[llvm] r291368 - [AVX-512] Add masked forms of the alternate MOVDDUP patterns.

[Craig Topper via llvm-commits]

Author: ctopper
Date: Sat Jan  7 16:20:23 2017
New Revision: 291368

URL: http://llvm.org/viewvc/llvm-project?rev=291368&view=rev
Log:
[AVX-512] Add masked forms of the alternate MOVDDUP patterns.

I'm not too sure how to get isel to select even all of the unmasked forms, but at least we have a consistent set now.

Modified:
    llvm/trunk/lib/Target/X86/X86InstrAVX512.td
    llvm/trunk/test/CodeGen/X86/vector-shuffle-masked.ll

Modified: llvm/trunk/lib/Target/X86/X86InstrAVX512.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86InstrAVX512.td?rev=291368&r1=291367&r2=291368&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86InstrAVX512.td (original)
+++ llvm/trunk/lib/Target/X86/X86InstrAVX512.td Sat Jan  7 16:20:23 2017
@@ -8646,6 +8646,28 @@ def : Pat<(v2f64 (X86VBroadcast (loadf64
           (VMOVDDUPZ128rm addr:$src)>;
 def : Pat<(v2f64 (X86VBroadcast f64:$src)),
           (VMOVDDUPZ128rr (COPY_TO_REGCLASS FR64X:$src, VR128X))>;
+
+def : Pat<(vselect (v2i1 VK2WM:$mask), (X86Movddup (loadv2f64 addr:$src)),
+                   (v2f64 VR128X:$src0)),
+          (VMOVDDUPZ128rmk VR128X:$src0, VK2WM:$mask, addr:$src)>;
+def : Pat<(vselect (v2i1 VK2WM:$mask), (X86Movddup (loadv2f64 addr:$src)),
+                   (bitconvert (v4i32 immAllZerosV))),
+          (VMOVDDUPZ128rmkz VK2WM:$mask, addr:$src)>;
+
+def : Pat<(vselect (v2i1 VK2WM:$mask), (v2f64 (X86VBroadcast f64:$src)),
+                   (v2f64 VR128X:$src0)),
+          (VMOVDDUPZ128rrk VR128X:$src0, VK2WM:$mask,
+                           (COPY_TO_REGCLASS FR64X:$src, VR128X))>;
+def : Pat<(vselect (v2i1 VK2WM:$mask), (v2f64 (X86VBroadcast f64:$src)),
+                   (bitconvert (v4i32 immAllZerosV))),
+          (VMOVDDUPZ128rrkz VK2WM:$mask, (COPY_TO_REGCLASS FR64X:$src, VR128X))>;
+
+def : Pat<(vselect (v2i1 VK2WM:$mask), (v2f64 (X86VBroadcast (loadf64 addr:$src))),
+                   (v2f64 VR128X:$src0)),
+          (VMOVDDUPZ128rmk VR128X:$src0, VK2WM:$mask, addr:$src)>;
+def : Pat<(vselect (v2i1 VK2WM:$mask), (v2f64 (X86VBroadcast (loadf64 addr:$src))),
+                   (bitconvert (v4i32 immAllZerosV))),
+          (VMOVDDUPZ128rmkz VK2WM:$mask, addr:$src)>;
 }
 
 //===----------------------------------------------------------------------===//

Modified: llvm/trunk/test/CodeGen/X86/vector-shuffle-masked.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-shuffle-masked.ll?rev=291368&r1=291367&r2=291368&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vector-shuffle-masked.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vector-shuffle-masked.ll Sat Jan  7 16:20:23 2017
@@ -686,3 +686,33 @@ define <2 x double> @mask_cast_extract_v
   %res = select <2 x i1> %mask.extract, <2 x double> %shuffle.cast, <2 x double> %passthru
   ret <2 x double> %res
 }
+
+define <2 x double> @broadcast_v4f32_0101_from_v2f32_mask(double* %x, <2 x double> %passthru, i8 %mask) {
+; CHECK-LABEL: broadcast_v4f32_0101_from_v2f32_mask:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    kmovb %esi, %k1
+; CHECK-NEXT:    vblendmpd (%rdi){1to2}, %xmm0, %xmm0 {%k1}
+; CHECK-NEXT:    retq
+  %q = load double, double* %x, align 1
+  %vecinit.i = insertelement <2 x double> undef, double %q, i32 0
+  %vecinit2.i = insertelement <2 x double> %vecinit.i, double %q, i32 1
+  %mask.cast = bitcast i8 %mask to <8 x i1>
+  %mask.extract = shufflevector <8 x i1> %mask.cast, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %res = select <2 x i1> %mask.extract, <2 x double> %vecinit2.i, <2 x double> %passthru
+  ret <2 x double> %res
+}
+
+define <2 x double> @broadcast_v4f32_0101_from_v2f32_maskz(double* %x, i8 %mask) {
+; CHECK-LABEL: broadcast_v4f32_0101_from_v2f32_maskz:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    kmovb %esi, %k1
+; CHECK-NEXT:    vmovddup {{.*#+}} xmm0 {%k1} {z} = mem[0,0]
+; CHECK-NEXT:    retq
+  %q = load double, double* %x, align 1
+  %vecinit.i = insertelement <2 x double> undef, double %q, i32 0
+  %vecinit2.i = insertelement <2 x double> %vecinit.i, double %q, i32 1
+  %mask.cast = bitcast i8 %mask to <8 x i1>
+  %mask.extract = shufflevector <8 x i1> %mask.cast, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %res = select <2 x i1> %mask.extract, <2 x double> %vecinit2.i, <2 x double> zeroinitializer
+  ret <2 x double> %res
+}