[llvm] 4f0fe6b - [ARM][MVE] Tail predicate bottom/top muls.

[Sam Parker via llvm-commits]

Author: Sam Parker
Date: 2019-12-20T08:33:01Z
New Revision: 4f0fe6b97e4463d5c8571ac71b23c63387251444

URL: https://github.com/llvm/llvm-project/commit/4f0fe6b97e4463d5c8571ac71b23c63387251444
DIFF: https://github.com/llvm/llvm-project/commit/4f0fe6b97e4463d5c8571ac71b23c63387251444.diff

LOG: [ARM][MVE] Tail predicate bottom/top muls.

Add VMULL and VQDMULL variants to our tail predication white list.

Differential Revision: https://reviews.llvm.org/D71465

Added: 
    

Modified: 
    llvm/lib/Target/ARM/ARMInstrMVE.td
    llvm/unittests/Target/ARM/MachineInstrTest.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/ARM/ARMInstrMVE.td b/llvm/lib/Target/ARM/ARMInstrMVE.td
index 82a622d799a5..1b631b4b9e1b 100644
--- a/llvm/lib/Target/ARM/ARMInstrMVE.td
+++ b/llvm/lib/Target/ARM/ARMInstrMVE.td
@@ -3967,6 +3967,7 @@ class MVE_VMULL<string iname, string suffix, bit bit_28, bits<2> bits_21_20,
   let Inst{8} = 0b0;
   let Inst{7} = Qn{3};
   let Inst{0} = 0b0;
+  let validForTailPredication = 1;
 }
 
 multiclass MVE_VMULL_m<MVEVectorVTInfo VTI,
@@ -4267,6 +4268,7 @@ class MVE_VQDMULL<string iname, string suffix, bit size, bit T,
   let Inst{8} = 0b1;
   let Inst{7} = Qn{3};
   let Inst{0} = 0b1;
+  let validForTailPredication = 1;
 }
 
 multiclass MVE_VQDMULL_halves<string suffix, bit size, string cstr=""> {
@@ -4382,6 +4384,7 @@ class MVE_VQDMULL_qr<string iname, string suffix, bit size,
   let Inst{12} = T;
   let Inst{8} = 0b1;
   let Inst{5} = 0b1;
+  let validForTailPredication = 1;
 }
 
 multiclass MVE_VQDMULL_qr_halves<string suffix, bit size, string cstr=""> {

diff  --git a/llvm/unittests/Target/ARM/MachineInstrTest.cpp b/llvm/unittests/Target/ARM/MachineInstrTest.cpp
index 8807a2f31826..581d8be26527 100644
--- a/llvm/unittests/Target/ARM/MachineInstrTest.cpp
+++ b/llvm/unittests/Target/ARM/MachineInstrTest.cpp
@@ -15,7 +15,6 @@ using namespace llvm;
 // descriptions. Currently we, conservatively, disallow:
 // - cross beat carries.
 // - narrowing of results.
-// - top/bottom operations.
 // - complex operations.
 // - horizontal operations.
 // - byte swapping.
@@ -243,6 +242,22 @@ TEST(MachineInstrValidTailPredication, IsCorrect) {
     case MVE_VMOVimmi32:	
     case MVE_VMOVimmi64:
     case MVE_VMOVimmi8:	
+    case MVE_VMULLBp16:
+    case MVE_VMULLBp8:
+    case MVE_VMULLBs16:
+    case MVE_VMULLBs32:
+    case MVE_VMULLBs8:
+    case MVE_VMULLBu16:
+    case MVE_VMULLBu32:
+    case MVE_VMULLBu8:
+    case MVE_VMULLTp16:
+    case MVE_VMULLTp8:
+    case MVE_VMULLTs16:
+    case MVE_VMULLTs32:
+    case MVE_VMULLTs8:
+    case MVE_VMULLTu16:
+    case MVE_VMULLTu32:
+    case MVE_VMULLTu8:	
     case MVE_VMUL_qr_f16:
     case MVE_VMUL_qr_f32:
     case MVE_VMUL_qr_i16:
@@ -287,6 +302,14 @@ TEST(MachineInstrValidTailPredication, IsCorrect) {
     case MVE_VQADDu16:
     case MVE_VQADDu32:
     case MVE_VQADDu8:
+    case MVE_VQDMULL_qr_s16bh:
+    case MVE_VQDMULL_qr_s16th:
+    case MVE_VQDMULL_qr_s32bh:
+    case MVE_VQDMULL_qr_s32th:
+    case MVE_VQDMULLs16bh:
+    case MVE_VQDMULLs16th:
+    case MVE_VQDMULLs32bh:
+    case MVE_VQDMULLs32th:
     case MVE_VQNEGs16:
     case MVE_VQNEGs32:
     case MVE_VQNEGs8: