2019-11-10 19:21:50 +08:00
|
|
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
|
|
|
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+sse2 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-SSE,CHECK-SSE2
|
|
|
|
|
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+sse4.1 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-SSE,CHECK-SSE41
|
|
|
|
|
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx < %s | FileCheck %s --check-prefixes=CHECK,CHECK-AVX,CHECK-AVX1
|
|
|
|
|
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-AVX,CHECK-AVX2
|
|
|
|
|
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512vl < %s | FileCheck %s --check-prefixes=CHECK,CHECK-AVX,CHECK-AVX512VL
|
|
|
|
|
|
|
|
|
|
define <4 x i1> @t0_all_tautological(<4 x i32> %X) nounwind {
|
|
|
|
|
; CHECK-SSE-LABEL: t0_all_tautological:
|
|
|
|
|
; CHECK-SSE: # %bb.0:
|
|
|
|
|
; CHECK-SSE-NEXT: pand {{.*}}(%rip), %xmm0
|
|
|
|
|
; CHECK-SSE-NEXT: pcmpeqd {{.*}}(%rip), %xmm0
|
|
|
|
|
; CHECK-SSE-NEXT: retq
|
|
|
|
|
;
|
|
|
|
|
; CHECK-AVX-LABEL: t0_all_tautological:
|
|
|
|
|
; CHECK-AVX: # %bb.0:
|
|
|
|
|
; CHECK-AVX-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX-NEXT: vpcmpeqd {{.*}}(%rip), %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX-NEXT: retq
|
|
|
|
|
%urem = urem <4 x i32> %X, <i32 1, i32 1, i32 2, i32 2>
|
|
|
|
|
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 1, i32 2, i32 3>
|
|
|
|
|
ret <4 x i1> %cmp
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define <4 x i1> @t1_all_odd_eq(<4 x i32> %X) nounwind {
|
|
|
|
|
; CHECK-SSE2-LABEL: t1_all_odd_eq:
|
|
|
|
|
; CHECK-SSE2: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [2863311531,2863311531,2863311531,2863311531]
|
|
|
|
|
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
|
|
|
|
|
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm0
|
|
|
|
|
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
|
|
|
|
|
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
|
|
|
|
|
; CHECK-SSE2-NEXT: pxor {{.*}}(%rip), %xmm0
|
|
|
|
|
; CHECK-SSE2-NEXT: pcmpgtd {{.*}}(%rip), %xmm0
|
|
|
|
|
; CHECK-SSE2-NEXT: pandn {{.*}}(%rip), %xmm0
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-SSE2-NEXT: retq
|
|
|
|
|
;
|
|
|
|
|
; CHECK-SSE41-LABEL: t1_all_odd_eq:
|
|
|
|
|
; CHECK-SSE41: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm0
|
|
|
|
|
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [1431655765,4294967295,4294967295,4294967295]
|
|
|
|
|
; CHECK-SSE41-NEXT: pminud %xmm0, %xmm1
|
|
|
|
|
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
|
|
|
|
|
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
|
|
|
|
|
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-SSE41-NEXT: retq
|
|
|
|
|
;
|
|
|
|
|
; CHECK-AVX1-LABEL: t1_all_odd_eq:
|
|
|
|
|
; CHECK-AVX1: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX1-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1
|
|
|
|
|
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
|
|
|
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-AVX1-NEXT: retq
|
|
|
|
|
;
|
|
|
|
|
; CHECK-AVX2-LABEL: t1_all_odd_eq:
|
|
|
|
|
; CHECK-AVX2: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [2863311531,2863311531,2863311531,2863311531]
|
|
|
|
|
; CHECK-AVX2-NEXT: vpmulld %xmm1, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX2-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1
|
|
|
|
|
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
|
|
|
; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-AVX2-NEXT: retq
|
|
|
|
|
;
|
|
|
|
|
; CHECK-AVX512VL-LABEL: t1_all_odd_eq:
|
|
|
|
|
; CHECK-AVX512VL: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-AVX512VL-NEXT: retq
|
|
|
|
|
%urem = urem <4 x i32> %X, <i32 3, i32 1, i32 1, i32 9>
|
|
|
|
|
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 42, i32 0, i32 42>
|
|
|
|
|
ret <4 x i1> %cmp
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define <4 x i1> @t1_all_odd_ne(<4 x i32> %X) nounwind {
|
|
|
|
|
; CHECK-SSE2-LABEL: t1_all_odd_ne:
|
|
|
|
|
; CHECK-SSE2: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [2863311531,2863311531,2863311531,2863311531]
|
|
|
|
|
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
|
|
|
|
|
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm0
|
|
|
|
|
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
|
|
|
|
|
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
|
|
|
|
|
; CHECK-SSE2-NEXT: pxor {{.*}}(%rip), %xmm0
|
|
|
|
|
; CHECK-SSE2-NEXT: pcmpgtd {{.*}}(%rip), %xmm0
|
|
|
|
|
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm1
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-SSE2-NEXT: retq
|
|
|
|
|
;
|
|
|
|
|
; CHECK-SSE41-LABEL: t1_all_odd_ne:
|
|
|
|
|
; CHECK-SSE41: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm0
|
|
|
|
|
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [1431655765,4294967295,4294967295,4294967295]
|
|
|
|
|
; CHECK-SSE41-NEXT: pminud %xmm0, %xmm1
|
|
|
|
|
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm1
|
|
|
|
|
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm0
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-SSE41-NEXT: retq
|
|
|
|
|
;
|
|
|
|
|
; CHECK-AVX1-LABEL: t1_all_odd_ne:
|
|
|
|
|
; CHECK-AVX1: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX1-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1
|
|
|
|
|
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
|
|
|
|
|
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-AVX1-NEXT: retq
|
|
|
|
|
;
|
|
|
|
|
; CHECK-AVX2-LABEL: t1_all_odd_ne:
|
|
|
|
|
; CHECK-AVX2: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [2863311531,2863311531,2863311531,2863311531]
|
|
|
|
|
; CHECK-AVX2-NEXT: vpmulld %xmm1, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX2-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1
|
|
|
|
|
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
|
|
|
|
|
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-AVX2-NEXT: retq
|
|
|
|
|
;
|
|
|
|
|
; CHECK-AVX512VL-LABEL: t1_all_odd_ne:
|
|
|
|
|
; CHECK-AVX512VL: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-AVX512VL-NEXT: vpternlogq $15, %xmm0, %xmm0, %xmm0
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-AVX512VL-NEXT: retq
|
|
|
|
|
%urem = urem <4 x i32> %X, <i32 3, i32 1, i32 1, i32 9>
|
|
|
|
|
%cmp = icmp ne <4 x i32> %urem, <i32 0, i32 42, i32 0, i32 42>
|
|
|
|
|
ret <4 x i1> %cmp
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define <8 x i1> @t2_narrow(<8 x i16> %X) nounwind {
|
|
|
|
|
; CHECK-SSE2-LABEL: t2_narrow:
|
|
|
|
|
; CHECK-SSE2: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-SSE2-NEXT: pmullw {{.*}}(%rip), %xmm0
|
|
|
|
|
; CHECK-SSE2-NEXT: psubusw {{.*}}(%rip), %xmm0
|
|
|
|
|
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
|
|
|
|
|
; CHECK-SSE2-NEXT: pcmpeqw %xmm1, %xmm0
|
|
|
|
|
; CHECK-SSE2-NEXT: pand {{.*}}(%rip), %xmm0
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-SSE2-NEXT: retq
|
|
|
|
|
;
|
|
|
|
|
; CHECK-SSE41-LABEL: t2_narrow:
|
|
|
|
|
; CHECK-SSE41: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-SSE41-NEXT: pmullw {{.*}}(%rip), %xmm0
|
|
|
|
|
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [21845,65535,65535,65535,21845,65535,65535,65535]
|
|
|
|
|
; CHECK-SSE41-NEXT: pminuw %xmm0, %xmm1
|
|
|
|
|
; CHECK-SSE41-NEXT: pcmpeqw %xmm1, %xmm0
|
|
|
|
|
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
|
|
|
|
|
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-SSE41-NEXT: retq
|
|
|
|
|
;
|
|
|
|
|
; CHECK-AVX1-LABEL: t2_narrow:
|
|
|
|
|
; CHECK-AVX1: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-AVX1-NEXT: vpmullw {{.*}}(%rip), %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX1-NEXT: vpminuw {{.*}}(%rip), %xmm0, %xmm1
|
|
|
|
|
; CHECK-AVX1-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
|
|
|
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-AVX1-NEXT: retq
|
|
|
|
|
;
|
|
|
|
|
; CHECK-AVX2-LABEL: t2_narrow:
|
|
|
|
|
; CHECK-AVX2: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-AVX2-NEXT: vpmullw {{.*}}(%rip), %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX2-NEXT: vpminuw {{.*}}(%rip), %xmm0, %xmm1
|
|
|
|
|
; CHECK-AVX2-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
|
|
|
; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-AVX2-NEXT: retq
|
|
|
|
|
;
|
|
|
|
|
; CHECK-AVX512VL-LABEL: t2_narrow:
|
|
|
|
|
; CHECK-AVX512VL: # %bb.0:
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-AVX512VL-NEXT: vpmullw {{.*}}(%rip), %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpminuw {{.*}}(%rip), %xmm0, %xmm1
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
|
2019-11-10 19:21:50 +08:00
|
|
|
; CHECK-AVX512VL-NEXT: retq
|
|
|
|
|
%urem = urem <8 x i16> %X, <i16 3, i16 1, i16 1, i16 9, i16 3, i16 1, i16 1, i16 9>
|
|
|
|
|
%cmp = icmp eq <8 x i16> %urem, <i16 0, i16 0, i16 42, i16 42, i16 0, i16 0, i16 42, i16 42>
|
|
|
|
|
ret <8 x i1> %cmp
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define <2 x i1> @t3_wide(<2 x i64> %X) nounwind {
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-SSE-LABEL: t3_wide:
|
|
|
|
|
; CHECK-SSE: # %bb.0:
|
|
|
|
|
; CHECK-SSE-NEXT: movdqa {{.*#+}} xmm1 = [12297829382473034411,12297829382473034411]
|
|
|
|
|
; CHECK-SSE-NEXT: movdqa %xmm0, %xmm2
|
|
|
|
|
; CHECK-SSE-NEXT: pmuludq %xmm1, %xmm2
|
|
|
|
|
; CHECK-SSE-NEXT: movdqa %xmm0, %xmm3
|
|
|
|
|
; CHECK-SSE-NEXT: psrlq $32, %xmm3
|
|
|
|
|
; CHECK-SSE-NEXT: pmuludq %xmm1, %xmm3
|
|
|
|
|
; CHECK-SSE-NEXT: pmuludq {{.*}}(%rip), %xmm0
|
|
|
|
|
; CHECK-SSE-NEXT: paddq %xmm3, %xmm0
|
|
|
|
|
; CHECK-SSE-NEXT: psllq $32, %xmm0
|
|
|
|
|
; CHECK-SSE-NEXT: paddq %xmm2, %xmm0
|
|
|
|
|
; CHECK-SSE-NEXT: pxor {{.*}}(%rip), %xmm0
|
|
|
|
|
; CHECK-SSE-NEXT: movdqa {{.*#+}} xmm1 = [15372286730238776661,9223372034707292159]
|
|
|
|
|
; CHECK-SSE-NEXT: movdqa %xmm0, %xmm2
|
|
|
|
|
; CHECK-SSE-NEXT: pcmpgtd %xmm1, %xmm2
|
|
|
|
|
; CHECK-SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm2[0,0,2,2]
|
|
|
|
|
; CHECK-SSE-NEXT: pcmpeqd %xmm1, %xmm0
|
|
|
|
|
; CHECK-SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
|
|
|
|
|
; CHECK-SSE-NEXT: pand %xmm3, %xmm0
|
|
|
|
|
; CHECK-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
|
|
|
|
|
; CHECK-SSE-NEXT: por %xmm0, %xmm1
|
|
|
|
|
; CHECK-SSE-NEXT: pcmpeqd %xmm0, %xmm0
|
|
|
|
|
; CHECK-SSE-NEXT: pxor %xmm1, %xmm0
|
|
|
|
|
; CHECK-SSE-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero
|
|
|
|
|
; CHECK-SSE-NEXT: retq
|
2019-11-10 19:21:50 +08:00
|
|
|
;
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-AVX1-LABEL: t3_wide:
|
|
|
|
|
; CHECK-AVX1: # %bb.0:
|
|
|
|
|
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm1 = [12297829382473034411,12297829382473034411]
|
|
|
|
|
; CHECK-AVX1-NEXT: vpmuludq %xmm1, %xmm0, %xmm2
|
|
|
|
|
; CHECK-AVX1-NEXT: vpsrlq $32, %xmm0, %xmm3
|
|
|
|
|
; CHECK-AVX1-NEXT: vpmuludq %xmm1, %xmm3, %xmm1
|
|
|
|
|
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX1-NEXT: vpaddq %xmm1, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX1-NEXT: vpsllq $32, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX1-NEXT: vpaddq %xmm0, %xmm2, %xmm0
|
|
|
|
|
; CHECK-AVX1-NEXT: vpxor {{.*}}(%rip), %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX1-NEXT: vpcmpgtq {{.*}}(%rip), %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
|
|
|
|
|
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX1-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
|
|
|
|
|
; CHECK-AVX1-NEXT: retq
|
2019-11-10 19:21:50 +08:00
|
|
|
;
|
[Codegen] TargetLowering::prepareUREMEqFold(): `x u% C1 ==/!= C2` with tautological C1 u<= C2 (PR35479)
Summary:
This is a preparatory cleanup before i add more
of this fold to deal with comparisons with non-zero.
In essence, the current lowering is:
```
Name: (X % C1) == 0 -> X * C3 <= C4
Pre: (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, 0
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%r = icmp ule i8 %n3, %C4
```
https://rise4fun.com/Alive/oqd
It kinda just works, really no weird edge-cases.
But it isn't all that great for when comparing with non-zero.
In particular, given `(X % C1) == C2`, there will be problems
in the always-false tautological case where `C2 u>= C1`:
https://rise4fun.com/Alive/pH3
That case is tautological, always-false:
```
Name: (X % Y) u>= Y
%o0 = urem i8 %x, %y
%r = icmp uge i8 %o0, %y
=>
%r = false
```
https://rise4fun.com/Alive/ofu
While we can't/shouldn't get such tautological case normally,
we do deal with non-splat vectors, so unless we want to give up
in this case, we need to fixup/short-circuit such lanes.
There are two lowering variants:
1. We can blend between whatever computed result and the correct tautological result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%res = icmp ule i8 %n3, %C4
%r = select i1 %is_tautologically_false, i1 0, i1 %res
```
https://rise4fun.com/Alive/PjT5
https://rise4fun.com/Alive/1KV
2. We can invert the comparison result
```
Name: (X % C1) == C2 -> X * C3 <= C4 || false
Pre: (C2 == 0 || C1 u<= C2) && (C1 u>> countTrailingZeros(C1)) * C3 == 1
%zz = and i8 C3, 0 ; trick alive into making C3 avaliable in precondition
%o0 = urem i8 %x, C1
%r = icmp eq i8 %o0, C2
=>
%zz = and i8 C3, 0 ; and silence it from complaining about said reg
%C4 = -1 /u C1
%n0 = mul i8 %x, C3
%n1 = lshr i8 %n0, countTrailingZeros(C1) ; rotate right
%n2 = shl i8 %n0, ((8-countTrailingZeros(C1)) %u 8) ; rotate right
%n3 = or i8 %n1, %n2 ; rotate right
%is_tautologically_false = icmp ule i8 C1, C2
%C4_fixed = select i1 %is_tautologically_false, i8 -1, i8 %C4
%res = icmp ule i8 %n3, %C4_fixed
%r = xor i1 %res, %is_tautologically_false
```
https://rise4fun.com/Alive/2xC
https://rise4fun.com/Alive/jpb5
3. We can expand into `and`/`or`:
https://rise4fun.com/Alive/WGn
https://rise4fun.com/Alive/lcb5
Blend-one is likely better since we avoid having to load the
replacement from constant pool. `xor` is second best since
it's still pretty general. I'm not adding `and`/`or` variants.
Reviewers: RKSimon, craig.topper, spatel
Reviewed By: RKSimon
Subscribers: nick, hiraditya, xbolva00, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70051
2019-11-22 20:16:03 +08:00
|
|
|
; CHECK-AVX2-LABEL: t3_wide:
|
|
|
|
|
; CHECK-AVX2: # %bb.0:
|
|
|
|
|
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [12297829382473034411,12297829382473034411]
|
|
|
|
|
; CHECK-AVX2-NEXT: vpmuludq %xmm1, %xmm0, %xmm2
|
|
|
|
|
; CHECK-AVX2-NEXT: vpsrlq $32, %xmm0, %xmm3
|
|
|
|
|
; CHECK-AVX2-NEXT: vpmuludq %xmm1, %xmm3, %xmm1
|
|
|
|
|
; CHECK-AVX2-NEXT: vpmuludq {{.*}}(%rip), %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX2-NEXT: vpaddq %xmm1, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX2-NEXT: vpsllq $32, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX2-NEXT: vpaddq %xmm0, %xmm2, %xmm0
|
|
|
|
|
; CHECK-AVX2-NEXT: vpxor {{.*}}(%rip), %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX2-NEXT: vpcmpgtq {{.*}}(%rip), %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
|
|
|
|
|
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX2-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
|
|
|
|
|
; CHECK-AVX2-NEXT: retq
|
|
|
|
|
;
|
|
|
|
|
; CHECK-AVX512VL-LABEL: t3_wide:
|
|
|
|
|
; CHECK-AVX512VL: # %bb.0:
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [12297829382473034411,12297829382473034411]
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpmuludq %xmm1, %xmm0, %xmm2
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpsrlq $32, %xmm0, %xmm3
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpmuludq %xmm1, %xmm3, %xmm1
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpmuludq {{.*}}(%rip), %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpaddq %xmm1, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpsllq $32, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpaddq %xmm0, %xmm2, %xmm0
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpminuq {{.*}}(%rip), %xmm0, %xmm1
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
|
|
|
|
|
; CHECK-AVX512VL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
|
|
|
|
|
; CHECK-AVX512VL-NEXT: retq
|
2019-11-10 19:21:50 +08:00
|
|
|
%urem = urem <2 x i64> %X, <i64 3, i64 1>
|
|
|
|
|
%cmp = icmp eq <2 x i64> %urem, <i64 0, i64 42>
|
|
|
|
|
ret <2 x i1> %cmp
|
|
|
|
|
}
|