RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-02-01 05:01:59 +01:00
Code Issues Projects Releases Wiki Activity
llvm-mirror/test/CodeGen/AArch64/hoist-and-by-const-from-lshr-in-eqcmp-zero.ll
Roman Lebedev a0525b084a [Codegen] (X & (C l>>/<< Y)) ==/!= 0 --> ((X <</l>> Y) & C) ==/!= 0 fold 
Summary:
This was originally reported in D62818.
https://rise4fun.com/Alive/oPH

InstCombine does the opposite fold, in hope that `C l>>/<< Y` expression
will be hoisted out of a loop if `Y` is invariant and `X` is not.
But as it is seen from the diffs here, if it didn't get hoisted,
the produced assembly is almost universally worse.

Much like with my recent "hoist add/sub by/from const" patches,
we should get almost universal win if we hoist constant,
there is almost always an "and/test by imm" instruction,
but "shift of imm" not so much, so we may avoid having to
materialize the immediate, and thus need one less register.
And since we now shift not by constant, but by something else,
the live-range of that something else may reduce.

Special care needs to be applied not to disturb x86 `BT` / hexagon `tstbit`
instruction pattern. And to not get into endless combine loop.

Reviewers: RKSimon, efriedma, t.p.northover, craig.topper, spatel, arsenm

Reviewed By: spatel

Subscribers: hiraditya, MaskRay, wuzish, xbolva00, nikic, nemanjai, jvesely, wdng, nhaehnle, javed.absar, tpr, kristof.beyls, jsji, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62871

llvm-svn: 366955
2019-07-24 22:57:22 +00:00

352 lines
10 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=aarch64-unknown-unknown < %s | FileCheck %s --check-prefixes=CHECK,AARCH64
; We are looking for the following pattern here:
; (X & (C l>> Y)) ==/!= 0
; It may be optimal to hoist the constant:
; ((X << Y) & C) ==/!= 0
;------------------------------------------------------------------------------;
; A few scalar test
;------------------------------------------------------------------------------;
; i8 scalar
define i1 @scalar_i8_signbit_eq(i8 %x, i8 %y) nounwind {
; CHECK-LABEL: scalar_i8_signbit_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: // kill: def $w1 killed $w1 def $x1
; CHECK-NEXT: lsl w8, w0, w1
; CHECK-NEXT: tst w8, #0x80
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i8 128, %y
%t1 = and i8 %t0, %x
%res = icmp eq i8 %t1, 0
ret i1 %res
}
define i1 @scalar_i8_lowestbit_eq(i8 %x, i8 %y) nounwind {
; CHECK-LABEL: scalar_i8_lowestbit_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: // kill: def $w1 killed $w1 def $x1
; CHECK-NEXT: lsl w8, w0, w1
; CHECK-NEXT: tst w8, #0x1
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i8 1, %y
%t1 = and i8 %t0, %x
%res = icmp eq i8 %t1, 0
ret i1 %res
}
define i1 @scalar_i8_bitsinmiddle_eq(i8 %x, i8 %y) nounwind {
; CHECK-LABEL: scalar_i8_bitsinmiddle_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: // kill: def $w1 killed $w1 def $x1
; CHECK-NEXT: lsl w8, w0, w1
; CHECK-NEXT: tst w8, #0x18
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i8 24, %y
%t1 = and i8 %t0, %x
%res = icmp eq i8 %t1, 0
ret i1 %res
}
; i16 scalar
define i1 @scalar_i16_signbit_eq(i16 %x, i16 %y) nounwind {
; CHECK-LABEL: scalar_i16_signbit_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: // kill: def $w1 killed $w1 def $x1
; CHECK-NEXT: lsl w8, w0, w1
; CHECK-NEXT: tst w8, #0x8000
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i16 32768, %y
%t1 = and i16 %t0, %x
%res = icmp eq i16 %t1, 0
ret i1 %res
}
define i1 @scalar_i16_lowestbit_eq(i16 %x, i16 %y) nounwind {
; CHECK-LABEL: scalar_i16_lowestbit_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: // kill: def $w1 killed $w1 def $x1
; CHECK-NEXT: lsl w8, w0, w1
; CHECK-NEXT: tst w8, #0x1
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i16 1, %y
%t1 = and i16 %t0, %x
%res = icmp eq i16 %t1, 0
ret i1 %res
}
define i1 @scalar_i16_bitsinmiddle_eq(i16 %x, i16 %y) nounwind {
; CHECK-LABEL: scalar_i16_bitsinmiddle_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: // kill: def $w1 killed $w1 def $x1
; CHECK-NEXT: lsl w8, w0, w1
; CHECK-NEXT: tst w8, #0xff0
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i16 4080, %y
%t1 = and i16 %t0, %x
%res = icmp eq i16 %t1, 0
ret i1 %res
}
; i32 scalar
define i1 @scalar_i32_signbit_eq(i32 %x, i32 %y) nounwind {
; CHECK-LABEL: scalar_i32_signbit_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: lsl w8, w0, w1
; CHECK-NEXT: tst w8, #0x80000000
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i32 2147483648, %y
%t1 = and i32 %t0, %x
%res = icmp eq i32 %t1, 0
ret i1 %res
}
define i1 @scalar_i32_lowestbit_eq(i32 %x, i32 %y) nounwind {
; CHECK-LABEL: scalar_i32_lowestbit_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: lsl w8, w0, w1
; CHECK-NEXT: tst w8, #0x1
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i32 1, %y
%t1 = and i32 %t0, %x
%res = icmp eq i32 %t1, 0
ret i1 %res
}
define i1 @scalar_i32_bitsinmiddle_eq(i32 %x, i32 %y) nounwind {
; CHECK-LABEL: scalar_i32_bitsinmiddle_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: lsl w8, w0, w1
; CHECK-NEXT: tst w8, #0xffff00
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i32 16776960, %y
%t1 = and i32 %t0, %x
%res = icmp eq i32 %t1, 0
ret i1 %res
}
; i64 scalar
define i1 @scalar_i64_signbit_eq(i64 %x, i64 %y) nounwind {
; CHECK-LABEL: scalar_i64_signbit_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: lsl x8, x0, x1
; CHECK-NEXT: tst x8, #0x8000000000000000
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i64 9223372036854775808, %y
%t1 = and i64 %t0, %x
%res = icmp eq i64 %t1, 0
ret i1 %res
}
define i1 @scalar_i64_lowestbit_eq(i64 %x, i64 %y) nounwind {
; CHECK-LABEL: scalar_i64_lowestbit_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: lsl x8, x0, x1
; CHECK-NEXT: tst x8, #0x1
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i64 1, %y
%t1 = and i64 %t0, %x
%res = icmp eq i64 %t1, 0
ret i1 %res
}
define i1 @scalar_i64_bitsinmiddle_eq(i64 %x, i64 %y) nounwind {
; CHECK-LABEL: scalar_i64_bitsinmiddle_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: lsl x8, x0, x1
; CHECK-NEXT: tst x8, #0xffffffff0000
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i64 281474976645120, %y
%t1 = and i64 %t0, %x
%res = icmp eq i64 %t1, 0
ret i1 %res
}
;------------------------------------------------------------------------------;
; A few trivial vector tests
;------------------------------------------------------------------------------;
define <4 x i1> @vec_4xi32_splat_eq(<4 x i32> %x, <4 x i32> %y) nounwind {
; CHECK-LABEL: vec_4xi32_splat_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: movi v2.4s, #1
; CHECK-NEXT: ushl v0.4s, v0.4s, v1.4s
; CHECK-NEXT: and v0.16b, v0.16b, v2.16b
; CHECK-NEXT: cmeq v0.4s, v0.4s, #0
; CHECK-NEXT: xtn v0.4h, v0.4s
; CHECK-NEXT: ret
%t0 = lshr <4 x i32> <i32 1, i32 1, i32 1, i32 1>, %y
%t1 = and <4 x i32> %t0, %x
%res = icmp eq <4 x i32> %t1, <i32 0, i32 0, i32 0, i32 0>
ret <4 x i1> %res
}
define <4 x i1> @vec_4xi32_nonsplat_eq(<4 x i32> %x, <4 x i32> %y) nounwind {
; CHECK-LABEL: vec_4xi32_nonsplat_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: adrp x8, .LCPI13_0
; CHECK-NEXT: ldr q2, [x8, :lo12:.LCPI13_0]
; CHECK-NEXT: neg v1.4s, v1.4s
; CHECK-NEXT: ushl v1.4s, v2.4s, v1.4s
; CHECK-NEXT: and v0.16b, v1.16b, v0.16b
; CHECK-NEXT: cmeq v0.4s, v0.4s, #0
; CHECK-NEXT: xtn v0.4h, v0.4s
; CHECK-NEXT: ret
%t0 = lshr <4 x i32> <i32 0, i32 1, i32 16776960, i32 2147483648>, %y
%t1 = and <4 x i32> %t0, %x
%res = icmp eq <4 x i32> %t1, <i32 0, i32 0, i32 0, i32 0>
ret <4 x i1> %res
}
define <4 x i1> @vec_4xi32_nonsplat_undef0_eq(<4 x i32> %x, <4 x i32> %y) nounwind {
; CHECK-LABEL: vec_4xi32_nonsplat_undef0_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: movi v2.4s, #1
; CHECK-NEXT: ushl v0.4s, v0.4s, v1.4s
; CHECK-NEXT: and v0.16b, v0.16b, v2.16b
; CHECK-NEXT: cmeq v0.4s, v0.4s, #0
; CHECK-NEXT: xtn v0.4h, v0.4s
; CHECK-NEXT: ret
%t0 = lshr <4 x i32> <i32 1, i32 1, i32 undef, i32 1>, %y
%t1 = and <4 x i32> %t0, %x
%res = icmp eq <4 x i32> %t1, <i32 0, i32 0, i32 0, i32 0>
ret <4 x i1> %res
}
define <4 x i1> @vec_4xi32_nonsplat_undef1_eq(<4 x i32> %x, <4 x i32> %y) nounwind {
; CHECK-LABEL: vec_4xi32_nonsplat_undef1_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: neg v1.4s, v1.4s
; CHECK-NEXT: movi v2.4s, #1
; CHECK-NEXT: ushl v1.4s, v2.4s, v1.4s
; CHECK-NEXT: and v0.16b, v1.16b, v0.16b
; CHECK-NEXT: cmeq v0.4s, v0.4s, #0
; CHECK-NEXT: xtn v0.4h, v0.4s
; CHECK-NEXT: ret
%t0 = lshr <4 x i32> <i32 1, i32 1, i32 1, i32 1>, %y
%t1 = and <4 x i32> %t0, %x
%res = icmp eq <4 x i32> %t1, <i32 0, i32 0, i32 undef, i32 0>
ret <4 x i1> %res
}
define <4 x i1> @vec_4xi32_nonsplat_undef2_eq(<4 x i32> %x, <4 x i32> %y) nounwind {
; CHECK-LABEL: vec_4xi32_nonsplat_undef2_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: neg v1.4s, v1.4s
; CHECK-NEXT: movi v2.4s, #1
; CHECK-NEXT: ushl v1.4s, v2.4s, v1.4s
; CHECK-NEXT: and v0.16b, v1.16b, v0.16b
; CHECK-NEXT: cmeq v0.4s, v0.4s, #0
; CHECK-NEXT: xtn v0.4h, v0.4s
; CHECK-NEXT: ret
%t0 = lshr <4 x i32> <i32 1, i32 1, i32 undef, i32 1>, %y
%t1 = and <4 x i32> %t0, %x
%res = icmp eq <4 x i32> %t1, <i32 0, i32 0, i32 undef, i32 0>
ret <4 x i1> %res
}
;------------------------------------------------------------------------------;
; A special tests
;------------------------------------------------------------------------------;
define i1 @scalar_i8_signbit_ne(i8 %x, i8 %y) nounwind {
; CHECK-LABEL: scalar_i8_signbit_ne:
; CHECK: // %bb.0:
; CHECK-NEXT: // kill: def $w1 killed $w1 def $x1
; CHECK-NEXT: lsl w8, w0, w1
; CHECK-NEXT: ubfx w0, w8, #7, #1
; CHECK-NEXT: ret
%t0 = lshr i8 128, %y
%t1 = and i8 %t0, %x
%res = icmp ne i8 %t1, 0 ; we are perfectly happy with 'ne' predicate
ret i1 %res
}
;------------------------------------------------------------------------------;
; What if X is a constant too?
;------------------------------------------------------------------------------;
define i1 @scalar_i32_x_is_const_eq(i32 %y) nounwind {
; CHECK-LABEL: scalar_i32_x_is_const_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: mov w8, #43605
; CHECK-NEXT: movk w8, #43605, lsl #16
; CHECK-NEXT: lsr w8, w8, w0
; CHECK-NEXT: tst w8, #0x1
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i32 2857740885, %y
%t1 = and i32 %t0, 1
%res = icmp eq i32 %t1, 0
ret i1 %res
}
define i1 @scalar_i32_x_is_const2_eq(i32 %y) nounwind {
; CHECK-LABEL: scalar_i32_x_is_const2_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: mov w8, #1
; CHECK-NEXT: mov w9, #43605
; CHECK-NEXT: lsr w8, w8, w0
; CHECK-NEXT: movk w9, #43605, lsl #16
; CHECK-NEXT: tst w8, w9
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i32 1, %y
%t1 = and i32 %t0, 2857740885
%res = icmp eq i32 %t1, 0
ret i1 %res
}
;------------------------------------------------------------------------------;
; A few negative tests
;------------------------------------------------------------------------------;
define i1 @negative_scalar_i8_bitsinmiddle_slt(i8 %x, i8 %y) nounwind {
; CHECK-LABEL: negative_scalar_i8_bitsinmiddle_slt:
; CHECK: // %bb.0:
; CHECK-NEXT: mov w8, #24
; CHECK-NEXT: // kill: def $w1 killed $w1 def $x1
; CHECK-NEXT: lsr w8, w8, w1
; CHECK-NEXT: and w8, w8, w0
; CHECK-NEXT: sxtb w8, w8
; CHECK-NEXT: cmp w8, #0 // =0
; CHECK-NEXT: cset w0, lt
; CHECK-NEXT: ret
%t0 = lshr i8 24, %y
%t1 = and i8 %t0, %x
%res = icmp slt i8 %t1, 0
ret i1 %res
}
define i1 @scalar_i8_signbit_eq_with_nonzero(i8 %x, i8 %y) nounwind {
; CHECK-LABEL: scalar_i8_signbit_eq_with_nonzero:
; CHECK: // %bb.0:
; CHECK-NEXT: mov w8, #128
; CHECK-NEXT: // kill: def $w1 killed $w1 def $x1
; CHECK-NEXT: lsr w8, w8, w1
; CHECK-NEXT: and w8, w8, w0
; CHECK-NEXT: and w8, w8, #0xff
; CHECK-NEXT: cmp w8, #1 // =1
; CHECK-NEXT: cset w0, eq
; CHECK-NEXT: ret
%t0 = lshr i8 128, %y
%t1 = and i8 %t0, %x
%res = icmp eq i8 %t1, 1 ; should be comparing with 0
ret i1 %res
}
Reference in New Issue View Git Blame Copy Permalink