From 8383b539ff4c039108ee0c202a27b787621d96cf Mon Sep 17 00:00:00 2001 From: Nadav Rotem Date: Tue, 9 Apr 2013 19:44:35 +0000 Subject: Add support for bottom-up SLP vectorization infrastructure. This commit adds the infrastructure for performing bottom-up SLP vectorization (and other optimizations) on parallel computations. The infrastructure has three potential users: 1. The loop vectorizer needs to be able to vectorize AOS data structures such as (sum += A[i] + A[i+1]). 2. The BB-vectorizer needs this infrastructure for bottom-up SLP vectorization, because bottom-up vectorization is faster to compute. 3. A loop-roller needs to be able to analyze consecutive chains and roll them into a loop, in order to reduce code size. A loop roller does not need to create vector instructions, and this infrastructure separates the chain analysis from the vectorization. This patch also includes a simple (100 LOC) bottom up SLP vectorizer that uses the infrastructure, and can vectorize this code: void SAXPY(int *x, int *y, int a, int i) { x[i] = a * x[i] + y[i]; x[i+1] = a * x[i+1] + y[i+1]; x[i+2] = a * x[i+2] + y[i+2]; x[i+3] = a * x[i+3] + y[i+3]; } git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@179117 91177308-0d34-0410-b5e6-96231b3b80d8 --- test/Transforms/SLPVectorizer/X86/simple-loop.ll | 100 +++++++++++++++++++++++ 1 file changed, 100 insertions(+) create mode 100644 test/Transforms/SLPVectorizer/X86/simple-loop.ll (limited to 'test/Transforms/SLPVectorizer/X86/simple-loop.ll') diff --git a/test/Transforms/SLPVectorizer/X86/simple-loop.ll b/test/Transforms/SLPVectorizer/X86/simple-loop.ll new file mode 100644 index 0000000..4c15ed0 --- /dev/null +++ b/test/Transforms/SLPVectorizer/X86/simple-loop.ll @@ -0,0 +1,100 @@ +; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s + +target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128" +target triple = "x86_64-apple-macosx10.8.0" + +;CHECK:rollable +define i32 @rollable(i32* noalias nocapture %in, i32* noalias nocapture %out, i64 %n) nounwind ssp uwtable { + %1 = icmp eq i64 %n, 0 + br i1 %1, label %._crit_edge, label %.lr.ph + +.lr.ph: ; preds = %0, %.lr.ph + %i.019 = phi i64 [ %26, %.lr.ph ], [ 0, %0 ] + %2 = shl i64 %i.019, 2 + %3 = getelementptr inbounds i32* %in, i64 %2 +;CHECK:load <4 x i32> + %4 = load i32* %3, align 4 + %5 = or i64 %2, 1 + %6 = getelementptr inbounds i32* %in, i64 %5 + %7 = load i32* %6, align 4 + %8 = or i64 %2, 2 + %9 = getelementptr inbounds i32* %in, i64 %8 + %10 = load i32* %9, align 4 + %11 = or i64 %2, 3 + %12 = getelementptr inbounds i32* %in, i64 %11 + %13 = load i32* %12, align 4 +;CHECK:mul <4 x i32> + %14 = mul i32 %4, 7 +;CHECK:add <4 x i32> + %15 = add i32 %14, 7 + %16 = mul i32 %7, 7 + %17 = add i32 %16, 14 + %18 = mul i32 %10, 7 + %19 = add i32 %18, 21 + %20 = mul i32 %13, 7 + %21 = add i32 %20, 28 + %22 = getelementptr inbounds i32* %out, i64 %2 +;CHECK:store <4 x i32> + store i32 %15, i32* %22, align 4 + %23 = getelementptr inbounds i32* %out, i64 %5 + store i32 %17, i32* %23, align 4 + %24 = getelementptr inbounds i32* %out, i64 %8 + store i32 %19, i32* %24, align 4 + %25 = getelementptr inbounds i32* %out, i64 %11 + store i32 %21, i32* %25, align 4 + %26 = add i64 %i.019, 1 + %exitcond = icmp eq i64 %26, %n + br i1 %exitcond, label %._crit_edge, label %.lr.ph + +._crit_edge: ; preds = %.lr.ph, %0 +;CHECK: ret + ret i32 undef +} + +;CHECK:unrollable +;CHECK-NOT: <4 x i32> +;CHECK: ret +define i32 @unrollable(i32* %in, i32* %out, i64 %n) nounwind ssp uwtable { + %1 = icmp eq i64 %n, 0 + br i1 %1, label %._crit_edge, label %.lr.ph + +.lr.ph: ; preds = %0, %.lr.ph + %i.019 = phi i64 [ %26, %.lr.ph ], [ 0, %0 ] + %2 = shl i64 %i.019, 2 + %3 = getelementptr inbounds i32* %in, i64 %2 + %4 = load i32* %3, align 4 + %5 = or i64 %2, 1 + %6 = getelementptr inbounds i32* %in, i64 %5 + %7 = load i32* %6, align 4 + %8 = or i64 %2, 2 + %9 = getelementptr inbounds i32* %in, i64 %8 + %10 = load i32* %9, align 4 + %11 = or i64 %2, 3 + %12 = getelementptr inbounds i32* %in, i64 %11 + %13 = load i32* %12, align 4 + %14 = mul i32 %4, 7 + %15 = add i32 %14, 7 + %16 = mul i32 %7, 7 + %17 = add i32 %16, 14 + %18 = mul i32 %10, 7 + %19 = add i32 %18, 21 + %20 = mul i32 %13, 7 + %21 = add i32 %20, 28 + %22 = getelementptr inbounds i32* %out, i64 %2 + store i32 %15, i32* %22, align 4 + %23 = getelementptr inbounds i32* %out, i64 %5 + store i32 %17, i32* %23, align 4 + %barrier = call i32 @goo(i32 0) ; <---------------- memory barrier. + %24 = getelementptr inbounds i32* %out, i64 %8 + store i32 %19, i32* %24, align 4 + %25 = getelementptr inbounds i32* %out, i64 %11 + store i32 %21, i32* %25, align 4 + %26 = add i64 %i.019, 1 + %exitcond = icmp eq i64 %26, %n + br i1 %exitcond, label %._crit_edge, label %.lr.ph + +._crit_edge: ; preds = %.lr.ph, %0 + ret i32 undef +} + +declare i32 @goo(i32) -- cgit v1.1 From 1f098af3648f7714dd0501f8ba97601e99471806 Mon Sep 17 00:00:00 2001 From: Nadav Rotem Date: Sun, 14 Apr 2013 05:47:04 +0000 Subject: Remove unused function attributes. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@179476 91177308-0d34-0410-b5e6-96231b3b80d8 --- test/Transforms/SLPVectorizer/X86/simple-loop.ll | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'test/Transforms/SLPVectorizer/X86/simple-loop.ll') diff --git a/test/Transforms/SLPVectorizer/X86/simple-loop.ll b/test/Transforms/SLPVectorizer/X86/simple-loop.ll index 4c15ed0..0111b94 100644 --- a/test/Transforms/SLPVectorizer/X86/simple-loop.ll +++ b/test/Transforms/SLPVectorizer/X86/simple-loop.ll @@ -4,7 +4,7 @@ target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f3 target triple = "x86_64-apple-macosx10.8.0" ;CHECK:rollable -define i32 @rollable(i32* noalias nocapture %in, i32* noalias nocapture %out, i64 %n) nounwind ssp uwtable { +define i32 @rollable(i32* noalias nocapture %in, i32* noalias nocapture %out, i64 %n) { %1 = icmp eq i64 %n, 0 br i1 %1, label %._crit_edge, label %.lr.ph -- cgit v1.1