[AArch64][GlobalISel] Select all-zero G_BUILD_VECTOR into a zero mov.

Unfortunately the leaf SDAG patterns aren't supported yet so we need to do
this manually, but it's not a significant amount of code anyway.

Differential Revision: https://reviews.llvm.org/D87924

diff --git a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
index 658ff94..1daa2b2 100644 (file)
--- a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
@@ -4703,8 +4703,9 @@ bool AArch64InstructionSelector::selectInsertElt(
 bool AArch64InstructionSelector::tryOptConstantBuildVec(
     MachineInstr &I, LLT DstTy, MachineRegisterInfo &MRI) const {
   assert(I.getOpcode() == TargetOpcode::G_BUILD_VECTOR);
-  assert(DstTy.getSizeInBits() <= 128 && "Unexpected build_vec type!");
-  if (DstTy.getSizeInBits() < 32)
+  unsigned DstSize = DstTy.getSizeInBits();
+  assert(DstSize <= 128 && "Unexpected build_vec type!");
+  if (DstSize < 32)
     return false;
   // Check if we're building a constant vector, in which case we want to
   // generate a constant pool load instead of a vector insert sequence.
@@ -4725,6 +4726,24 @@ bool AArch64InstructionSelector::tryOptConstantBuildVec(
   }
   Constant *CV = ConstantVector::get(Csts);
   MachineIRBuilder MIB(I);
+  if (CV->isNullValue()) {
+    // Until the importer can support immAllZerosV in pattern leaf nodes,
+    // select a zero move manually here.
+    Register DstReg = I.getOperand(0).getReg();
+    if (DstSize == 128) {
+      auto Mov = MIB.buildInstr(AArch64::MOVIv2d_ns, {DstReg}, {}).addImm(0);
+      I.eraseFromParent();
+      return constrainSelectedInstRegOperands(*Mov, TII, TRI, RBI);
+    } else if (DstSize == 64) {
+      auto Mov =
+          MIB.buildInstr(AArch64::MOVIv2d_ns, {&AArch64::FPR128RegClass}, {})
+              .addImm(0);
+      MIB.buildInstr(TargetOpcode::COPY, {DstReg}, {})
+          .addReg(Mov.getReg(0), 0, AArch64::dsub);
+      I.eraseFromParent();
+      return RBI.constrainGenericRegister(DstReg, AArch64::FPR64RegClass, MRI);
+    }
+  }
   auto *CPLoad = emitLoadFromConstantPool(CV, MIB);
   if (!CPLoad) {
     LLVM_DEBUG(dbgs() << "Could not generate cp load for build_vector");

diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/select-build-vector.mir b/llvm/test/CodeGen/AArch64/GlobalISel/select-build-vector.mir
index d3ec9a8..af186b4 100644 (file)
--- a/llvm/test/CodeGen/AArch64/GlobalISel/select-build-vector.mir
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/select-build-vector.mir
@@ -1,28 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 # RUN: llc -mtriple=aarch64-- -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck %s
---- |
-  target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
-  target triple = "aarch64"
-
-  define <4 x float> @test_f32(float %a, float %b, float %c, float %d) {
-    ret <4 x float> undef
-  }
-
-  define <2 x double> @test_f64(double %a, double %b) {
-    ret <2 x double> undef
-  }
-
-  define <4 x i32> @test_i32(i32 %a, i32 %b, i32 %c, i32 %d) {
-    ret <4 x i32> undef
-  }
-
-  define <2 x i64> @test_i64(i64 %a, i64 %b) {
-    ret <2 x i64> undef
-  }
-
-  define void @test_p0(i64 *%a, i64 *%b) { ret void }
-
-...
 ---
 name:            test_f32
 alignment:       4
@@ -33,7 +10,7 @@ selected:        false
 failedISel:      false
 tracksRegLiveness: true
 body:             |
-  bb.0 (%ir-block.0):
+  bb.0:
     liveins: $s0, $s1, $s2, $s3
 
     ; CHECK-LABEL: name: test_f32
@@ -74,7 +51,7 @@ selected:        false
 failedISel:      false
 tracksRegLiveness: true
 body:             |
-  bb.0 (%ir-block.0):
+  bb.0:
     liveins: $d0, $d1, $d2, $d3
 
     ; CHECK-LABEL: name: test_f64
@@ -105,7 +82,7 @@ selected:        false
 failedISel:      false
 tracksRegLiveness: true
 body:             |
-  bb.0 (%ir-block.0):
+  bb.0:
     liveins: $w0, $w1, $w2, $w3
 
     ; CHECK-LABEL: name: test_i32
@@ -140,7 +117,7 @@ selected:        false
 failedISel:      false
 tracksRegLiveness: true
 body:             |
-  bb.0 (%ir-block.0):
+  bb.0:
     liveins: $x0, $x1
 
     ; CHECK-LABEL: name: test_i64
@@ -169,7 +146,7 @@ selected:        false
 failedISel:      false
 tracksRegLiveness: true
 body:             |
-  bb.0 (%ir-block.0):
+  bb.0:
     liveins: $x0, $x1
 
     ; CHECK-LABEL: name: test_p0
@@ -188,3 +165,66 @@ body:             |
     RET_ReallyLR implicit $q0
 
 ...
+---
+name:            test_v4s32_zero
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+liveins:
+  - { reg: '$x0' }
+frameInfo:
+  maxAlignment:    1
+machineFunctionInfo: {}
+body:             |
+  bb.1:
+    liveins: $x0
+
+    ; CHECK-LABEL: name: test_v4s32_zero
+    ; CHECK: liveins: $x0
+    ; CHECK: [[MOVIv2d_ns:%[0-9]+]]:fpr128 = MOVIv2d_ns 0
+    ; CHECK: $q0 = COPY [[MOVIv2d_ns]]
+    ; CHECK: RET_ReallyLR
+    %0:gpr(p0) = COPY $x0
+    %2:gpr(s32) = G_CONSTANT i32 0
+    %3:fpr(s32) = COPY %2(s32)
+    %4:fpr(s32) = COPY %2(s32)
+    %5:fpr(s32) = COPY %2(s32)
+    %6:fpr(s32) = COPY %2(s32)
+    %1:fpr(<4 x s32>) = G_BUILD_VECTOR %3(s32), %4(s32), %5(s32), %6(s32)
+    $q0 = COPY %1(<4 x s32>)
+    RET_ReallyLR
+...
+---
+name:            test_v8s8_zero
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+liveins:
+  - { reg: '$x0' }
+frameInfo:
+  maxAlignment:    1
+machineFunctionInfo: {}
+body:             |
+  bb.1:
+    liveins: $x0
+
+    ; CHECK-LABEL: name: test_v8s8_zero
+    ; CHECK: liveins: $x0
+    ; CHECK: [[MOVIv2d_ns:%[0-9]+]]:fpr128 = MOVIv2d_ns 0
+    ; CHECK: [[COPY:%[0-9]+]]:fpr64 = COPY [[MOVIv2d_ns]].dsub
+    ; CHECK: $d0 = COPY [[COPY]]
+    ; CHECK: RET_ReallyLR
+    %0:gpr(p0) = COPY $x0
+    %2:gpr(s8) = G_CONSTANT i8 0
+    %3:fpr(s8) = COPY %2(s8)
+    %4:fpr(s8) = COPY %2(s8)
+    %5:fpr(s8) = COPY %2(s8)
+    %6:fpr(s8) = COPY %2(s8)
+    %7:fpr(s8) = COPY %2(s8)
+    %8:fpr(s8) = COPY %2(s8)
+    %9:fpr(s8) = COPY %2(s8)
+    %10:fpr(s8) = COPY %2(s8)
+    %1:fpr(<8 x s8>) = G_BUILD_VECTOR %3(s8), %4(s8), %5(s8), %6(s8), %7(s8), %8(s8), %9(s8), %10(s8)
+    $d0 = COPY %1(<8 x s8>)
+    RET_ReallyLR
+...

diff --git a/llvm/test/CodeGen/AArch64/arm64-vabs.ll b/llvm/test/CodeGen/AArch64/arm64-vabs.ll
index 2e59ffc..6365229 100644 (file)
--- a/llvm/test/CodeGen/AArch64/arm64-vabs.ll
+++ b/llvm/test/CodeGen/AArch64/arm64-vabs.ll
@@ -956,8 +956,8 @@ define <2 x i32> @abspattern1(<2 x i32> %a) nounwind {
 ; DAG: abs.2s
 ; DAG-NEXT: ret
 
-; GISEL: neg.2s
-; GISEL: cmge.2s
+; GISEL-DAG: neg.2s
+; GISEL-DAG: cmge.2s
 ; GISEL: bif.8b
         %tmp1neg = sub <2 x i32> zeroinitializer, %a
         %b = icmp sge <2 x i32> %a, zeroinitializer
@@ -1035,8 +1035,8 @@ define <2 x i64> @abspattern7(<2 x i64> %a) nounwind {
 ; DAG: abs.2d
 ; DAG-NEXT: ret
 
-; GISEL: neg.2d
-; GISEL: cmge.2d
+; GISEL-DAG: neg.2d
+; GISEL-DAG: cmge.2d
 ; GISEL: bit.16b
         %tmp1neg = sub <2 x i64> zeroinitializer, %a
         %b = icmp sle <2 x i64> %a, zeroinitializer

diff --git a/llvm/test/CodeGen/AArch64/combine-loads.ll b/llvm/test/CodeGen/AArch64/combine-loads.ll
index c94751d..2e88c3c 100644 (file)
--- a/llvm/test/CodeGen/AArch64/combine-loads.ll
+++ b/llvm/test/CodeGen/AArch64/combine-loads.ll
@@ -4,8 +4,7 @@
 define <2 x i64> @z(i64* nocapture nonnull readonly %p) {
 ; CHECK-LABEL: z:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    adrp x8, .LCPI0_0
-; CHECK-NEXT:    ldr q0, [x8, :lo12:.LCPI0_0]
+; CHECK-NEXT:    movi v0.2d, #0000000000000000
 ; CHECK-NEXT:    ldr x9, [x0]
 ; CHECK-NEXT:    ldr x8, [x0, #8]
 ; CHECK-NEXT:    mov v0.d[0], x9