
; CHECK-NEXT: i32 0, label [[CASE0:%.*]]
; CHECK-NEXT: i32 99, label [[CASE99]]
; CHECK-NEXT: ]
; CHECK: case0:
; CHECK-NEXT: [[HELPER:%.*]] = tail call i32 @test6_helper()
; CHECK-NEXT: [[ACCUMULATOR_RET_TR2:%.*]] = add i32 [[ACCUMULATOR_TR]], [[HELPER]]
; CHECK-NEXT: [[CURRENT_RET_TR1:%.*]] = select i1 [[RET_KNOWN_TR]], i32 [[RET_TR]], i32 [[ACCUMULATOR_RET_TR2]]
; CHECK-NEXT: ret i32 [[CURRENT_RET_TR1]]
; CHECK: case99:
; CHECK-NEXT: [[SUB1]] = add i32 [[X_TR]], -1
; CHECK-NEXT: [[ACCUMULATOR_RET_TR:%.*]] = add i32 [[ACCUMULATOR_TR]], 18
; CHECK-NEXT: [[CURRENT_RET_TR]] = select i1 [[RET_KNOWN_TR]], i32 [[RET_TR]], i32 [[ACCUMULATOR_RET_TR]]
; CHECK-NEXT: br label [[TAILRECURSE]]
; CHECK: default:
; CHECK-NEXT: [[SUB2]] = add i32 [[X_TR]], -1
; CHECK-NEXT: [[ACCUMULATE]] = add i32 [[ACCUMULATOR_TR]], [[Y]]
; CHECK-NEXT: br label [[TAILRECURSE]]
;
entry:
switch i32 %x, label %default [
i32 0, label %case0
i32 99, label %case99
]
case0:
%helper = call i32 @test6_helper()
ret i32 %helper
case99:
%sub1 = add i32 %x, -1
%recurse1 = call i32 @test6_multiple_returns(i32 %sub1, i32 %y)
ret i32 18
default:
%sub2 = add i32 %x, -1
%recurse2 = call i32 @test6_multiple_returns(i32 %sub2, i32 %y)
%accumulate = add i32 %recurse2, %y
ret i32 %accumulate
}
declare i32 @test6_helper()
; It is only safe to transform one accumulator per function, make sure we don't
; try to remove more.
define i32 @test7_multiple_accumulators(i32 %a) local_unnamed_addr {
; CHECK-LABEL: define i32 @test7_multiple_accumulators(
; CHECK-SAME: i32 [[A:%.*]]) local_unnamed_addr {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[TAILRECURSE:%.*]]
; CHECK: tailrecurse:
; CHECK-NEXT: [[ACCUMULATOR_TR:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[ACCUMULATE1:%.*]], [[IF_THEN2:%.*]] ]
; CHECK-NEXT: [[A_TR:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[SUB:%.*]], [[IF_THEN2]] ]
; CHECK-NEXT: [[TOBOOL:%.*]] = icmp eq i32 [[A_TR]], 0
; CHECK-NEXT: br i1 [[TOBOOL]], label [[RETURN:%.*]], label [[IF_END:%.*]]
; CHECK: if.end:
; CHECK-NEXT: [[AND:%.*]] = and i32 [[A_TR]], 1
; CHECK-NEXT: [[TOBOOL1:%.*]] = icmp eq i32 [[AND]], 0
; CHECK-NEXT: [[SUB]] = add nsw i32 [[A_TR]], -1
; CHECK-NEXT: br i1 [[TOBOOL1]], label [[IF_END3:%.*]], label [[IF_THEN2]]
; CHECK: if.then2:
; CHECK-NEXT: [[ACCUMULATE1]] = add nsw i32 [[ACCUMULATOR_TR]], 1
; CHECK-NEXT: br label [[TAILRECURSE]]
; CHECK: if.end3:
; CHECK-NEXT: [[RECURSE2:%.*]] = tail call i32 @test7_multiple_accumulators(i32 [[SUB]])
; CHECK-NEXT: [[ACCUMULATE2:%.*]] = mul nsw i32 [[RECURSE2]], 2
; CHECK-NEXT: [[ACCUMULATOR_RET_TR:%.*]] = add nsw i32 [[ACCUMULATOR_TR]], [[ACCUMULATE2]]
; CHECK-NEXT: ret i32 [[ACCUMULATOR_RET_TR]]
; CHECK: return:
; CHECK-NEXT: [[ACCUMULATOR_RET_TR1:%.*]] = add nsw i32 [[ACCUMULATOR_TR]], 0
; CHECK-NEXT: ret i32 [[ACCUMULATOR_RET_TR1]]
;
entry:
%tobool = icmp eq i32 %a, 0
br i1 %tobool, label %return, label %if.end
if.end:
%and = and i32 %a, 1
%tobool1 = icmp eq i32 %and, 0
%sub = add nsw i32 %a, -1
br i1 %tobool1, label %if.end3, label %if.then2
if.then2:
%recurse1 = tail call i32 @test7_multiple_accumulators(i32 %sub)
%accumulate1 = add nsw i32 %recurse1, 1
br label %return
if.end3:
%recurse2 = tail call i32 @test7_multiple_accumulators(i32 %sub)
%accumulate2 = mul nsw i32 %recurse2, 2
br label %return
return:
%retval.0 = phi i32 [ %accumulate1, %if.then2 ], [ %accumulate2, %if.end3 ], [ 0, %entry ]
ret i32 %retval.0
}
%struct.ListNode = type { i32, ptr }
; We cannot TRE commutative, non-associative intrinsics
define i32 @test_non_associative_sadd_sat(ptr %a) local_unnamed_addr {
; CHECK-LABEL: define i32 @test_non_associative_sadd_sat(
; CHECK-SAME: ptr [[A:%.*]]) local_unnamed_addr {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq ptr [[A]], null
; CHECK-NEXT: br i1 [[TOBOOL_NOT]], label [[COMMON_RET6:%.*]], label [[IF_END:%.*]]
; CHECK: common.ret6:
; CHECK-NEXT: ret i32 -1
; CHECK: if.end:
; CHECK-NEXT: [[TMP0:%.*]] = load i32, ptr [[A]], align 4
; CHECK-NEXT: [[NEXT:%.*]] = getelementptr inbounds [[STRUCT_LISTNODE:%.*]], ptr [[A]], i64 0, i32 1
; CHECK-NEXT: [[TMP1:%.*]] = load ptr, ptr [[NEXT]], align 8
; CHECK-NEXT: [[CALL:%.*]] = tail call i32 @test_non_associative_sadd_sat(ptr [[TMP1]])
; CHECK-NEXT: [[DOTSROA_SPECULATED:%.*]] = tail call i32 @llvm.sadd.sat.i32(i32 [[TMP0]], i32 [[CALL]])
; CHECK-NEXT: ret i32 [[DOTSROA_SPECULATED]]
;
entry:
%tobool.not = icmp eq ptr %a, null
br i1 %tobool.not, label %common.ret6, label %if.end
common.ret6: ; preds = %entry, %if.end
%common.ret6.op = phi i32 [ %.sroa.speculated, %if.end ], [ -1, %entry ]
ret i32 %common.ret6.op
if.end: ; preds = %entry
%0 = load i32, ptr %a
%next = getelementptr inbounds %struct.ListNode, ptr %a, i64 0, i32 1
%1 = load ptr, ptr %next
%call = tail call i32 @test_non_associative_sadd_sat(ptr %1)
%.sroa.speculated = tail call i32 @llvm.sadd.sat.i32(i32 %0, i32 %call)
br label %common.ret6
}
declare i32 @llvm.sadd.sat.i32(i32, i32)
Reference in New Issue View Git Blame Copy Permalink