[AVX512] Adding support for load truncate store of I1

store operation on a truncated memory (load) of vXi1 is poorly supported by LLVM and most of the time end with an assertion.
This patch fixes this issue.

Differential Revision: https://reviews.llvm.org/D39547

Change-Id: Ida5523dd09c1ad384acc0a27e9e59273d28cbdc9
llvm-svn: 320691

2 files changed, 154 insertions, 0 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 648e0d9deee..58bcc031ce3 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -1222,6 +1222,9 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
       }
     }
 
+    if (Subtarget.hasAVX512())
+      setOperationAction(ISD::SELECT, MVT::v1i1, Custom);
+
     if (Subtarget.hasDQI()) {
       for (auto VT : { MVT::v2i64, MVT::v4i64, MVT::v8i64 }) {
         setOperationAction(ISD::SINT_TO_FP,     VT, Legal);
diff --git a/llvm/test/CodeGen/X86/avx512-load-trunc-store-i1.ll b/llvm/test/CodeGen/X86/avx512-load-trunc-store-i1.ll
new file mode 100644
index 00000000000..bfcac893512
--- /dev/null
+++ b/llvm/test/CodeGen/X86/avx512-load-trunc-store-i1.ll
@@ -0,0 +1,151 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl,+avx512dq -O2 | FileCheck %s --check-prefix=AVX512-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f -O2| FileCheck %s --check-prefix=AVX512-ONLY
+
+define void @load_v1i2_trunc_v1i1_store(<1 x i2>* %a0,<1 x i1>* %a1) {
+; AVX512-ALL-LABEL: load_v1i2_trunc_v1i1_store:
+; AVX512-ALL:       # %bb.0:
+; AVX512-ALL-NEXT:    movb (%rdi), %al
+; AVX512-ALL-NEXT:    testb %al, %al
+; AVX512-ALL-NEXT:    setne %al
+; AVX512-ALL-NEXT:    kmovd %eax, %k0
+; AVX512-ALL-NEXT:    kmovb %k0, (%rsi)
+; AVX512-ALL-NEXT:    retq
+;
+; AVX512-ONLY-LABEL: load_v1i2_trunc_v1i1_store:
+; AVX512-ONLY:       # %bb.0:
+; AVX512-ONLY-NEXT:    movb (%rdi), %al
+; AVX512-ONLY-NEXT:    testb %al, %al
+; AVX512-ONLY-NEXT:    setne %al
+; AVX512-ONLY-NEXT:    movb %al, (%rsi)
+; AVX512-ONLY-NEXT:    retq
+    %d0 = load <1 x i2>, <1 x i2>* %a0
+    %d1 = trunc <1 x i2> %d0 to <1 x i1>
+    store <1 x i1> %d1, <1 x i1>* %a1
+    ret void
+}
+define void @load_v1i3_trunc_v1i1_store(<1 x i3>* %a0,<1 x i1>* %a1) {
+; AVX512-ALL-LABEL: load_v1i3_trunc_v1i1_store:
+; AVX512-ALL:       # %bb.0:
+; AVX512-ALL-NEXT:    movb (%rdi), %al
+; AVX512-ALL-NEXT:    testb %al, %al
+; AVX512-ALL-NEXT:    setne %al
+; AVX512-ALL-NEXT:    kmovd %eax, %k0
+; AVX512-ALL-NEXT:    kmovb %k0, (%rsi)
+; AVX512-ALL-NEXT:    retq
+;
+; AVX512-ONLY-LABEL: load_v1i3_trunc_v1i1_store:
+; AVX512-ONLY:       # %bb.0:
+; AVX512-ONLY-NEXT:    movb (%rdi), %al
+; AVX512-ONLY-NEXT:    testb %al, %al
+; AVX512-ONLY-NEXT:    setne %al
+; AVX512-ONLY-NEXT:    movb %al, (%rsi)
+; AVX512-ONLY-NEXT:    retq
+    %d0 = load <1 x i3>, <1 x i3>* %a0
+    %d1 = trunc <1 x i3> %d0 to <1 x i1>
+    store <1 x i1> %d1, <1 x i1>* %a1
+    ret void
+}
+define void @load_v1i4_trunc_v1i1_store(<1 x i4>* %a0,<1 x i1>* %a1) {
+; AVX512-ALL-LABEL: load_v1i4_trunc_v1i1_store:
+; AVX512-ALL:       # %bb.0:
+; AVX512-ALL-NEXT:    movb (%rdi), %al
+; AVX512-ALL-NEXT:    testb %al, %al
+; AVX512-ALL-NEXT:    setne %al
+; AVX512-ALL-NEXT:    kmovd %eax, %k0
+; AVX512-ALL-NEXT:    kmovb %k0, (%rsi)
+; AVX512-ALL-NEXT:    retq
+;
+; AVX512-ONLY-LABEL: load_v1i4_trunc_v1i1_store:
+; AVX512-ONLY:       # %bb.0:
+; AVX512-ONLY-NEXT:    movb (%rdi), %al
+; AVX512-ONLY-NEXT:    testb %al, %al
+; AVX512-ONLY-NEXT:    setne %al
+; AVX512-ONLY-NEXT:    movb %al, (%rsi)
+; AVX512-ONLY-NEXT:    retq
+    %d0 = load <1 x i4>, <1 x i4>* %a0
+    %d1 = trunc <1 x i4> %d0 to <1 x i1>
+    store <1 x i1> %d1, <1 x i1>* %a1
+    ret void
+}
+define void @load_v1i8_trunc_v1i1_store(<1 x i8>* %a0,<1 x i1>* %a1) {
+; AVX512-ALL-LABEL: load_v1i8_trunc_v1i1_store:
+; AVX512-ALL:       # %bb.0:
+; AVX512-ALL-NEXT:    cmpb $0, (%rdi)
+; AVX512-ALL-NEXT:    setne %al
+; AVX512-ALL-NEXT:    kmovd %eax, %k0
+; AVX512-ALL-NEXT:    kmovb %k0, (%rsi)
+; AVX512-ALL-NEXT:    retq
+;
+; AVX512-ONLY-LABEL: load_v1i8_trunc_v1i1_store:
+; AVX512-ONLY:       # %bb.0:
+; AVX512-ONLY-NEXT:    cmpb $0, (%rdi)
+; AVX512-ONLY-NEXT:    setne %al
+; AVX512-ONLY-NEXT:    movb %al, (%rsi)
+; AVX512-ONLY-NEXT:    retq
+    %d0 = load <1 x i8>, <1 x i8>* %a0
+    %d1 = trunc <1 x i8> %d0 to <1 x i1>
+    store <1 x i1> %d1, <1 x i1>* %a1
+    ret void
+}
+define void @load_v1i16_trunc_v1i1_store(<1 x i16>* %a0,<1 x i1>* %a1) {
+; AVX512-ALL-LABEL: load_v1i16_trunc_v1i1_store:
+; AVX512-ALL:       # %bb.0:
+; AVX512-ALL-NEXT:    cmpb $0, (%rdi)
+; AVX512-ALL-NEXT:    setne %al
+; AVX512-ALL-NEXT:    kmovd %eax, %k0
+; AVX512-ALL-NEXT:    kmovb %k0, (%rsi)
+; AVX512-ALL-NEXT:    retq
+;
+; AVX512-ONLY-LABEL: load_v1i16_trunc_v1i1_store:
+; AVX512-ONLY:       # %bb.0:
+; AVX512-ONLY-NEXT:    cmpb $0, (%rdi)
+; AVX512-ONLY-NEXT:    setne %al
+; AVX512-ONLY-NEXT:    movb %al, (%rsi)
+; AVX512-ONLY-NEXT:    retq
+    %d0 = load <1 x i16>, <1 x i16>* %a0
+    %d1 = trunc <1 x i16> %d0 to <1 x i1>
+    store <1 x i1> %d1, <1 x i1>* %a1
+    ret void
+}
+define void @load_v1i32_trunc_v1i1_store(<1 x i32>* %a0,<1 x i1>* %a1) {
+; AVX512-ALL-LABEL: load_v1i32_trunc_v1i1_store:
+; AVX512-ALL:       # %bb.0:
+; AVX512-ALL-NEXT:    cmpb $0, (%rdi)
+; AVX512-ALL-NEXT:    setne %al
+; AVX512-ALL-NEXT:    kmovd %eax, %k0
+; AVX512-ALL-NEXT:    kmovb %k0, (%rsi)
+; AVX512-ALL-NEXT:    retq
+;
+; AVX512-ONLY-LABEL: load_v1i32_trunc_v1i1_store:
+; AVX512-ONLY:       # %bb.0:
+; AVX512-ONLY-NEXT:    cmpb $0, (%rdi)
+; AVX512-ONLY-NEXT:    setne %al
+; AVX512-ONLY-NEXT:    movb %al, (%rsi)
+; AVX512-ONLY-NEXT:    retq
+    %d0 = load <1 x i32>, <1 x i32>* %a0
+    %d1 = trunc <1 x i32> %d0 to <1 x i1>
+    store <1 x i1> %d1, <1 x i1>* %a1
+    ret void
+}
+define void @load_v1i64_trunc_v1i1_store(<1 x i64>* %a0,<1 x i1>* %a1) {
+; AVX512-ALL-LABEL: load_v1i64_trunc_v1i1_store:
+; AVX512-ALL:       # %bb.0:
+; AVX512-ALL-NEXT:    cmpb $0, (%rdi)
+; AVX512-ALL-NEXT:    setne %al
+; AVX512-ALL-NEXT:    kmovd %eax, %k0
+; AVX512-ALL-NEXT:    kmovb %k0, (%rsi)
+; AVX512-ALL-NEXT:    retq
+;
+; AVX512-ONLY-LABEL: load_v1i64_trunc_v1i1_store:
+; AVX512-ONLY:       # %bb.0:
+; AVX512-ONLY-NEXT:    cmpb $0, (%rdi)
+; AVX512-ONLY-NEXT:    setne %al
+; AVX512-ONLY-NEXT:    movb %al, (%rsi)
+; AVX512-ONLY-NEXT:    retq
+    %d0 = load <1 x i64>, <1 x i64>* %a0
+    %d1 = trunc <1 x i64> %d0 to <1 x i1>
+    store <1 x i1> %d1, <1 x i1>* %a1
+    ret void
+}
+