diff options
author | Tim Northover <tnorthover@apple.com> | 2014-04-04 14:49:21 +0000 |
---|---|---|
committer | Tim Northover <tnorthover@apple.com> | 2014-04-04 14:49:21 +0000 |
commit | 4a4d62bfb923f81d0cc8203bdca487a764bb0945 (patch) | |
tree | 0b01c2b83297b39f2289aa8da591138f30412acb /test/CodeGen/ARM64 | |
parent | 148692bafe85b8b9b6c96917d1aeafe2862bfe56 (diff) | |
download | llvm-4a4d62bfb923f81d0cc8203bdca487a764bb0945.tar.gz llvm-4a4d62bfb923f81d0cc8203bdca487a764bb0945.tar.bz2 llvm-4a4d62bfb923f81d0cc8203bdca487a764bb0945.tar.xz |
ARM64: handle v1i1 types arising from setcc properly.
There were several overlapping problems here, and this solution is
closely inspired by the one adopted in AArch64 in r201381.
Firstly, scalarisation of v1i1 setcc operations simply fails if the
input types are legal. This is fixed in LegalizeVectorTypes.cpp this
time, and allows AArch64 code to be simplified slightly.
Second, vselect with such a setcc feeding into it ends up in
ScalarizeVectorOperand, where it's not handled. I experimented with an
implementation, but found that whatever DAG came out was rather
horrific. I think Hao's DAG combine approach is a good one for
quality, though there are edge cases it won't catch (to be fixed
separately).
Should fix PR19335.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@205625 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/CodeGen/ARM64')
-rw-r--r-- | test/CodeGen/ARM64/neon-v1i1-setcc.ll | 65 |
1 files changed, 65 insertions, 0 deletions
diff --git a/test/CodeGen/ARM64/neon-v1i1-setcc.ll b/test/CodeGen/ARM64/neon-v1i1-setcc.ll new file mode 100644 index 0000000000..a7e59fbc00 --- /dev/null +++ b/test/CodeGen/ARM64/neon-v1i1-setcc.ll @@ -0,0 +1,65 @@ +; RUN: llc %s -o - -verify-machineinstrs -mtriple=arm64-none-linux-gnu | FileCheck %s + +; This is the analogue of AArch64's file of the same name. It's mostly testing +; some form of correct lowering occurs, the tests are a little artificial but I +; strongly suspect there's room for improved CodeGen (FIXME). + +define i64 @test_sext_extr_cmp_0(<1 x i64> %v1, <1 x i64> %v2) { +; CHECK-LABEL: test_sext_extr_cmp_0: +; CHECK: cmp {{x[0-9]+}}, {{x[0-9]+}} +; CHECK: csinc + %1 = icmp sge <1 x i64> %v1, %v2 + %2 = extractelement <1 x i1> %1, i32 0 + %vget_lane = sext i1 %2 to i64 + ret i64 %vget_lane +} + +define i64 @test_sext_extr_cmp_1(<1 x double> %v1, <1 x double> %v2) { +; CHECK-LABEL: test_sext_extr_cmp_1: +; CHECK: fcmp {{d[0-9]+}}, {{d[0-9]+}} + %1 = fcmp oeq <1 x double> %v1, %v2 + %2 = extractelement <1 x i1> %1, i32 0 + %vget_lane = sext i1 %2 to i64 + ret i64 %vget_lane +} + +define <1 x i64> @test_select_v1i1_0(<1 x i64> %v1, <1 x i64> %v2, <1 x i64> %v3) { +; CHECK-LABEL: test_select_v1i1_0: +; CHECK: cmeq d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}} +; CHECK: bic v{{[0-9]+}}.8b, v{{[0-9]+}}.8b, v{{[0-9]+}}.8b + %1 = icmp eq <1 x i64> %v1, %v2 + %res = select <1 x i1> %1, <1 x i64> zeroinitializer, <1 x i64> %v3 + ret <1 x i64> %res +} + +define <1 x i64> @test_select_v1i1_1(<1 x double> %v1, <1 x double> %v2, <1 x i64> %v3) { +; CHECK-LABEL: test_select_v1i1_1: +; CHECK: fcmeq d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}} +; CHECK: bic v{{[0-9]+}}.8b, v{{[0-9]+}}.8b, v{{[0-9]+}}.8b + %1 = fcmp oeq <1 x double> %v1, %v2 + %res = select <1 x i1> %1, <1 x i64> zeroinitializer, <1 x i64> %v3 + ret <1 x i64> %res +} + +define <1 x double> @test_select_v1i1_2(<1 x i64> %v1, <1 x i64> %v2, <1 x double> %v3) { +; CHECK-LABEL: test_select_v1i1_2: +; CHECK: cmeq d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}} +; CHECK: bic v{{[0-9]+}}.8b, v{{[0-9]+}}.8b, v{{[0-9]+}}.8b + %1 = icmp eq <1 x i64> %v1, %v2 + %res = select <1 x i1> %1, <1 x double> zeroinitializer, <1 x double> %v3 + ret <1 x double> %res +} + +define i32 @test_br_extr_cmp(<1 x i64> %v1, <1 x i64> %v2) { +; CHECK-LABEL: test_br_extr_cmp: +; CHECK: cmp x{{[0-9]+}}, x{{[0-9]+}} + %1 = icmp eq <1 x i64> %v1, %v2 + %2 = extractelement <1 x i1> %1, i32 0 + br i1 %2, label %if.end, label %if.then + +if.then: + ret i32 0; + +if.end: + ret i32 1; +} |