Precommit test for avgr pattern

badumbatish · badumbatish · commit 4a82aea026ed · 2025-08-12T12:04:14.000-07:00
diff --git a/llvm/test/CodeGen/WebAssembly/simd-avgr.ll b/llvm/test/CodeGen/WebAssembly/simd-avgr.ll
@@ -0,0 +1,230 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -O2 -mtriple=wasm32 -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mattr=+simd128 | FileCheck %s
+
+;void f(unsigned char *x, unsigned char *y, int n) {
+;  for (int i = 0; i < n; i++)
+;    x[i] = (x[i] + y[i] + 1) / 2;
+;}
+
+define void @f(ptr %x, ptr %y, i32 %n) {
+; CHECK-LABEL: f:
+; CHECK:         .functype f (i32, i32, i32) -> ()
+; CHECK-NEXT:  # %bb.0: # %entry
+; CHECK-NEXT:    block
+; CHECK-NEXT:    i32.const $push0=, 1
+; CHECK-NEXT:    i32.lt_s $push1=, $2, $pop0
+; CHECK-NEXT:    br_if 0, $pop1 # 0: down to label0
+; CHECK-NEXT:  # %bb.1: # %for.body.preheader
+; CHECK-NEXT:    i32.const $5=, 0
+; CHECK-NEXT:    block
+; CHECK-NEXT:    i32.const $push2=, 16
+; CHECK-NEXT:    i32.lt_u $push3=, $2, $pop2
+; CHECK-NEXT:    br_if 0, $pop3 # 0: down to label1
+; CHECK-NEXT:  # %bb.2: # %vector.memcheck
+; CHECK-NEXT:    block
+; CHECK-NEXT:    i32.add $push5=, $1, $2
+; CHECK-NEXT:    i32.ge_u $push6=, $0, $pop5
+; CHECK-NEXT:    br_if 0, $pop6 # 0: down to label2
+; CHECK-NEXT:  # %bb.3: # %vector.memcheck
+; CHECK-NEXT:    i32.add $push4=, $0, $2
+; CHECK-NEXT:    i32.lt_u $push7=, $1, $pop4
+; CHECK-NEXT:    br_if 1, $pop7 # 1: down to label1
+; CHECK-NEXT:  .LBB0_4: # %vector.ph
+; CHECK-NEXT:    end_block # label2:
+; CHECK-NEXT:    local.copy $6=, $0
+; CHECK-NEXT:    local.copy $7=, $1
+; CHECK-NEXT:    i32.const $push8=, 2147483632
+; CHECK-NEXT:    i32.and $push34=, $2, $pop8
+; CHECK-NEXT:    local.tee $push33=, $5=, $pop34
+; CHECK-NEXT:    local.copy $8=, $pop33
+; CHECK-NEXT:  .LBB0_5: # %vector.body
+; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT:    loop # label3:
+; CHECK-NEXT:    v128.load $push44=, 0($6):p2align=0
+; CHECK-NEXT:    local.tee $push43=, $4=, $pop44
+; CHECK-NEXT:    v128.load $push42=, 0($7):p2align=0
+; CHECK-NEXT:    local.tee $push41=, $3=, $pop42
+; CHECK-NEXT:    v128.or $push9=, $pop43, $pop41
+; CHECK-NEXT:    v128.xor $push10=, $4, $3
+; CHECK-NEXT:    i32.const $push40=, 1
+; CHECK-NEXT:    i8x16.shr_u $push11=, $pop10, $pop40
+; CHECK-NEXT:    i8x16.sub $push12=, $pop9, $pop11
+; CHECK-NEXT:    v128.store 0($6):p2align=0, $pop12
+; CHECK-NEXT:    i32.const $push39=, 16
+; CHECK-NEXT:    i32.add $6=, $6, $pop39
+; CHECK-NEXT:    i32.const $push38=, 16
+; CHECK-NEXT:    i32.add $7=, $7, $pop38
+; CHECK-NEXT:    i32.const $push37=, -16
+; CHECK-NEXT:    i32.add $push36=, $8, $pop37
+; CHECK-NEXT:    local.tee $push35=, $8=, $pop36
+; CHECK-NEXT:    br_if 0, $pop35 # 0: up to label3
+; CHECK-NEXT:  # %bb.6: # %middle.block
+; CHECK-NEXT:    end_loop
+; CHECK-NEXT:    i32.eq $push13=, $2, $5
+; CHECK-NEXT:    br_if 1, $pop13 # 1: down to label0
+; CHECK-NEXT:  .LBB0_7: # %for.body.preheader16
+; CHECK-NEXT:    end_block # label1:
+; CHECK-NEXT:    i32.const $push46=, 1
+; CHECK-NEXT:    i32.or $6=, $5, $pop46
+; CHECK-NEXT:    block
+; CHECK-NEXT:    i32.const $push45=, 1
+; CHECK-NEXT:    i32.and $push14=, $2, $pop45
+; CHECK-NEXT:    i32.eqz $push64=, $pop14
+; CHECK-NEXT:    br_if 0, $pop64 # 0: down to label4
+; CHECK-NEXT:  # %bb.8: # %for.body.prol
+; CHECK-NEXT:    i32.add $push50=, $0, $5
+; CHECK-NEXT:    local.tee $push49=, $7=, $pop50
+; CHECK-NEXT:    i32.load8_u $push17=, 0($7)
+; CHECK-NEXT:    i32.add $push15=, $1, $5
+; CHECK-NEXT:    i32.load8_u $push16=, 0($pop15)
+; CHECK-NEXT:    i32.add $push18=, $pop17, $pop16
+; CHECK-NEXT:    i32.const $push48=, 1
+; CHECK-NEXT:    i32.add $push19=, $pop18, $pop48
+; CHECK-NEXT:    i32.const $push47=, 1
+; CHECK-NEXT:    i32.shr_u $push20=, $pop19, $pop47
+; CHECK-NEXT:    i32.store8 0($pop49), $pop20
+; CHECK-NEXT:    local.copy $5=, $6
+; CHECK-NEXT:  .LBB0_9: # %for.body.prol.loopexit
+; CHECK-NEXT:    end_block # label4:
+; CHECK-NEXT:    i32.eq $push21=, $2, $6
+; CHECK-NEXT:    br_if 0, $pop21 # 0: down to label0
+; CHECK-NEXT:  # %bb.10: # %for.body.preheader1
+; CHECK-NEXT:    i32.add $6=, $0, $5
+; CHECK-NEXT:    i32.add $7=, $1, $5
+; CHECK-NEXT:    i32.sub $8=, $2, $5
+; CHECK-NEXT:  .LBB0_11: # %for.body
+; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT:    loop # label5:
+; CHECK-NEXT:    i32.load8_u $push23=, 0($6)
+; CHECK-NEXT:    i32.load8_u $push22=, 0($7)
+; CHECK-NEXT:    i32.add $push24=, $pop23, $pop22
+; CHECK-NEXT:    i32.const $push63=, 1
+; CHECK-NEXT:    i32.add $push25=, $pop24, $pop63
+; CHECK-NEXT:    i32.const $push62=, 1
+; CHECK-NEXT:    i32.shr_u $push26=, $pop25, $pop62
+; CHECK-NEXT:    i32.store8 0($6), $pop26
+; CHECK-NEXT:    i32.const $push61=, 1
+; CHECK-NEXT:    i32.add $push60=, $6, $pop61
+; CHECK-NEXT:    local.tee $push59=, $2=, $pop60
+; CHECK-NEXT:    i32.load8_u $push27=, 0($2)
+; CHECK-NEXT:    i32.const $push58=, 1
+; CHECK-NEXT:    i32.add $push28=, $7, $pop58
+; CHECK-NEXT:    i32.load8_u $push29=, 0($pop28)
+; CHECK-NEXT:    i32.add $push30=, $pop27, $pop29
+; CHECK-NEXT:    i32.const $push57=, 1
+; CHECK-NEXT:    i32.add $push31=, $pop30, $pop57
+; CHECK-NEXT:    i32.const $push56=, 1
+; CHECK-NEXT:    i32.shr_u $push32=, $pop31, $pop56
+; CHECK-NEXT:    i32.store8 0($pop59), $pop32
+; CHECK-NEXT:    i32.const $push55=, 2
+; CHECK-NEXT:    i32.add $6=, $6, $pop55
+; CHECK-NEXT:    i32.const $push54=, 2
+; CHECK-NEXT:    i32.add $7=, $7, $pop54
+; CHECK-NEXT:    i32.const $push53=, -2
+; CHECK-NEXT:    i32.add $push52=, $8, $pop53
+; CHECK-NEXT:    local.tee $push51=, $8=, $pop52
+; CHECK-NEXT:    br_if 0, $pop51 # 0: up to label5
+; CHECK-NEXT:  .LBB0_12: # %for.cond.cleanup
+; CHECK-NEXT:    end_loop
+; CHECK-NEXT:    end_block # label0:
+; CHECK-NEXT:    return
+entry:
+  %cmp12 = icmp sgt i32 %n, 0
+  br i1 %cmp12, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:
+  %min.iters.check = icmp ult i32 %n, 16
+  br i1 %min.iters.check, label %for.body.preheader16, label %vector.memcheck
+
+vector.memcheck:
+  %scevgep = getelementptr i8, ptr %x, i32 %n
+  %scevgep14 = getelementptr i8, ptr %y, i32 %n
+  %bound0 = icmp ult ptr %x, %scevgep14
+  %bound1 = icmp ult ptr %y, %scevgep
+  %found.conflict = and i1 %bound0, %bound1
+  br i1 %found.conflict, label %for.body.preheader16, label %vector.ph
+
+vector.ph:
+  %n.vec = and i32 %n, 2147483632
+  br label %vector.body
+
+vector.body:
+  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]
+  %0 = getelementptr inbounds nuw i8, ptr %x, i32 %index
+  %wide.load = load <16 x i8>, ptr %0, align 1
+  %1 = zext <16 x i8> %wide.load to <16 x i16>
+  %2 = getelementptr inbounds nuw i8, ptr %y, i32 %index
+  %wide.load15 = load <16 x i8>, ptr %2, align 1
+  %3 = zext <16 x i8> %wide.load15 to <16 x i16>
+  %4 = add nuw nsw <16 x i16> %1, splat (i16 1)
+  %5 = add nuw nsw <16 x i16> %4, %3
+  %6 = lshr <16 x i16> %5, splat (i16 1)
+  %7 = trunc nuw <16 x i16> %6 to <16 x i8>
+  store <16 x i8> %7, ptr %0, align 1
+  %index.next = add nuw i32 %index, 16
+  %8 = icmp eq i32 %index.next, %n.vec
+  br i1 %8, label %middle.block, label %vector.body
+
+middle.block:
+  %cmp.n = icmp eq i32 %n, %n.vec
+  br i1 %cmp.n, label %for.cond.cleanup, label %for.body.preheader16
+
+for.body.preheader16:
+  %i.013.ph = phi i32 [ 0, %vector.memcheck ], [ 0, %for.body.preheader ], [ %n.vec, %middle.block ]
+  %.neg = or disjoint i32 %i.013.ph, 1
+  %xtraiter = and i32 %n, 1
+  %lcmp.mod.not = icmp eq i32 %xtraiter, 0
+  br i1 %lcmp.mod.not, label %for.body.prol.loopexit, label %for.body.prol
+
+for.body.prol:
+  %arrayidx.prol = getelementptr inbounds nuw i8, ptr %x, i32 %i.013.ph
+  %9 = load i8, ptr %arrayidx.prol, align 1
+  %conv.prol = zext i8 %9 to i16
+  %arrayidx1.prol = getelementptr inbounds nuw i8, ptr %y, i32 %i.013.ph
+  %10 = load i8, ptr %arrayidx1.prol, align 1
+  %conv2.prol = zext i8 %10 to i16
+  %add.prol = add nuw nsw i16 %conv.prol, 1
+  %add3.prol = add nuw nsw i16 %add.prol, %conv2.prol
+  %div11.prol = lshr i16 %add3.prol, 1
+  %conv4.prol = trunc nuw i16 %div11.prol to i8
+  store i8 %conv4.prol, ptr %arrayidx.prol, align 1
+  %inc.prol = or disjoint i32 %i.013.ph, 1
+  br label %for.body.prol.loopexit
+
+for.body.prol.loopexit:
+  %i.013.unr = phi i32 [ %i.013.ph, %for.body.preheader16 ], [ %inc.prol, %for.body.prol ]
+  %11 = icmp eq i32 %n, %.neg
+  br i1 %11, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+  ret void
+
+for.body:
+  %i.013 = phi i32 [ %inc.1, %for.body ], [ %i.013.unr, %for.body.prol.loopexit ]
+  %arrayidx = getelementptr inbounds nuw i8, ptr %x, i32 %i.013
+  %12 = load i8, ptr %arrayidx, align 1
+  %conv = zext i8 %12 to i16
+  %arrayidx1 = getelementptr inbounds nuw i8, ptr %y, i32 %i.013
+  %13 = load i8, ptr %arrayidx1, align 1
+  %conv2 = zext i8 %13 to i16
+  %add = add nuw nsw i16 %conv, 1
+  %add3 = add nuw nsw i16 %add, %conv2
+  %div11 = lshr i16 %add3, 1
+  %conv4 = trunc nuw i16 %div11 to i8
+  store i8 %conv4, ptr %arrayidx, align 1
+  %inc = add nuw nsw i32 %i.013, 1
+  %arrayidx.1 = getelementptr inbounds nuw i8, ptr %x, i32 %inc
+  %14 = load i8, ptr %arrayidx.1, align 1
+  %conv.1 = zext i8 %14 to i16
+  %arrayidx1.1 = getelementptr inbounds nuw i8, ptr %y, i32 %inc
+  %15 = load i8, ptr %arrayidx1.1, align 1
+  %conv2.1 = zext i8 %15 to i16
+  %add.1 = add nuw nsw i16 %conv.1, 1
+  %add3.1 = add nuw nsw i16 %add.1, %conv2.1
+  %div11.1 = lshr i16 %add3.1, 1
+  %conv4.1 = trunc nuw i16 %div11.1 to i8
+  store i8 %conv4.1, ptr %arrayidx.1, align 1
+  %inc.1 = add nuw nsw i32 %i.013, 2
+  %exitcond.not.1 = icmp eq i32 %inc.1, %n
+  br i1 %exitcond.not.1, label %for.cond.cleanup, label %for.body
+}
