AVX-512: added calling conventions for i1 vectors.

Fixed bug: https://llvm.org/bugs/show_bug.cgi?id=20724



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@235889 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 71 insertions, 3 deletions

diff --git a/lib/Target/X86/X86CallingConv.td b/lib/Target/X86/X86CallingConv.td
index 41c759a52ee..8ce911079f4 100644
--- a/lib/Target/X86/X86CallingConv.td
+++ b/lib/Target/X86/X86CallingConv.td
@@ -39,6 +39,16 @@ def RetCC_X86Common : CallingConv<[
   CCIfType<[i32], CCAssignToReg<[EAX, EDX, ECX]>>,
   CCIfType<[i64], CCAssignToReg<[RAX, RDX, RCX]>>,
 
+  // Boolean vectors of AVX-512 are returned in SIMD registers.
+  // The call from AVX to AVX-512 function should work,
+  // since the boolean types in AVX/AVX2 are promoted by default.
+  CCIfType<[v2i1],  CCPromoteToType<v2i64>>,
+  CCIfType<[v4i1],  CCPromoteToType<v4i32>>,
+  CCIfType<[v8i1],  CCPromoteToType<v8i16>>,
+  CCIfType<[v16i1], CCPromoteToType<v16i8>>,
+  CCIfType<[v32i1], CCPromoteToType<v32i8>>,
+  CCIfType<[v64i1], CCPromoteToType<v64i8>>,
+
   // Vector types are returned in XMM0 and XMM1, when they fit.  XMM2 and XMM3
   // can only be used by ABI non-compliant code. If the target doesn't have XMM
   // registers, it won't have vector types.
@@ -258,6 +268,16 @@ def CC_X86_64_C : CallingConv<[
             CCIfSubtarget<"hasSSE2()",
             CCPromoteToType<v2i64>>>>,
 
+  // Boolean vectors of AVX-512 are returned in SIMD registers.
+  // The call from AVX to AVX-512 function should work,
+  // since the boolean types in AVX/AVX2 are promoted by default.
+  CCIfType<[v2i1],  CCPromoteToType<v2i64>>,
+  CCIfType<[v4i1],  CCPromoteToType<v4i32>>,
+  CCIfType<[v8i1],  CCPromoteToType<v8i16>>,
+  CCIfType<[v16i1], CCPromoteToType<v16i8>>,
+  CCIfType<[v32i1], CCPromoteToType<v32i8>>,
+  CCIfType<[v64i1], CCPromoteToType<v64i8>>,
+
   // The first 8 FP/Vector arguments are passed in XMM registers.
   CCIfType<[f32, f64, v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
             CCIfSubtarget<"hasSSE1()",
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index bf61ab8ba4b..079880ef4a1 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -1907,8 +1907,12 @@ X86TargetLowering::LowerReturn(SDValue Chain,
       ValToCopy = DAG.getNode(ISD::SIGN_EXTEND, dl, VA.getLocVT(), ValToCopy);
     else if (VA.getLocInfo() == CCValAssign::ZExt)
       ValToCopy = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), ValToCopy);
-    else if (VA.getLocInfo() == CCValAssign::AExt)
-      ValToCopy = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), ValToCopy);
+    else if (VA.getLocInfo() == CCValAssign::AExt) {
+      if (ValVT.getScalarType() == MVT::i1)
+        ValToCopy = DAG.getNode(ISD::SIGN_EXTEND, dl, VA.getLocVT(), ValToCopy);
+      else
+        ValToCopy = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), ValToCopy);
+    }   
     else if (VA.getLocInfo() == CCValAssign::BCvt)
       ValToCopy = DAG.getNode(ISD::BITCAST, dl, VA.getLocVT(), ValToCopy);
 
@@ -2376,7 +2380,7 @@ X86TargetLowering::LowerFormalArguments(SDValue Chain,
 
       if (VA.isExtInLoc()) {
         // Handle MMX values passed in XMM regs.
-        if (RegVT.isVector())
+        if (RegVT.isVector() && VA.getValVT().getScalarType() != MVT::i1)
           ArgValue = DAG.getNode(X86ISD::MOVDQ2Q, dl, VA.getValVT(), ArgValue);
         else
           ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
diff --git a/test/CodeGen/X86/avx512-calling-conv.ll b/test/CodeGen/X86/avx512-calling-conv.ll
new file mode 100644
index 00000000000..cd09218810e
--- /dev/null
+++ b/test/CodeGen/X86/avx512-calling-conv.ll
@@ -0,0 +1,44 @@
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl | FileCheck %s --check-prefix=KNL
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx | FileCheck %s --check-prefix=SKX
+
+; KNL-LABEL: test1
+; KNL: vxorps
+define <16 x i1> @test1() {
+  ret <16 x i1> zeroinitializer
+}
+
+; SKX-LABEL: test2
+; SKX: vpmovb2m
+; SKX: vpmovb2m
+; SKX: kandw
+; SKX: vpmovm2b
+; KNL-LABEL: test2
+; KNL: vpmovsxbd
+; KNL: vpmovsxbd
+; KNL: vpandd
+; KNL: vpmovdb
+define <16 x i1> @test2(<16 x i1>%a, <16 x i1>%b) {
+  %c = and <16 x i1>%a, %b
+  ret <16 x i1> %c
+}
+
+; SKX-LABEL: test3
+; SKX: vpmovw2m
+; SKX: vpmovw2m
+; SKX: kandb
+; SKX: vpmovm2w
+define <8 x i1> @test3(<8 x i1>%a, <8 x i1>%b) {
+  %c = and <8 x i1>%a, %b
+  ret <8 x i1> %c
+}
+
+; SKX-LABEL: test4
+; SKX: vpmovd2m
+; SKX: vpmovd2m
+; SKX: kandw
+; SKX: vpmovm2d
+define <4 x i1> @test4(<4 x i1>%a, <4 x i1>%b) {
+  %c = and <4 x i1>%a, %b
+  ret <4 x i1> %c
+}
+