8ebec3d551
This patch improves constant folding through `llvm.vector.insert`. It does not change anything for fixed-length vectors (which can already be folded to ConstantVectors for these cases), but folds scalable vectors that otherwise would not be folded. These folds preserve the destination vector (which could be undef or poison), giving targets more freedom in lowering the operations.
157 lines
7.1 KiB
LLVM
157 lines
7.1 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
|
|
; RUN: opt -S -passes=instcombine %s | FileCheck %s
|
|
; RUN: opt -S -passes=instcombine %s \
|
|
; RUN: -use-constant-int-for-fixed-length-splat \
|
|
; RUN -use-constant-fp-for-fixed-length-splat \
|
|
; RUN: -use-constant-int-for-scalable-splat \
|
|
; RUN: -use-constant-fp-for-scalable-splat | FileCheck %s
|
|
|
|
define <vscale x 4 x i32> @insert_div() {
|
|
; CHECK-LABEL: @insert_div(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[DIV:%.*]] = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> poison, <4 x i32> splat (i32 3), i64 0)
|
|
; CHECK-NEXT: ret <vscale x 4 x i32> [[DIV]]
|
|
;
|
|
entry:
|
|
%0 = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> poison, <4 x i32> splat (i32 9), i64 0)
|
|
%div = udiv <vscale x 4 x i32> %0, splat (i32 3)
|
|
ret <vscale x 4 x i32> %div
|
|
}
|
|
|
|
define <vscale x 4 x i32> @insert_div_splat_lhs() {
|
|
; CHECK-LABEL: @insert_div_splat_lhs(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[DIV:%.*]] = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> splat (i32 5), <4 x i32> splat (i32 2), i64 0)
|
|
; CHECK-NEXT: ret <vscale x 4 x i32> [[DIV]]
|
|
;
|
|
entry:
|
|
%0 = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> splat(i32 2), <4 x i32> splat (i32 5), i64 0)
|
|
%div = udiv <vscale x 4 x i32> splat (i32 10), %0
|
|
ret <vscale x 4 x i32> %div
|
|
}
|
|
|
|
define <vscale x 4 x i32> @insert_div_mixed_splat() {
|
|
; CHECK-LABEL: @insert_div_mixed_splat(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[DIV:%.*]] = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> splat (i32 6), <4 x i32> splat (i32 3), i64 0)
|
|
; CHECK-NEXT: ret <vscale x 4 x i32> [[DIV]]
|
|
;
|
|
entry:
|
|
%0 = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> splat (i32 18), <4 x i32> splat (i32 9), i64 0)
|
|
%div = udiv <vscale x 4 x i32> %0, splat (i32 3)
|
|
ret <vscale x 4 x i32> %div
|
|
}
|
|
|
|
define <vscale x 4 x i32> @insert_mul() {
|
|
; CHECK-LABEL: @insert_mul(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[MUL:%.*]] = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> poison, <4 x i32> splat (i32 7), i64 4)
|
|
; CHECK-NEXT: ret <vscale x 4 x i32> [[MUL]]
|
|
;
|
|
entry:
|
|
%0 = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> poison, <4 x i32> splat (i32 1), i64 4)
|
|
%mul = mul <vscale x 4 x i32> %0, splat (i32 7)
|
|
ret <vscale x 4 x i32> %mul
|
|
}
|
|
|
|
define <vscale x 4 x i32> @insert_add() {
|
|
; CHECK-LABEL: @insert_add(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[ADD:%.*]] = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> poison, <4 x i32> splat (i32 16), i64 0)
|
|
; CHECK-NEXT: ret <vscale x 4 x i32> [[ADD]]
|
|
;
|
|
entry:
|
|
%0 = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> poison, <4 x i32> splat (i32 5), i64 0)
|
|
%add = add <vscale x 4 x i32> %0, splat (i32 11)
|
|
ret <vscale x 4 x i32> %add
|
|
}
|
|
|
|
define <vscale x 4 x i32> @insert_add_non_splat_subvector() {
|
|
; CHECK-LABEL: @insert_add_non_splat_subvector(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[ADD:%.*]] = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> poison, <4 x i32> <i32 101, i32 102, i32 103, i32 104>, i64 0)
|
|
; CHECK-NEXT: ret <vscale x 4 x i32> [[ADD]]
|
|
;
|
|
entry:
|
|
%0 = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 4>, i64 0)
|
|
%add = add <vscale x 4 x i32> %0, splat (i32 100)
|
|
ret <vscale x 4 x i32> %add
|
|
}
|
|
|
|
define <vscale x 4 x float> @insert_add_fp() {
|
|
; CHECK-LABEL: @insert_add_fp(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[ADD:%.*]] = call <vscale x 4 x float> @llvm.vector.insert.nxv4f32.v4f32(<vscale x 4 x float> splat (float 6.250000e+00), <4 x float> splat (float 5.500000e+00), i64 0)
|
|
; CHECK-NEXT: ret <vscale x 4 x float> [[ADD]]
|
|
;
|
|
entry:
|
|
%0 = call <vscale x 4 x float> @llvm.vector.insert.nxv4f32.v4f32(<vscale x 4 x float> splat(float 1.25), <4 x float> splat (float 0.5), i64 0)
|
|
%add = fadd <vscale x 4 x float> %0, splat (float 5.0)
|
|
ret <vscale x 4 x float> %add
|
|
}
|
|
|
|
define <vscale x 8 x i32> @insert_add_scalable_subvector() {
|
|
; CHECK-LABEL: @insert_add_scalable_subvector(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[ADD:%.*]] = call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> splat (i32 20), <vscale x 4 x i32> splat (i32 -4), i64 0)
|
|
; CHECK-NEXT: ret <vscale x 8 x i32> [[ADD]]
|
|
;
|
|
entry:
|
|
%0 = call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> splat(i32 16), <vscale x 4 x i32> splat (i32 -8), i64 0)
|
|
%add = add <vscale x 8 x i32> %0, splat (i32 4)
|
|
ret <vscale x 8 x i32> %add
|
|
}
|
|
|
|
define <vscale x 4 x i32> @insert_sub() {
|
|
; CHECK-LABEL: @insert_sub(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[SUB:%.*]] = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> poison, <4 x i32> zeroinitializer, i64 8)
|
|
; CHECK-NEXT: ret <vscale x 4 x i32> [[SUB]]
|
|
;
|
|
entry:
|
|
%0 = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> poison, <4 x i32> splat (i32 11), i64 8)
|
|
%sub = add <vscale x 4 x i32> %0, splat (i32 -11)
|
|
ret <vscale x 4 x i32> %sub
|
|
}
|
|
|
|
define <vscale x 4 x i32> @insert_and_partially_undef() {
|
|
; CHECK-LABEL: @insert_and_partially_undef(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[AND:%.*]] = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> zeroinitializer, <4 x i32> splat (i32 4), i64 0)
|
|
; CHECK-NEXT: ret <vscale x 4 x i32> [[AND]]
|
|
;
|
|
entry:
|
|
%0 = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> undef, <4 x i32> splat (i32 6), i64 0)
|
|
%and = and <vscale x 4 x i32> %0, splat (i32 4)
|
|
ret <vscale x 4 x i32> %and
|
|
}
|
|
|
|
define <vscale x 4 x i32> @insert_fold_chain() {
|
|
; CHECK-LABEL: @insert_fold_chain(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[ADD:%.*]] = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> splat (i32 11), <4 x i32> splat (i32 8), i64 0)
|
|
; CHECK-NEXT: ret <vscale x 4 x i32> [[ADD]]
|
|
;
|
|
entry:
|
|
%0 = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> splat (i32 21), <4 x i32> splat (i32 12), i64 0)
|
|
%div = udiv <vscale x 4 x i32> %0, splat (i32 3)
|
|
%add = add <vscale x 4 x i32> %div, splat (i32 4)
|
|
ret <vscale x 4 x i32> %add
|
|
}
|
|
|
|
; TODO: This could be folded more.
|
|
define <vscale x 4 x i32> @insert_add_both_insert_vector() {
|
|
; CHECK-LABEL: @insert_add_both_insert_vector(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: [[TMP0:%.*]] = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> splat (i32 10), <4 x i32> splat (i32 5), i64 0)
|
|
; CHECK-NEXT: [[TMP1:%.*]] = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> splat (i32 -1), <4 x i32> splat (i32 2), i64 0)
|
|
; CHECK-NEXT: [[ADD:%.*]] = add <vscale x 4 x i32> [[TMP0]], [[TMP1]]
|
|
; CHECK-NEXT: ret <vscale x 4 x i32> [[ADD]]
|
|
;
|
|
entry:
|
|
%0 = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> splat(i32 10), <4 x i32> splat (i32 5), i64 0)
|
|
%1 = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v4i32(<vscale x 4 x i32> splat(i32 -1), <4 x i32> splat (i32 2), i64 0)
|
|
%add = add <vscale x 4 x i32> %0, %1
|
|
ret <vscale x 4 x i32> %add
|
|
}
|