ARM Cost Model: We need to detect the max bitwidth of types in the loop in order to select the max vectorization factor.

We don't have a detailed analysis on which values are vectorized and which stay scalars in the vectorized loop so we use
another method. We look at reduction variables, loads and stores, which are the only ways to get information in and out
of loop iterations. If the data types are extended and truncated then the cost model will catch the cost of the vector
zext/sext/trunc operations.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@172178 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 64 insertions, 6 deletions

diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp
index 4803a9d038..4bb8c43656 100644
--- a/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -2635,7 +2635,7 @@ LoopVectorizationCostModel::selectVectorizationFactor(bool OptForSize,
 
   assert(MaxVectorSize <= 32 && "Did not expect to pack so many elements"
          " into one vector.");
-  
+
   unsigned VF = MaxVectorSize;
 
   // If we optimize the program for size, avoid creating the tail loop.
@@ -2697,17 +2697,23 @@ unsigned LoopVectorizationCostModel::getWidestType() {
 
     // For each instruction in the loop.
     for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) {
-      if (Legal->isUniformAfterVectorization(it))
+      Type *T = it->getType();
+
+      // Only examine Loads, Stores and PHINodes.
+      if (!isa<LoadInst>(it) && !isa<StoreInst>(it) && !isa<PHINode>(it))
         continue;
 
-      Type *T = it->getType();
+      // Examine PHI nodes that are reduction variables.
+      if (PHINode *PN = dyn_cast<PHINode>(it))
+        if (!Legal->getReductionVars()->count(PN))
+          continue;
 
+      // Examine the stored values.
       if (StoreInst *ST = dyn_cast<StoreInst>(it))
         T = ST->getValueOperand()->getType();
 
-      // PHINodes and pointers are difficult to analyze, but we catch all other
-      // uses of the types in other instructions.
-      if (isa<PHINode>(it) || T->isPointerTy() || T->isVoidTy())
+      // Ignore stored/loaded pointer types.
+      if (T->isPointerTy())
         continue;
 
       MaxWidth = std::max(MaxWidth, T->getScalarSizeInBits());
diff --git a/test/Transforms/LoopVectorize/ARM/width-detect.ll b/test/Transforms/LoopVectorize/ARM/width-detect.ll
new file mode 100644
index 0000000000..c0795b6a79
--- /dev/null
+++ b/test/Transforms/LoopVectorize/ARM/width-detect.ll
@@ -0,0 +1,52 @@
+; RUN: opt < %s  -loop-vectorize -mtriple=thumbv7-apple-ios3.0.0 -S | FileCheck %s
+
+target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:32:64-v128:32:128-a0:0:32-n32-S32"
+target triple = "thumbv7-apple-ios3.0.0"
+
+;CHECK:foo_F64
+;CHECK: <2 x double>
+;CHECK:ret
+define double @foo_F64(double* nocapture %A, i32 %n) nounwind uwtable readonly ssp {
+  %1 = icmp sgt i32 %n, 0
+  br i1 %1, label %.lr.ph, label %._crit_edge
+
+.lr.ph:                                           ; preds = %0, %.lr.ph
+  %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 0, %0 ]
+  %prod.01 = phi double [ %4, %.lr.ph ], [ 0.000000e+00, %0 ]
+  %2 = getelementptr inbounds double* %A, i64 %indvars.iv
+  %3 = load double* %2, align 8
+  %4 = fmul fast double %prod.01, %3
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, %n
+  br i1 %exitcond, label %._crit_edge, label %.lr.ph
+
+._crit_edge:                                      ; preds = %.lr.ph, %0
+  %prod.0.lcssa = phi double [ 0.000000e+00, %0 ], [ %4, %.lr.ph ]
+  ret double %prod.0.lcssa
+}
+
+;CHECK:foo_I8
+;CHECK: xor <16 x i8>
+;CHECK:ret
+define signext i8 @foo_I8(i8* nocapture %A, i32 %n) nounwind uwtable readonly ssp {
+  %1 = icmp sgt i32 %n, 0
+  br i1 %1, label %.lr.ph, label %._crit_edge
+
+.lr.ph:                                           ; preds = %0, %.lr.ph
+  %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 0, %0 ]
+  %red.01 = phi i8 [ %4, %.lr.ph ], [ 0, %0 ]
+  %2 = getelementptr inbounds i8* %A, i64 %indvars.iv
+  %3 = load i8* %2, align 1
+  %4 = xor i8 %3, %red.01
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, %n
+  br i1 %exitcond, label %._crit_edge, label %.lr.ph
+
+._crit_edge:                                      ; preds = %.lr.ph, %0
+  %red.0.lcssa = phi i8 [ 0, %0 ], [ %4, %.lr.ph ]
+  ret i8 %red.0.lcssa
+}
+
+