[polly] r245888 - [Vectorizer] Detect strides in multi-dimensional arrays

[Tobias Grosser via llvm-commits]

Author: grosser
Date: Mon Aug 24 17:20:46 2015
New Revision: 245888

URL: http://llvm.org/viewvc/llvm-project?rev=245888&view=rev
Log:
[Vectorizer] Detect strides in multi-dimensional arrays

The original code was only correct for one-dimensional arrays, but derived
incorrect strides for multi-dimensional arrays.

Added:
    polly/trunk/test/Isl/CodeGen/simple_vec_strides_multidim.ll
Modified:
    polly/trunk/lib/Analysis/ScopInfo.cpp

Modified: polly/trunk/lib/Analysis/ScopInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/polly/trunk/lib/Analysis/ScopInfo.cpp?rev=245888&r1=245887&r2=245888&view=diff
==============================================================================
--- polly/trunk/lib/Analysis/ScopInfo.cpp (original)
+++ polly/trunk/lib/Analysis/ScopInfo.cpp Mon Aug 24 17:20:46 2015
@@ -642,7 +642,10 @@ bool MemoryAccess::isStrideX(__isl_take
 
   Stride = getStride(Schedule);
   StrideX = isl_set_universe(isl_set_get_space(Stride));
-  StrideX = isl_set_fix_si(StrideX, isl_dim_set, 0, StrideWidth);
+  for (unsigned i = 0; i < isl_set_dim(StrideX, isl_dim_set) - 1; i++)
+    StrideX = isl_set_fix_si(StrideX, isl_dim_set, i, 0);
+  StrideX = isl_set_fix_si(StrideX, isl_dim_set,
+                           isl_set_dim(StrideX, isl_dim_set) - 1, StrideWidth);
   IsStrideX = isl_set_is_subset(Stride, StrideX);
 
   isl_set_free(StrideX);

Added: polly/trunk/test/Isl/CodeGen/simple_vec_strides_multidim.ll
URL: http://llvm.org/viewvc/llvm-project/polly/trunk/test/Isl/CodeGen/simple_vec_strides_multidim.ll?rev=245888&view=auto
==============================================================================
--- polly/trunk/test/Isl/CodeGen/simple_vec_strides_multidim.ll (added)
+++ polly/trunk/test/Isl/CodeGen/simple_vec_strides_multidim.ll Mon Aug 24 17:20:46 2015
@@ -0,0 +1,76 @@
+; RUN: opt %loadPolly -polly-codegen -polly-vectorizer=polly -S -dce < %s | FileCheck %s
+;
+;    void foo(long n, float A[restrict][n], float B[restrict][n],
+;             float C[restrict][n], float D[restrict][n]) {
+;      for (long i = 0; i < 4; i++)
+;        for (long j = 0; j < 4; j++)
+;          A[i][j] += B[i][0] + C[i][2 * j] + D[j][0];
+;    }
+;
+
+; CHECK: shufflevector
+; CHECK: insertelement
+; CHECK: insertelement
+; CHECK: insertelement
+; CHECK: insertelement
+; CHECK: insertelement
+; CHECK: insertelement
+; CHECK: insertelement
+; CHECK: insertelement
+; CHECK: store <4 x float>
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+
+define void @foo(i64 %n, float* noalias %A, float* noalias %B, float* noalias %C, float* noalias %D) {
+bb:
+  br label %bb3
+
+bb3:                                              ; preds = %bb25, %bb
+  %i.0 = phi i64 [ 0, %bb ], [ %tmp26, %bb25 ]
+  %exitcond2 = icmp ne i64 %i.0, 4
+  br i1 %exitcond2, label %bb4, label %bb27
+
+bb4:                                              ; preds = %bb3
+  br label %bb5
+
+bb5:                                              ; preds = %bb22, %bb4
+  %j.0 = phi i64 [ 0, %bb4 ], [ %tmp23, %bb22 ]
+  %exitcond = icmp ne i64 %j.0, 4
+  br i1 %exitcond, label %bb6, label %bb24
+
+bb6:                                              ; preds = %bb5
+  %tmp = mul nsw i64 %i.0, %n
+  %tmp7 = getelementptr inbounds float, float* %B, i64 %tmp
+  %tmp8 = load float, float* %tmp7, align 4
+  %tmp9 = shl nsw i64 %j.0, 1
+  %tmp10 = mul nsw i64 %i.0, %n
+  %.sum = add i64 %tmp10, %tmp9
+  %tmp11 = getelementptr inbounds float, float* %C, i64 %.sum
+  %tmp12 = load float, float* %tmp11, align 4
+  %tmp13 = fadd float %tmp8, %tmp12
+  %tmp14 = mul nsw i64 %j.0, %n
+  %tmp15 = getelementptr inbounds float, float* %D, i64 %tmp14
+  %tmp16 = load float, float* %tmp15, align 4
+  %tmp17 = fadd float %tmp13, %tmp16
+  %tmp18 = mul nsw i64 %i.0, %n
+  %.sum1 = add i64 %tmp18, %j.0
+  %tmp19 = getelementptr inbounds float, float* %A, i64 %.sum1
+  %tmp20 = load float, float* %tmp19, align 4
+  %tmp21 = fadd float %tmp20, %tmp17
+  store float %tmp21, float* %tmp19, align 4
+  br label %bb22
+
+bb22:                                             ; preds = %bb6
+  %tmp23 = add nsw i64 %j.0, 1
+  br label %bb5
+
+bb24:                                             ; preds = %bb5
+  br label %bb25
+
+bb25:                                             ; preds = %bb24
+  %tmp26 = add nsw i64 %i.0, 1
+  br label %bb3
+
+bb27:                                             ; preds = %bb3
+  ret void
+}