maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AArch64/srem-seteq-vec-splat.ll

227 lines
8.2 KiB
LLVM
Raw Normal View History

[NFC][Codegen][X86][AArch64] Add "(x s% C) == 0" tests Much like with `urem`, the same optimization (albeit with slightly different algorithm) applies for the signed case, too. I'm simply copying the test coverage from `urem` case for now, i believe it should be (close to?) sufficient. llvm-svn: 366640
2019-07-21 03:25:44 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=aarch64-unknown-linux-gnu < %s | FileCheck %s
; Odd divisor
define <4 x i32> @test_srem_odd_25(<4 x i32> %X) nounwind {
; CHECK-LABEL: test_srem_odd_25:
; CHECK: // %bb.0:
[CodeGen][SelectionDAG] More efficient code for X % C == 0 (SREM case) Summary: This implements an optimization described in Hacker's Delight 10-17: when `C` is constant, the result of `X % C == 0` can be computed more cheaply without actually calculating the remainder. The motivation is discussed here: https://bugs.llvm.org/show_bug.cgi?id=35479. One huge caveat: this signed case is only valid for positive divisors. While we can freely negate negative divisors, we can't negate `INT_MIN`, so for now if `INT_MIN` is encountered, we bailout. As a follow-up, it should be possible to handle that more gracefully via extra `and`+`setcc`+`select`. This passes llvm's test-suite, and from cursory(!) cross-examination the folds (the assembly) match those of GCC, and manual checking via alive did not reveal any issues (other than the `INT_MIN` case) Reviewers: RKSimon, spatel, hermord, craig.topper, xbolva00 Reviewed By: RKSimon, xbolva00 Subscribers: xbolva00, thakis, javed.absar, hiraditya, dexonsmith, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65366 llvm-svn: 368702
2019-08-13 22:57:37 +08:00
; CHECK-NEXT: mov w8, #23593
; CHECK-NEXT: mov w9, #47185
; CHECK-NEXT: movk w8, #49807, lsl #16
; CHECK-NEXT: movk w9, #1310, lsl #16
; CHECK-NEXT: mov w10, #28834
; CHECK-NEXT: movk w10, #2621, lsl #16
; CHECK-NEXT: dup v1.4s, w8
; CHECK-NEXT: dup v2.4s, w9
; CHECK-NEXT: dup v3.4s, w10
; CHECK-NEXT: mla v2.4s, v0.4s, v1.4s
; CHECK-NEXT: cmhs v0.4s, v3.4s, v2.4s
[NFC][Codegen][X86][AArch64] Add "(x s% C) == 0" tests Much like with `urem`, the same optimization (albeit with slightly different algorithm) applies for the signed case, too. I'm simply copying the test coverage from `urem` case for now, i believe it should be (close to?) sufficient. llvm-svn: 366640
2019-07-21 03:25:44 +08:00
; CHECK-NEXT: movi v1.4s, #1
; CHECK-NEXT: and v0.16b, v0.16b, v1.16b
; CHECK-NEXT: ret
%srem = srem <4 x i32> %X, <i32 25, i32 25, i32 25, i32 25>
%cmp = icmp eq <4 x i32> %srem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; Even divisors
define <4 x i32> @test_srem_even_100(<4 x i32> %X) nounwind {
; CHECK-LABEL: test_srem_even_100:
; CHECK: // %bb.0:
; CHECK-NEXT: mov w8, #34079
; CHECK-NEXT: movk w8, #20971, lsl #16
; CHECK-NEXT: dup v2.4s, w8
; CHECK-NEXT: smull2 v3.2d, v0.4s, v2.4s
; CHECK-NEXT: smull v2.2d, v0.2s, v2.2s
; CHECK-NEXT: uzp2 v2.4s, v2.4s, v3.4s
; CHECK-NEXT: sshr v3.4s, v2.4s, #5
; CHECK-NEXT: movi v1.4s, #100
; CHECK-NEXT: usra v3.4s, v2.4s, #31
; CHECK-NEXT: mls v0.4s, v3.4s, v1.4s
; CHECK-NEXT: cmeq v0.4s, v0.4s, #0
; CHECK-NEXT: movi v1.4s, #1
; CHECK-NEXT: and v0.16b, v0.16b, v1.16b
; CHECK-NEXT: ret
%srem = srem <4 x i32> %X, <i32 100, i32 100, i32 100, i32 100>
%cmp = icmp eq <4 x i32> %srem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
[NFC][X86][AArch64] Revisit test coverage for X s% C == 0 fold - add tests for negative divisors, INT_MIN divisors As discussed in the review, that fold is only valid for positive divisors, so while we can negate negative divisors, we have to special-case INT_MIN. llvm-svn: 367294
2019-07-30 16:00:49 +08:00
; Negative divisors should be negated, and thus this is still splat vectors.
; Odd divisor
define <4 x i32> @test_srem_odd_neg25(<4 x i32> %X) nounwind {
; CHECK-LABEL: test_srem_odd_neg25:
; CHECK: // %bb.0:
[CodeGen][SelectionDAG] More efficient code for X % C == 0 (SREM case) Summary: This implements an optimization described in Hacker's Delight 10-17: when `C` is constant, the result of `X % C == 0` can be computed more cheaply without actually calculating the remainder. The motivation is discussed here: https://bugs.llvm.org/show_bug.cgi?id=35479. One huge caveat: this signed case is only valid for positive divisors. While we can freely negate negative divisors, we can't negate `INT_MIN`, so for now if `INT_MIN` is encountered, we bailout. As a follow-up, it should be possible to handle that more gracefully via extra `and`+`setcc`+`select`. This passes llvm's test-suite, and from cursory(!) cross-examination the folds (the assembly) match those of GCC, and manual checking via alive did not reveal any issues (other than the `INT_MIN` case) Reviewers: RKSimon, spatel, hermord, craig.topper, xbolva00 Reviewed By: RKSimon, xbolva00 Subscribers: xbolva00, thakis, javed.absar, hiraditya, dexonsmith, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65366 llvm-svn: 368702
2019-08-13 22:57:37 +08:00
; CHECK-NEXT: mov w8, #23593
; CHECK-NEXT: mov w9, #47185
; CHECK-NEXT: movk w8, #49807, lsl #16
; CHECK-NEXT: movk w9, #1310, lsl #16
; CHECK-NEXT: mov w10, #28834
; CHECK-NEXT: movk w10, #2621, lsl #16
; CHECK-NEXT: dup v1.4s, w8
; CHECK-NEXT: dup v2.4s, w9
; CHECK-NEXT: dup v3.4s, w10
; CHECK-NEXT: mla v2.4s, v0.4s, v1.4s
; CHECK-NEXT: cmhs v0.4s, v3.4s, v2.4s
[NFC][X86][AArch64] Revisit test coverage for X s% C == 0 fold - add tests for negative divisors, INT_MIN divisors As discussed in the review, that fold is only valid for positive divisors, so while we can negate negative divisors, we have to special-case INT_MIN. llvm-svn: 367294
2019-07-30 16:00:49 +08:00
; CHECK-NEXT: movi v1.4s, #1
; CHECK-NEXT: and v0.16b, v0.16b, v1.16b
; CHECK-NEXT: ret
%srem = srem <4 x i32> %X, <i32 25, i32 -25, i32 -25, i32 25>
%cmp = icmp eq <4 x i32> %srem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; Even divisors
define <4 x i32> @test_srem_even_neg100(<4 x i32> %X) nounwind {
; CHECK-LABEL: test_srem_even_neg100:
; CHECK: // %bb.0:
; CHECK-NEXT: adrp x8, .LCPI3_0
; CHECK-NEXT: ldr q1, [x8, :lo12:.LCPI3_0]
; CHECK-NEXT: adrp x8, .LCPI3_1
; CHECK-NEXT: ldr q2, [x8, :lo12:.LCPI3_1]
; CHECK-NEXT: smull2 v3.2d, v0.4s, v1.4s
; CHECK-NEXT: smull v1.2d, v0.2s, v1.2s
; CHECK-NEXT: uzp2 v1.4s, v1.4s, v3.4s
; CHECK-NEXT: sshr v3.4s, v1.4s, #5
; CHECK-NEXT: usra v3.4s, v1.4s, #31
; CHECK-NEXT: mls v0.4s, v3.4s, v2.4s
; CHECK-NEXT: cmeq v0.4s, v0.4s, #0
; CHECK-NEXT: movi v1.4s, #1
; CHECK-NEXT: and v0.16b, v0.16b, v1.16b
; CHECK-NEXT: ret
%srem = srem <4 x i32> %X, <i32 -100, i32 100, i32 -100, i32 100>
%cmp = icmp eq <4 x i32> %srem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
[NFC][Codegen][X86][AArch64] Add "(x s% C) == 0" tests Much like with `urem`, the same optimization (albeit with slightly different algorithm) applies for the signed case, too. I'm simply copying the test coverage from `urem` case for now, i believe it should be (close to?) sufficient. llvm-svn: 366640
2019-07-21 03:25:44 +08:00
;------------------------------------------------------------------------------;
; Comparison constant has undef elements.
;------------------------------------------------------------------------------;
define <4 x i32> @test_srem_odd_undef1(<4 x i32> %X) nounwind {
; CHECK-LABEL: test_srem_odd_undef1:
; CHECK: // %bb.0:
; CHECK-NEXT: mov w8, #34079
; CHECK-NEXT: movk w8, #20971, lsl #16
; CHECK-NEXT: dup v2.4s, w8
; CHECK-NEXT: smull2 v3.2d, v0.4s, v2.4s
; CHECK-NEXT: smull v2.2d, v0.2s, v2.2s
; CHECK-NEXT: uzp2 v2.4s, v2.4s, v3.4s
; CHECK-NEXT: sshr v3.4s, v2.4s, #3
; CHECK-NEXT: movi v1.4s, #25
; CHECK-NEXT: usra v3.4s, v2.4s, #31
; CHECK-NEXT: mls v0.4s, v3.4s, v1.4s
; CHECK-NEXT: cmeq v0.4s, v0.4s, #0
; CHECK-NEXT: movi v1.4s, #1
; CHECK-NEXT: and v0.16b, v0.16b, v1.16b
; CHECK-NEXT: ret
%srem = srem <4 x i32> %X, <i32 25, i32 25, i32 25, i32 25>
%cmp = icmp eq <4 x i32> %srem, <i32 0, i32 0, i32 undef, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
define <4 x i32> @test_srem_even_undef1(<4 x i32> %X) nounwind {
; CHECK-LABEL: test_srem_even_undef1:
; CHECK: // %bb.0:
; CHECK-NEXT: mov w8, #34079
; CHECK-NEXT: movk w8, #20971, lsl #16
; CHECK-NEXT: dup v2.4s, w8
; CHECK-NEXT: smull2 v3.2d, v0.4s, v2.4s
; CHECK-NEXT: smull v2.2d, v0.2s, v2.2s
; CHECK-NEXT: uzp2 v2.4s, v2.4s, v3.4s
; CHECK-NEXT: sshr v3.4s, v2.4s, #5
; CHECK-NEXT: movi v1.4s, #100
; CHECK-NEXT: usra v3.4s, v2.4s, #31
; CHECK-NEXT: mls v0.4s, v3.4s, v1.4s
; CHECK-NEXT: cmeq v0.4s, v0.4s, #0
; CHECK-NEXT: movi v1.4s, #1
; CHECK-NEXT: and v0.16b, v0.16b, v1.16b
; CHECK-NEXT: ret
%srem = srem <4 x i32> %X, <i32 100, i32 100, i32 100, i32 100>
%cmp = icmp eq <4 x i32> %srem, <i32 0, i32 0, i32 undef, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
;------------------------------------------------------------------------------;
; Negative tests
;------------------------------------------------------------------------------;
[NFC][X86][AArch64] Revisit test coverage for X s% C == 0 fold - add tests for negative divisors, INT_MIN divisors As discussed in the review, that fold is only valid for positive divisors, so while we can negate negative divisors, we have to special-case INT_MIN. llvm-svn: 367294
2019-07-30 16:00:49 +08:00
define <4 x i32> @test_srem_one_eq(<4 x i32> %X) nounwind {
; CHECK-LABEL: test_srem_one_eq:
; CHECK: // %bb.0:
; CHECK-NEXT: movi v0.4s, #1
; CHECK-NEXT: ret
%srem = srem <4 x i32> %X, <i32 1, i32 1, i32 1, i32 1>
%cmp = icmp eq <4 x i32> %srem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
define <4 x i32> @test_srem_one_ne(<4 x i32> %X) nounwind {
; CHECK-LABEL: test_srem_one_ne:
; CHECK: // %bb.0:
; CHECK-NEXT: movi v0.2d, #0000000000000000
; CHECK-NEXT: ret
%srem = srem <4 x i32> %X, <i32 1, i32 1, i32 1, i32 1>
%cmp = icmp ne <4 x i32> %srem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
[NFC][Codegen][X86][AArch64] Add "(x s% C) == 0" tests Much like with `urem`, the same optimization (albeit with slightly different algorithm) applies for the signed case, too. I'm simply copying the test coverage from `urem` case for now, i believe it should be (close to?) sufficient. llvm-svn: 366640
2019-07-21 03:25:44 +08:00
; We can lower remainder of division by powers of two much better elsewhere.
define <4 x i32> @test_srem_pow2(<4 x i32> %X) nounwind {
; CHECK-LABEL: test_srem_pow2:
; CHECK: // %bb.0:
; CHECK-NEXT: sshr v1.4s, v0.4s, #31
; CHECK-NEXT: mov v2.16b, v0.16b
; CHECK-NEXT: usra v2.4s, v1.4s, #28
; CHECK-NEXT: bic v2.4s, #15
; CHECK-NEXT: sub v0.4s, v0.4s, v2.4s
; CHECK-NEXT: cmeq v0.4s, v0.4s, #0
; CHECK-NEXT: movi v1.4s, #1
; CHECK-NEXT: and v0.16b, v0.16b, v1.16b
; CHECK-NEXT: ret
%srem = srem <4 x i32> %X, <i32 16, i32 16, i32 16, i32 16>
%cmp = icmp eq <4 x i32> %srem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
[NFC][X86][AArch64] Revisit test coverage for X s% C == 0 fold - add tests for negative divisors, INT_MIN divisors As discussed in the review, that fold is only valid for positive divisors, so while we can negate negative divisors, we have to special-case INT_MIN. llvm-svn: 367294
2019-07-30 16:00:49 +08:00
; We could lower remainder of division by INT_MIN much better elsewhere.
define <4 x i32> @test_srem_int_min(<4 x i32> %X) nounwind {
; CHECK-LABEL: test_srem_int_min:
[NFC][Codegen][X86][AArch64] Add "(x s% C) == 0" tests Much like with `urem`, the same optimization (albeit with slightly different algorithm) applies for the signed case, too. I'm simply copying the test coverage from `urem` case for now, i believe it should be (close to?) sufficient. llvm-svn: 366640
2019-07-21 03:25:44 +08:00
; CHECK: // %bb.0:
[NFC][X86][AArch64] Revisit test coverage for X s% C == 0 fold - add tests for negative divisors, INT_MIN divisors As discussed in the review, that fold is only valid for positive divisors, so while we can negate negative divisors, we have to special-case INT_MIN. llvm-svn: 367294
2019-07-30 16:00:49 +08:00
; CHECK-NEXT: sshr v1.4s, v0.4s, #31
; CHECK-NEXT: mov v2.16b, v0.16b
; CHECK-NEXT: movi v3.4s, #128, lsl #24
; CHECK-NEXT: usra v2.4s, v1.4s, #1
; CHECK-NEXT: and v1.16b, v2.16b, v3.16b
; CHECK-NEXT: sub v0.4s, v0.4s, v1.4s
; CHECK-NEXT: cmeq v0.4s, v0.4s, #0
; CHECK-NEXT: movi v1.4s, #1
; CHECK-NEXT: and v0.16b, v0.16b, v1.16b
[NFC][Codegen][X86][AArch64] Add "(x s% C) == 0" tests Much like with `urem`, the same optimization (albeit with slightly different algorithm) applies for the signed case, too. I'm simply copying the test coverage from `urem` case for now, i believe it should be (close to?) sufficient. llvm-svn: 366640
2019-07-21 03:25:44 +08:00
; CHECK-NEXT: ret
[NFC][X86][AArch64] Revisit test coverage for X s% C == 0 fold - add tests for negative divisors, INT_MIN divisors As discussed in the review, that fold is only valid for positive divisors, so while we can negate negative divisors, we have to special-case INT_MIN. llvm-svn: 367294
2019-07-30 16:00:49 +08:00
%srem = srem <4 x i32> %X, <i32 2147483648, i32 2147483648, i32 2147483648, i32 2147483648>
[NFC][Codegen][X86][AArch64] Add "(x s% C) == 0" tests Much like with `urem`, the same optimization (albeit with slightly different algorithm) applies for the signed case, too. I'm simply copying the test coverage from `urem` case for now, i believe it should be (close to?) sufficient. llvm-svn: 366640
2019-07-21 03:25:44 +08:00
%cmp = icmp eq <4 x i32> %srem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
[NFC][X86][AArch64] Revisit test coverage for X s% C == 0 fold - add tests for negative divisors, INT_MIN divisors As discussed in the review, that fold is only valid for positive divisors, so while we can negate negative divisors, we have to special-case INT_MIN. llvm-svn: 367294
2019-07-30 16:00:49 +08:00
; We could lower remainder of division by all-ones much better elsewhere.
define <4 x i32> @test_srem_allones(<4 x i32> %X) nounwind {
; CHECK-LABEL: test_srem_allones:
[NFC][Codegen][X86][AArch64] Add "(x s% C) == 0" tests Much like with `urem`, the same optimization (albeit with slightly different algorithm) applies for the signed case, too. I'm simply copying the test coverage from `urem` case for now, i believe it should be (close to?) sufficient. llvm-svn: 366640
2019-07-21 03:25:44 +08:00
; CHECK: // %bb.0:
; CHECK-NEXT: movi v0.4s, #1
; CHECK-NEXT: ret
[NFC][X86][AArch64] Revisit test coverage for X s% C == 0 fold - add tests for negative divisors, INT_MIN divisors As discussed in the review, that fold is only valid for positive divisors, so while we can negate negative divisors, we have to special-case INT_MIN. llvm-svn: 367294
2019-07-30 16:00:49 +08:00
%srem = srem <4 x i32> %X, <i32 4294967295, i32 4294967295, i32 4294967295, i32 4294967295>
[NFC][Codegen][X86][AArch64] Add "(x s% C) == 0" tests Much like with `urem`, the same optimization (albeit with slightly different algorithm) applies for the signed case, too. I'm simply copying the test coverage from `urem` case for now, i believe it should be (close to?) sufficient. llvm-svn: 366640
2019-07-21 03:25:44 +08:00
%cmp = icmp eq <4 x i32> %srem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}