aboutsummaryrefslogtreecommitdiffstats
path: root/test/Transforms/BBVectorize
diff options
context:
space:
mode:
Diffstat (limited to 'test/Transforms/BBVectorize')
-rw-r--r--test/Transforms/BBVectorize/no-ldstr-conn.ll23
-rw-r--r--test/Transforms/BBVectorize/simple-int.ll44
-rw-r--r--test/Transforms/BBVectorize/simple-ldstr-ptrs.ll81
-rw-r--r--test/Transforms/BBVectorize/simple-sel.ll30
4 files changed, 178 insertions, 0 deletions
diff --git a/test/Transforms/BBVectorize/no-ldstr-conn.ll b/test/Transforms/BBVectorize/no-ldstr-conn.ll
new file mode 100644
index 0000000..ada2a71
--- /dev/null
+++ b/test/Transforms/BBVectorize/no-ldstr-conn.ll
@@ -0,0 +1,23 @@
+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"
+; RUN: opt < %s -bb-vectorize -bb-vectorize-req-chain-depth=2 -instcombine -gvn -S | FileCheck %s
+
+; Make sure that things (specifically getelementptr) are not connected to loads
+; and stores via the address operand (which would be bad because the address
+; is really a scalar even after vectorization)
+define i64 @test2(i64 %a) nounwind uwtable readonly {
+entry:
+ %a1 = inttoptr i64 %a to i64*
+ %a2 = getelementptr i64* %a1, i64 1
+ %a3 = getelementptr i64* %a1, i64 2
+ %v2 = load i64* %a2, align 8
+ %v3 = load i64* %a3, align 8
+ %v2a = add i64 %v2, 5
+ %v3a = add i64 %v3, 7
+ store i64 %v2a, i64* %a2, align 8
+ store i64 %v3a, i64* %a3, align 8
+ %r = add i64 %v2, %v3
+ ret i64 %r
+; CHECK: @test2
+; CHECK-NOT: getelementptr <2 x i64*>
+}
+
diff --git a/test/Transforms/BBVectorize/simple-int.ll b/test/Transforms/BBVectorize/simple-int.ll
index b2ef27b..6844977 100644
--- a/test/Transforms/BBVectorize/simple-int.ll
+++ b/test/Transforms/BBVectorize/simple-int.ll
@@ -3,6 +3,7 @@ target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f3
declare double @llvm.fma.f64(double, double, double)
declare double @llvm.cos.f64(double)
+declare double @llvm.powi.f64(double, i32)
; Basic depth-3 chain with fma
define double @test1(double %A1, double %A2, double %B1, double %B2, double %C1, double %C2) {
@@ -54,6 +55,49 @@ define double @test2(double %A1, double %A2, double %B1, double %B2) {
; CHECK: ret double %R
}
+; Basic depth-3 chain with powi
+define double @test3(double %A1, double %A2, double %B1, double %B2, i32 %P) {
+
+ %X1 = fsub double %A1, %B1
+ %X2 = fsub double %A2, %B2
+ %Y1 = call double @llvm.powi.f64(double %X1, i32 %P)
+ %Y2 = call double @llvm.powi.f64(double %X2, i32 %P)
+ %Z1 = fadd double %Y1, %B1
+ %Z2 = fadd double %Y2, %B2
+ %R = fmul double %Z1, %Z2
+ ret double %R
+; CHECK: @test3
+; CHECK: %X1.v.i1.1 = insertelement <2 x double> undef, double %B1, i32 0
+; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0
+; CHECK: %X1.v.i1.2 = insertelement <2 x double> %X1.v.i1.1, double %B2, i32 1
+; CHECK: %X1.v.i0.2 = insertelement <2 x double> %X1.v.i0.1, double %A2, i32 1
+; CHECK: %X1 = fsub <2 x double> %X1.v.i0.2, %X1.v.i1.2
+; CHECK: %Y1 = call <2 x double> @llvm.powi.v2f64(<2 x double> %X1, i32 %P)
+; CHECK: %Z1 = fadd <2 x double> %Y1, %X1.v.i1.2
+; CHECK: %Z1.v.r1 = extractelement <2 x double> %Z1, i32 0
+; CHECK: %Z1.v.r2 = extractelement <2 x double> %Z1, i32 1
+; CHECK: %R = fmul double %Z1.v.r1, %Z1.v.r2
+; CHECK: ret double %R
+}
+
+; Basic depth-3 chain with powi (different powers: should not vectorize)
+define double @test4(double %A1, double %A2, double %B1, double %B2, i32 %P) {
+
+ %X1 = fsub double %A1, %B1
+ %X2 = fsub double %A2, %B2
+ %P2 = add i32 %P, 1
+ %Y1 = call double @llvm.powi.f64(double %X1, i32 %P)
+ %Y2 = call double @llvm.powi.f64(double %X2, i32 %P2)
+ %Z1 = fadd double %Y1, %B1
+ %Z2 = fadd double %Y2, %B2
+ %R = fmul double %Z1, %Z2
+ ret double %R
+; CHECK: @test4
+; CHECK-NOT: <2 x double>
+; CHECK: ret double %R
+}
+
; CHECK: declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>) nounwind readnone
; CHECK: declare <2 x double> @llvm.cos.v2f64(<2 x double>) nounwind readonly
+; CHECK: declare <2 x double> @llvm.powi.v2f64(<2 x double>, i32) nounwind readonly
diff --git a/test/Transforms/BBVectorize/simple-ldstr-ptrs.ll b/test/Transforms/BBVectorize/simple-ldstr-ptrs.ll
new file mode 100644
index 0000000..f992d41
--- /dev/null
+++ b/test/Transforms/BBVectorize/simple-ldstr-ptrs.ll
@@ -0,0 +1,81 @@
+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"
+; RUN: opt < %s -bb-vectorize -bb-vectorize-req-chain-depth=3 -instcombine -gvn -S | FileCheck %s
+; RUN: opt < %s -bb-vectorize -bb-vectorize-req-chain-depth=3 -bb-vectorize-aligned-only -instcombine -gvn -S | FileCheck %s -check-prefix=CHECK-AO
+
+; Simple 3-pair chain also with loads and stores (using ptrs and gep)
+define double @test1(i64* %a, i64* %b, i64* %c) nounwind uwtable readonly {
+entry:
+ %i0 = load i64* %a, align 8
+ %i1 = load i64* %b, align 8
+ %mul = mul i64 %i0, %i1
+ %arrayidx3 = getelementptr inbounds i64* %a, i64 1
+ %i3 = load i64* %arrayidx3, align 8
+ %arrayidx4 = getelementptr inbounds i64* %b, i64 1
+ %i4 = load i64* %arrayidx4, align 8
+ %mul5 = mul i64 %i3, %i4
+ %ptr = inttoptr i64 %mul to double*
+ %ptr5 = inttoptr i64 %mul5 to double*
+ %aptr = getelementptr inbounds double* %ptr, i64 2
+ %aptr5 = getelementptr inbounds double* %ptr5, i64 3
+ %av = load double* %aptr, align 16
+ %av5 = load double* %aptr5, align 16
+ %r = fmul double %av, %av5
+ store i64 %mul, i64* %c, align 8
+ %arrayidx5 = getelementptr inbounds i64* %c, i64 1
+ store i64 %mul5, i64* %arrayidx5, align 8
+ ret double %r
+; CHECK: @test1
+; CHECK: %i0.v.i0 = bitcast i64* %a to <2 x i64>*
+; CHECK: %i1.v.i0 = bitcast i64* %b to <2 x i64>*
+; CHECK: %i0 = load <2 x i64>* %i0.v.i0, align 8
+; CHECK: %i1 = load <2 x i64>* %i1.v.i0, align 8
+; CHECK: %mul = mul <2 x i64> %i0, %i1
+; CHECK: %ptr = inttoptr <2 x i64> %mul to <2 x double*>
+; CHECK: %aptr = getelementptr inbounds <2 x double*> %ptr, <2 x i64> <i64 2, i64 3>
+; CHECK: %aptr.v.r1 = extractelement <2 x double*> %aptr, i32 0
+; CHECK: %aptr.v.r2 = extractelement <2 x double*> %aptr, i32 1
+; CHECK: %av = load double* %aptr.v.r1, align 16
+; CHECK: %av5 = load double* %aptr.v.r2, align 16
+; CHECK: %r = fmul double %av, %av5
+; CHECK: %0 = bitcast i64* %c to <2 x i64>*
+; CHECK: store <2 x i64> %mul, <2 x i64>* %0, align 8
+; CHECK: ret double %r
+; CHECK-AO: @test1
+; CHECK-AO-NOT: load <2 x
+}
+
+; Simple 3-pair chain with loads and stores (using ptrs and gep)
+define void @test2(i64** %a, i64** %b, i64** %c) nounwind uwtable readonly {
+entry:
+ %i0 = load i64** %a, align 8
+ %i1 = load i64** %b, align 8
+ %arrayidx3 = getelementptr inbounds i64** %a, i64 1
+ %i3 = load i64** %arrayidx3, align 8
+ %arrayidx4 = getelementptr inbounds i64** %b, i64 1
+ %i4 = load i64** %arrayidx4, align 8
+ %o1 = load i64* %i1, align 8
+ %o4 = load i64* %i4, align 8
+ %ptr0 = getelementptr inbounds i64* %i0, i64 %o1
+ %ptr3 = getelementptr inbounds i64* %i3, i64 %o4
+ store i64* %ptr0, i64** %c, align 8
+ %arrayidx5 = getelementptr inbounds i64** %c, i64 1
+ store i64* %ptr3, i64** %arrayidx5, align 8
+ ret void
+; CHECK: @test2
+; CHECK: %i0.v.i0 = bitcast i64** %a to <2 x i64*>*
+; CHECK: %i1 = load i64** %b, align 8
+; CHECK: %i0 = load <2 x i64*>* %i0.v.i0, align 8
+; CHECK: %arrayidx4 = getelementptr inbounds i64** %b, i64 1
+; CHECK: %i4 = load i64** %arrayidx4, align 8
+; CHECK: %o1 = load i64* %i1, align 8
+; CHECK: %o4 = load i64* %i4, align 8
+; CHECK: %ptr0.v.i1.1 = insertelement <2 x i64> undef, i64 %o1, i32 0
+; CHECK: %ptr0.v.i1.2 = insertelement <2 x i64> %ptr0.v.i1.1, i64 %o4, i32 1
+; CHECK: %ptr0 = getelementptr inbounds <2 x i64*> %i0, <2 x i64> %ptr0.v.i1.2
+; CHECK: %0 = bitcast i64** %c to <2 x i64*>*
+; CHECK: store <2 x i64*> %ptr0, <2 x i64*>* %0, align 8
+; CHECK: ret void
+; CHECK-AO: @test2
+; CHECK-AO-NOT: <2 x
+}
+
diff --git a/test/Transforms/BBVectorize/simple-sel.ll b/test/Transforms/BBVectorize/simple-sel.ll
new file mode 100644
index 0000000..4daa571
--- /dev/null
+++ b/test/Transforms/BBVectorize/simple-sel.ll
@@ -0,0 +1,30 @@
+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"
+; RUN: opt < %s -bb-vectorize -bb-vectorize-req-chain-depth=3 -instcombine -gvn -S | FileCheck %s
+
+; Basic depth-3 chain with select
+define double @test1(double %A1, double %A2, double %B1, double %B2, i1 %C1, i1 %C2) {
+; CHECK: @test1
+; CHECK: %X1.v.i1.1 = insertelement <2 x double> undef, double %B1, i32 0
+; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0
+; CHECK: %X1.v.i1.2 = insertelement <2 x double> %X1.v.i1.1, double %B2, i32 1
+; CHECK: %X1.v.i0.2 = insertelement <2 x double> %X1.v.i0.1, double %A2, i32 1
+ %X1 = fsub double %A1, %B1
+ %X2 = fsub double %A2, %B2
+; CHECK: %X1 = fsub <2 x double> %X1.v.i0.2, %X1.v.i1.2
+ %Y1 = fmul double %X1, %A1
+ %Y2 = fmul double %X2, %A2
+; CHECK: %Y1 = fmul <2 x double> %X1, %X1.v.i0.2
+ %Z1 = select i1 %C1, double %Y1, double %B1
+ %Z2 = select i1 %C2, double %Y2, double %B2
+; CHECK: %Z1.v.i0.1 = insertelement <2 x i1> undef, i1 %C1, i32 0
+; CHECK: %Z1.v.i0.2 = insertelement <2 x i1> %Z1.v.i0.1, i1 %C2, i32 1
+; CHECK: %Z1 = select <2 x i1> %Z1.v.i0.2, <2 x double> %Y1, <2 x double> %X1.v.i1.2
+ %R = fmul double %Z1, %Z2
+; CHECK: %Z1.v.r1 = extractelement <2 x double> %Z1, i32 0
+; CHECK: %Z1.v.r2 = extractelement <2 x double> %Z1, i32 1
+; CHECK: %R = fmul double %Z1.v.r1, %Z1.v.r2
+ ret double %R
+; CHECK: ret double %R
+}
+
+
generated by cgit v1.1 at 2025-08-15 14:19:21 +0000