mirror of
https://github.com/apple/swift.git
synced 2025-12-21 12:14:44 +01:00
2e5b820063
It is more profitable in most cases to inline the big callee rather than unroll the loop, because unrolling it would create a lot of calls which cannot be further optimized due to increased size of the function containing a loop.
284 lines
8.4 KiB
Plaintext
284 lines
8.4 KiB
Plaintext
// RUN: %target-sil-opt -assume-parsing-unqualified-ownership-sil -enable-sil-verify-all -loop-unroll %s | %FileCheck %s
|
|
|
|
sil_stage canonical
|
|
|
|
import Builtin
|
|
|
|
// CHECK-LABEL: sil @loop_unroll_1
|
|
// CHECK: bb0
|
|
// CHECK: br bb1
|
|
// CHECK: bb1
|
|
// CHECK: builtin "sadd_with_overflow_Int64
|
|
// CHECK: cond_br {{.*}}, bb2, bb3
|
|
// CHECK: bb2:
|
|
// CHECK: return
|
|
// CHECK: bb3
|
|
// CHECK: builtin "sadd_with_overflow_Int64
|
|
// CHECK: br bb2
|
|
|
|
sil @loop_unroll_1 : $@convention(thin) () -> () {
|
|
bb0:
|
|
%0 = integer_literal $Builtin.Int64, 0
|
|
%1 = integer_literal $Builtin.Int64, 1
|
|
%2 = integer_literal $Builtin.Int64, 2
|
|
%3 = integer_literal $Builtin.Int1, 1
|
|
br bb1(%0 : $Builtin.Int64)
|
|
|
|
bb1(%4 : $Builtin.Int64):
|
|
%5 = builtin "sadd_with_overflow_Int64"(%4 : $Builtin.Int64, %1 : $Builtin.Int64, %3 : $Builtin.Int1) : $(Builtin.Int64, Builtin.Int1)
|
|
%6 = tuple_extract %5 : $(Builtin.Int64, Builtin.Int1), 0
|
|
%7 = builtin "cmp_eq_Int64"(%6 : $Builtin.Int64, %2 : $Builtin.Int64) : $Builtin.Int1
|
|
cond_br %7, bb2, bb1(%6 : $Builtin.Int64)
|
|
|
|
bb2:
|
|
%8 = tuple()
|
|
return %8 : $()
|
|
}
|
|
|
|
// CHECK-LABEL: sil @loop_unroll_2
|
|
// CHECK: bb0:
|
|
// CHECK: br bb1
|
|
// CHECK: bb1
|
|
// CHECK: = builtin "sadd_with_overflow_Int64
|
|
// CHECK: cond_br {{.*}}, bb3, bb2
|
|
// CHECK: bb2:
|
|
// CHECK: br bb4
|
|
// CHECK: bb3:
|
|
// CHECK: return
|
|
// CHECK: bb4
|
|
// CHECK: = builtin "sadd_with_overflow_Int64
|
|
// CHECK: br bb3
|
|
|
|
sil @loop_unroll_2 : $@convention(thin) () -> () {
|
|
bb0:
|
|
%0 = integer_literal $Builtin.Int64, 0
|
|
%1 = integer_literal $Builtin.Int64, 1
|
|
%2 = integer_literal $Builtin.Int64, 2
|
|
%3 = integer_literal $Builtin.Int1, 1
|
|
br bb1(%0 : $Builtin.Int64)
|
|
|
|
bb1(%4 : $Builtin.Int64):
|
|
%5 = builtin "sadd_with_overflow_Int64"(%4 : $Builtin.Int64, %1 : $Builtin.Int64, %3 : $Builtin.Int1) : $(Builtin.Int64, Builtin.Int1)
|
|
%6 = tuple_extract %5 : $(Builtin.Int64, Builtin.Int1), 0
|
|
%7 = builtin "cmp_eq_Int64"(%6 : $Builtin.Int64, %2 : $Builtin.Int64) : $Builtin.Int1
|
|
cond_br %7, bb3, bb2
|
|
|
|
bb2:
|
|
br bb1(%6 : $Builtin.Int64)
|
|
|
|
bb3:
|
|
%8 = tuple()
|
|
return %8 : $()
|
|
}
|
|
|
|
// CHECK-LABEL: sil @loop_unroll_3
|
|
// CHECK: bb0:
|
|
// CHECK: br bb1
|
|
// CHECK: bb1
|
|
// CHECK: = builtin "sadd_with_overflow_Int64
|
|
// CHECK: cond_br {{.*}}, bb3, bb2
|
|
// CHECK: bb2:
|
|
// CHECK: br bb4
|
|
// CHECK: bb3:
|
|
// CHECK: return
|
|
// CHECK: bb4
|
|
// CHECK: = builtin "sadd_with_overflow_Int64
|
|
// CHECK: br bb3
|
|
|
|
sil @loop_unroll_3 : $@convention(thin) () -> () {
|
|
bb0:
|
|
%0 = integer_literal $Builtin.Int64, 0
|
|
%1 = integer_literal $Builtin.Int64, 1
|
|
%2 = integer_literal $Builtin.Int64, 2
|
|
%3 = integer_literal $Builtin.Int1, 1
|
|
br bb1(%0 : $Builtin.Int64)
|
|
|
|
bb1(%4 : $Builtin.Int64):
|
|
%5 = builtin "sadd_with_overflow_Int64"(%4 : $Builtin.Int64, %1 : $Builtin.Int64, %3 : $Builtin.Int1) : $(Builtin.Int64, Builtin.Int1)
|
|
%6 = tuple_extract %5 : $(Builtin.Int64, Builtin.Int1), 0
|
|
%7 = builtin "cmp_sge_Int64"(%6 : $Builtin.Int64, %2 : $Builtin.Int64) : $Builtin.Int1
|
|
cond_br %7, bb3, bb2
|
|
|
|
bb2:
|
|
br bb1(%6 : $Builtin.Int64)
|
|
|
|
bb3:
|
|
%8 = tuple()
|
|
return %8 : $()
|
|
}
|
|
|
|
// CHECK-LABEL: sil @loop_unroll_4
|
|
// CHECK: bb0:
|
|
// CHECK: br bb1
|
|
// CHECK: bb1
|
|
// CHECK: = builtin "sadd_with_overflow_Int64
|
|
// CHECK: cond_br {{.*}}, bb3, bb2
|
|
// CHECK: bb2:
|
|
// CHECK: br bb4
|
|
// CHECK: bb3:
|
|
// CHECK: return
|
|
// CHECK: bb4
|
|
// CHECK: = builtin "sadd_with_overflow_Int64
|
|
// CHECK: br bb3
|
|
|
|
sil @loop_unroll_4 : $@convention(thin) () -> () {
|
|
bb0:
|
|
%0 = integer_literal $Builtin.Int64, 0
|
|
%1 = integer_literal $Builtin.Int64, 1
|
|
%2 = integer_literal $Builtin.Int64, 1
|
|
%3 = integer_literal $Builtin.Int1, 1
|
|
br bb1(%0 : $Builtin.Int64)
|
|
|
|
bb1(%4 : $Builtin.Int64):
|
|
%5 = builtin "sadd_with_overflow_Int64"(%4 : $Builtin.Int64, %1 : $Builtin.Int64, %3 : $Builtin.Int1) : $(Builtin.Int64, Builtin.Int1)
|
|
%6 = tuple_extract %5 : $(Builtin.Int64, Builtin.Int1), 0
|
|
%7 = builtin "cmp_sgt_Int64"(%6 : $Builtin.Int64, %2 : $Builtin.Int64) : $Builtin.Int1
|
|
cond_br %7, bb3, bb2
|
|
|
|
bb2:
|
|
br bb1(%6 : $Builtin.Int64)
|
|
|
|
bb3:
|
|
%8 = tuple()
|
|
return %8 : $()
|
|
}
|
|
|
|
class B {}
|
|
|
|
// CHECK-LABEL: sil @unroll_with_stack_allocation
|
|
// CHECK: = alloc_ref [stack]
|
|
// CHECK: dealloc_ref [stack]
|
|
// CHECK: = alloc_ref [stack]
|
|
// CHECK: dealloc_ref [stack]
|
|
// CHECK: }
|
|
sil @unroll_with_stack_allocation : $@convention(thin) () -> () {
|
|
bb0:
|
|
%0 = integer_literal $Builtin.Int64, 0
|
|
%1 = integer_literal $Builtin.Int64, 1
|
|
%2 = integer_literal $Builtin.Int64, 2
|
|
%3 = integer_literal $Builtin.Int1, 1
|
|
br bb1(%0 : $Builtin.Int64)
|
|
|
|
bb1(%4 : $Builtin.Int64):
|
|
%a = alloc_ref [stack] $B
|
|
br bb2
|
|
|
|
bb2:
|
|
%5 = builtin "sadd_with_overflow_Int64"(%4 : $Builtin.Int64, %1 : $Builtin.Int64, %3 : $Builtin.Int1) : $(Builtin.Int64, Builtin.Int1)
|
|
%6 = tuple_extract %5 : $(Builtin.Int64, Builtin.Int1), 0
|
|
%7 = builtin "cmp_eq_Int64"(%6 : $Builtin.Int64, %2 : $Builtin.Int64) : $Builtin.Int1
|
|
dealloc_ref [stack] %a : $B
|
|
cond_br %7, bb4, bb3
|
|
|
|
bb3:
|
|
br bb1(%6 : $Builtin.Int64)
|
|
|
|
bb4:
|
|
%8 = tuple()
|
|
return %8 : $()
|
|
}
|
|
|
|
// CHECK-LABEL: sil @dont_copy_stack_allocation_with_dealloc_outside_loop
|
|
// CHECK: = alloc_ref [stack]
|
|
// CHECK: bb2:
|
|
// CHECK: dealloc_ref [stack]
|
|
// CHECK: bb3:
|
|
// CHECK: dealloc_ref [stack]
|
|
// CHECK-NEXT: tuple
|
|
// CHECK-NEXT: return
|
|
sil @dont_copy_stack_allocation_with_dealloc_outside_loop : $@convention(thin) () -> () {
|
|
bb0:
|
|
%0 = integer_literal $Builtin.Int64, 0
|
|
%1 = integer_literal $Builtin.Int64, 1
|
|
%2 = integer_literal $Builtin.Int64, 2
|
|
%3 = integer_literal $Builtin.Int1, 1
|
|
br bb1(%0 : $Builtin.Int64)
|
|
|
|
bb1(%4 : $Builtin.Int64):
|
|
%a = alloc_ref [stack] $B
|
|
%5 = builtin "sadd_with_overflow_Int64"(%4 : $Builtin.Int64, %1 : $Builtin.Int64, %3 : $Builtin.Int1) : $(Builtin.Int64, Builtin.Int1)
|
|
%6 = tuple_extract %5 : $(Builtin.Int64, Builtin.Int1), 0
|
|
%7 = builtin "cmp_eq_Int64"(%6 : $Builtin.Int64, %2 : $Builtin.Int64) : $Builtin.Int1
|
|
cond_br %7, bb3, bb2
|
|
|
|
bb2:
|
|
dealloc_ref [stack] %a : $B
|
|
br bb1(%6 : $Builtin.Int64)
|
|
|
|
bb3:
|
|
dealloc_ref [stack] %a : $B
|
|
%8 = tuple()
|
|
return %8 : $()
|
|
}
|
|
|
|
sil @big_func: $@convention(thin) () -> Builtin.Int64 {
|
|
bb0:
|
|
%x0 = integer_literal $Builtin.Int64, 1
|
|
%overflow_check = integer_literal $Builtin.Int1, 0
|
|
%sum1 = builtin "sadd_with_overflow_Int64"(%x0 : $Builtin.Int64, %x0 : $Builtin.Int64, %overflow_check : $Builtin.Int1) : $(Builtin.Int64, Builtin.Int1)
|
|
%x1 = tuple_extract %sum1 : $(Builtin.Int64, Builtin.Int1), 0
|
|
br bb1
|
|
|
|
bb1:
|
|
%sum2 = builtin "sadd_with_overflow_Int64"(%x1 : $Builtin.Int64, %x1 : $Builtin.Int64, %overflow_check : $Builtin.Int1) : $(Builtin.Int64, Builtin.Int1)
|
|
%x2 = tuple_extract %sum2 : $(Builtin.Int64, Builtin.Int1), 0
|
|
br bb2
|
|
|
|
bb2:
|
|
%sum3 = builtin "sadd_with_overflow_Int64"(%x2 : $Builtin.Int64, %x2 : $Builtin.Int64, %overflow_check : $Builtin.Int1) : $(Builtin.Int64, Builtin.Int1)
|
|
%x3 = tuple_extract %sum3 : $(Builtin.Int64, Builtin.Int1), 0
|
|
br bb3
|
|
|
|
bb3:
|
|
%sum4 = builtin "sadd_with_overflow_Int64"(%x3 : $Builtin.Int64, %x3 : $Builtin.Int64, %overflow_check : $Builtin.Int1) : $(Builtin.Int64, Builtin.Int1)
|
|
%x4 = tuple_extract %sum4 : $(Builtin.Int64, Builtin.Int1), 0
|
|
br bb4
|
|
|
|
bb4:
|
|
%sum5 = builtin "sadd_with_overflow_Int64"(%x4 : $Builtin.Int64, %x4 : $Builtin.Int64, %overflow_check : $Builtin.Int1) : $(Builtin.Int64, Builtin.Int1)
|
|
%x5 = tuple_extract %sum5 : $(Builtin.Int64, Builtin.Int1), 0
|
|
br bb5
|
|
|
|
bb5:
|
|
%sum6 = builtin "sadd_with_overflow_Int64"(%x5 : $Builtin.Int64, %x5 : $Builtin.Int64, %overflow_check : $Builtin.Int1) : $(Builtin.Int64, Builtin.Int1)
|
|
%x6 = tuple_extract %sum6 : $(Builtin.Int64, Builtin.Int1), 0
|
|
br bb6
|
|
|
|
bb6:
|
|
return %x6 : $Builtin.Int64
|
|
}
|
|
|
|
// Check that the compiler does not unroll loops containing calls
|
|
// of big inlineable functions.
|
|
//
|
|
// CHECK-LABEL: sil @unroll_with_apply
|
|
// CHECK: apply
|
|
// CHECK: // end sil function 'unroll_with_apply'
|
|
sil @unroll_with_apply : $@convention(thin) () -> () {
|
|
bb0:
|
|
%0 = integer_literal $Builtin.Int64, 0
|
|
%1 = integer_literal $Builtin.Int64, 1
|
|
%2 = integer_literal $Builtin.Int64, 20
|
|
%3 = integer_literal $Builtin.Int1, 1
|
|
%f = function_ref @big_func: $@convention(thin) () -> Builtin.Int64
|
|
br bb1(%0 : $Builtin.Int64)
|
|
|
|
bb1(%4 : $Builtin.Int64):
|
|
%r = apply %f() : $@convention(thin) () -> Builtin.Int64
|
|
br bb2
|
|
|
|
bb2:
|
|
%5 = builtin "sadd_with_overflow_Int64"(%4 : $Builtin.Int64, %1 : $Builtin.Int64, %3 : $Builtin.Int1) : $(Builtin.Int64, Builtin.Int1)
|
|
%6 = tuple_extract %5 : $(Builtin.Int64, Builtin.Int1), 0
|
|
%7 = builtin "cmp_eq_Int64"(%6 : $Builtin.Int64, %2 : $Builtin.Int64) : $Builtin.Int1
|
|
cond_br %7, bb4, bb3
|
|
|
|
bb3:
|
|
br bb1(%6 : $Builtin.Int64)
|
|
|
|
bb4:
|
|
%8 = tuple()
|
|
return %8 : $()
|
|
}
|
|
|