Build on top of https://github.com/llvm/llvm-project/pull/154510 to completely remove the blocks of dead scalar loops. Depends on https://github.com/llvm/llvm-project/pull/154510. PR: https://github.com/llvm/llvm-project/pull/155497
85 lines
5.1 KiB
LLVM
85 lines
5.1 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 5
|
|
; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v -S | FileCheck %s -check-prefix=NO-ZVFHMIN
|
|
; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v -S -prefer-predicate-over-epilogue=scalar-epilogue | FileCheck %s -check-prefix=NO-ZVFHMIN
|
|
; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v -S -prefer-predicate-over-epilogue=predicate-else-scalar-epilogue | FileCheck %s -check-prefix=NO-ZVFHMIN
|
|
; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v,+zvfhmin -S | FileCheck %s -check-prefix=ZVFHMIN
|
|
|
|
define void @fadd(ptr noalias %a, ptr noalias %b, i64 %n) {
|
|
; NO-ZVFHMIN-LABEL: define void @fadd(
|
|
; NO-ZVFHMIN-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
|
|
; NO-ZVFHMIN-NEXT: [[ENTRY:.*]]:
|
|
; NO-ZVFHMIN-NEXT: br label %[[LOOP:.*]]
|
|
; NO-ZVFHMIN: [[LOOP]]:
|
|
; NO-ZVFHMIN-NEXT: [[I:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[I_NEXT:%.*]], %[[LOOP]] ]
|
|
; NO-ZVFHMIN-NEXT: [[A_GEP:%.*]] = getelementptr half, ptr [[A]], i64 [[I]]
|
|
; NO-ZVFHMIN-NEXT: [[B_GEP:%.*]] = getelementptr half, ptr [[B]], i64 [[I]]
|
|
; NO-ZVFHMIN-NEXT: [[X:%.*]] = load half, ptr [[A_GEP]], align 2
|
|
; NO-ZVFHMIN-NEXT: [[Y:%.*]] = load half, ptr [[B_GEP]], align 2
|
|
; NO-ZVFHMIN-NEXT: [[Z:%.*]] = fadd half [[X]], [[Y]]
|
|
; NO-ZVFHMIN-NEXT: store half [[Z]], ptr [[A_GEP]], align 2
|
|
; NO-ZVFHMIN-NEXT: [[I_NEXT]] = add i64 [[I]], 1
|
|
; NO-ZVFHMIN-NEXT: [[DONE:%.*]] = icmp eq i64 [[I_NEXT]], [[N]]
|
|
; NO-ZVFHMIN-NEXT: br i1 [[DONE]], label %[[EXIT:.*]], label %[[LOOP]]
|
|
; NO-ZVFHMIN: [[EXIT]]:
|
|
; NO-ZVFHMIN-NEXT: ret void
|
|
;
|
|
; ZVFHMIN-LABEL: define void @fadd(
|
|
; ZVFHMIN-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
|
|
; ZVFHMIN-NEXT: [[ENTRY:.*:]]
|
|
; ZVFHMIN-NEXT: br label %[[VECTOR_PH:.*]]
|
|
; ZVFHMIN: [[VECTOR_PH]]:
|
|
; ZVFHMIN-NEXT: br label %[[VECTOR_BODY:.*]]
|
|
; ZVFHMIN: [[VECTOR_BODY]]:
|
|
; ZVFHMIN-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_EVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
|
|
; ZVFHMIN-NEXT: [[AVL:%.*]] = phi i64 [ [[N]], %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
|
|
; ZVFHMIN-NEXT: [[TMP6:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 8, i1 true)
|
|
; ZVFHMIN-NEXT: [[TMP1:%.*]] = getelementptr half, ptr [[A]], i64 [[INDEX]]
|
|
; ZVFHMIN-NEXT: [[TMP2:%.*]] = getelementptr half, ptr [[B]], i64 [[INDEX]]
|
|
; ZVFHMIN-NEXT: [[WIDE_LOAD:%.*]] = call <vscale x 8 x half> @llvm.vp.load.nxv8f16.p0(ptr align 2 [[TMP1]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP6]])
|
|
; ZVFHMIN-NEXT: [[WIDE_LOAD1:%.*]] = call <vscale x 8 x half> @llvm.vp.load.nxv8f16.p0(ptr align 2 [[TMP2]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP6]])
|
|
; ZVFHMIN-NEXT: [[TMP11:%.*]] = fadd <vscale x 8 x half> [[WIDE_LOAD]], [[WIDE_LOAD1]]
|
|
; ZVFHMIN-NEXT: call void @llvm.vp.store.nxv8f16.p0(<vscale x 8 x half> [[TMP11]], ptr align 2 [[TMP1]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP6]])
|
|
; ZVFHMIN-NEXT: [[TMP13:%.*]] = zext i32 [[TMP6]] to i64
|
|
; ZVFHMIN-NEXT: [[INDEX_EVL_NEXT]] = add i64 [[TMP13]], [[INDEX]]
|
|
; ZVFHMIN-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP13]]
|
|
; ZVFHMIN-NEXT: [[TMP7:%.*]] = icmp eq i64 [[AVL_NEXT]], 0
|
|
; ZVFHMIN-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
|
|
; ZVFHMIN: [[MIDDLE_BLOCK]]:
|
|
; ZVFHMIN-NEXT: br label %[[EXIT:.*]]
|
|
; ZVFHMIN: [[EXIT]]:
|
|
; ZVFHMIN-NEXT: ret void
|
|
;
|
|
; NO-ZVFHMIN-PREDICATED-LABEL: define void @fadd(
|
|
; NO-ZVFHMIN-PREDICATED-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
|
|
; NO-ZVFHMIN-PREDICATED-NEXT: [[ENTRY:.*]]:
|
|
; NO-ZVFHMIN-PREDICATED-NEXT: br label %[[LOOP:.*]]
|
|
; NO-ZVFHMIN-PREDICATED: [[LOOP]]:
|
|
; NO-ZVFHMIN-PREDICATED-NEXT: [[I:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[I_NEXT:%.*]], %[[LOOP]] ]
|
|
; NO-ZVFHMIN-PREDICATED-NEXT: [[A_GEP:%.*]] = getelementptr half, ptr [[A]], i64 [[I]]
|
|
; NO-ZVFHMIN-PREDICATED-NEXT: [[B_GEP:%.*]] = getelementptr half, ptr [[B]], i64 [[I]]
|
|
; NO-ZVFHMIN-PREDICATED-NEXT: [[X:%.*]] = load half, ptr [[A_GEP]], align 2
|
|
; NO-ZVFHMIN-PREDICATED-NEXT: [[Y:%.*]] = load half, ptr [[B_GEP]], align 2
|
|
; NO-ZVFHMIN-PREDICATED-NEXT: [[Z:%.*]] = fadd half [[X]], [[Y]]
|
|
; NO-ZVFHMIN-PREDICATED-NEXT: store half [[Z]], ptr [[A_GEP]], align 2
|
|
; NO-ZVFHMIN-PREDICATED-NEXT: [[I_NEXT]] = add i64 [[I]], 1
|
|
; NO-ZVFHMIN-PREDICATED-NEXT: [[DONE:%.*]] = icmp eq i64 [[I_NEXT]], [[N]]
|
|
; NO-ZVFHMIN-PREDICATED-NEXT: br i1 [[DONE]], label %[[EXIT:.*]], label %[[LOOP]]
|
|
; NO-ZVFHMIN-PREDICATED: [[EXIT]]:
|
|
; NO-ZVFHMIN-PREDICATED-NEXT: ret void
|
|
entry:
|
|
br label %loop
|
|
loop:
|
|
%i = phi i64 [0, %entry], [%i.next, %loop]
|
|
%a.gep = getelementptr half, ptr %a, i64 %i
|
|
%b.gep = getelementptr half, ptr %b, i64 %i
|
|
%x = load half, ptr %a.gep
|
|
%y = load half, ptr %b.gep
|
|
%z = fadd half %x, %y
|
|
store half %z, ptr %a.gep
|
|
%i.next = add i64 %i, 1
|
|
%done = icmp eq i64 %i.next, %n
|
|
br i1 %done, label %exit, label %loop
|
|
exit:
|
|
ret void
|
|
}
|