| ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 6 |
| ; RUN: opt -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -S %s | FileCheck %s |
| |
| ; Tests for FindLastIV reductions where the selected value is an expression |
| ; of the induction variable (e.g., IV * const, IV << const, IV | const). |
| |
| define i64 @findlast_shl_expr(ptr %a, i64 %n) { |
| ; CHECK-LABEL: define i64 @findlast_shl_expr( |
| ; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*]]: |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] |
| ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] |
| ; CHECK: [[VECTOR_BODY]]: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -1), %[[VECTOR_PH]] ], [ [[TMP7:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 8 |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD]], splat (i64 42) |
| ; CHECK-NEXT: [[TMP3:%.*]] = shl <4 x i64> [[VEC_IND]], splat (i64 2) |
| ; CHECK-NEXT: [[TMP4:%.*]] = freeze <4 x i1> [[TMP2]] |
| ; CHECK-NEXT: [[TMP5:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP4]]) |
| ; CHECK-NEXT: [[TMP6]] = select i1 [[TMP5]], <4 x i1> [[TMP2]], <4 x i1> [[TMP0]] |
| ; CHECK-NEXT: [[TMP7]] = select i1 [[TMP5]], <4 x i64> [[TMP3]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP9:%.*]] = call i64 @llvm.experimental.vector.extract.last.active.v4i64(<4 x i64> [[TMP7]], <4 x i1> [[TMP6]], i64 -1) |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label %[[DONE:.*]], label %[[SCALAR_PH]] |
| ; CHECK: [[SCALAR_PH]]: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP9]], %[[MIDDLE_BLOCK]] ], [ -1, %[[ENTRY]] ] |
| ; CHECK-NEXT: br label %[[LOOP:.*]] |
| ; CHECK: [[LOOP]]: |
| ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[RDX:%.*]] = phi i64 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[SEL:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] |
| ; CHECK-NEXT: [[LOAD:%.*]] = load i64, ptr [[GEP]], align 8 |
| ; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[LOAD]], 42 |
| ; CHECK-NEXT: [[EXPR:%.*]] = shl i64 [[IV]], 2 |
| ; CHECK-NEXT: [[SEL]] = select i1 [[CMP]], i64 [[EXPR]], i64 [[RDX]] |
| ; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 |
| ; CHECK-NEXT: [[EXIT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[EXIT]], label %[[DONE]], label %[[LOOP]], !llvm.loop [[LOOP3:![0-9]+]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: [[SEL_LCSSA:%.*]] = phi i64 [ [[SEL]], %[[LOOP]] ], [ [[TMP9]], %[[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret i64 [[SEL_LCSSA]] |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] |
| %rdx = phi i64 [ -1, %entry ], [ %sel, %loop ] |
| %gep = getelementptr inbounds i64, ptr %a, i64 %iv |
| %load = load i64, ptr %gep, align 8 |
| %cmp = icmp eq i64 %load, 42 |
| %expr = shl i64 %iv, 2 |
| %sel = select i1 %cmp, i64 %expr, i64 %rdx |
| %iv.next = add nuw nsw i64 %iv, 1 |
| %exit = icmp eq i64 %iv.next, %n |
| br i1 %exit, label %done, label %loop |
| |
| done: |
| ret i64 %sel |
| } |
| |
| define i64 @findlast_mul_expr(ptr %a, i64 %n) { |
| ; CHECK-LABEL: define i64 @findlast_mul_expr( |
| ; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*]]: |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] |
| ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] |
| ; CHECK: [[VECTOR_BODY]]: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -1), %[[VECTOR_PH]] ], [ [[TMP7:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 8 |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD]], splat (i64 42) |
| ; CHECK-NEXT: [[TMP3:%.*]] = mul <4 x i64> [[VEC_IND]], splat (i64 3) |
| ; CHECK-NEXT: [[TMP4:%.*]] = freeze <4 x i1> [[TMP2]] |
| ; CHECK-NEXT: [[TMP5:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP4]]) |
| ; CHECK-NEXT: [[TMP6]] = select i1 [[TMP5]], <4 x i1> [[TMP2]], <4 x i1> [[TMP0]] |
| ; CHECK-NEXT: [[TMP7]] = select i1 [[TMP5]], <4 x i64> [[TMP3]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP9:%.*]] = call i64 @llvm.experimental.vector.extract.last.active.v4i64(<4 x i64> [[TMP7]], <4 x i1> [[TMP6]], i64 -1) |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label %[[DONE:.*]], label %[[SCALAR_PH]] |
| ; CHECK: [[SCALAR_PH]]: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP9]], %[[MIDDLE_BLOCK]] ], [ -1, %[[ENTRY]] ] |
| ; CHECK-NEXT: br label %[[LOOP:.*]] |
| ; CHECK: [[LOOP]]: |
| ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[RDX:%.*]] = phi i64 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[SEL:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] |
| ; CHECK-NEXT: [[LOAD:%.*]] = load i64, ptr [[GEP]], align 8 |
| ; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[LOAD]], 42 |
| ; CHECK-NEXT: [[EXPR:%.*]] = mul i64 [[IV]], 3 |
| ; CHECK-NEXT: [[SEL]] = select i1 [[CMP]], i64 [[EXPR]], i64 [[RDX]] |
| ; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 |
| ; CHECK-NEXT: [[EXIT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[EXIT]], label %[[DONE]], label %[[LOOP]], !llvm.loop [[LOOP5:![0-9]+]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: [[SEL_LCSSA:%.*]] = phi i64 [ [[SEL]], %[[LOOP]] ], [ [[TMP9]], %[[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret i64 [[SEL_LCSSA]] |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] |
| %rdx = phi i64 [ -1, %entry ], [ %sel, %loop ] |
| %gep = getelementptr inbounds i64, ptr %a, i64 %iv |
| %load = load i64, ptr %gep, align 8 |
| %cmp = icmp eq i64 %load, 42 |
| %expr = mul i64 %iv, 3 |
| %sel = select i1 %cmp, i64 %expr, i64 %rdx |
| %iv.next = add nuw nsw i64 %iv, 1 |
| %exit = icmp eq i64 %iv.next, %n |
| br i1 %exit, label %done, label %loop |
| |
| done: |
| ret i64 %sel |
| } |
| |
| ; Negative test: the mul operand is loop-varying (a load), so expression |
| ; sinking should NOT apply. Uses extract-last-active instead. |
| define i64 @findlast_mul_loop_varying_operand(ptr %a, ptr %b, i64 %n) { |
| ; CHECK-LABEL: define i64 @findlast_mul_loop_varying_operand( |
| ; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*]]: |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] |
| ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] |
| ; CHECK: [[VECTOR_BODY]]: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -1), %[[VECTOR_PH]] ], [ [[TMP7:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 8 |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD]], splat (i64 42) |
| ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <4 x i64>, ptr [[TMP3]], align 8 |
| ; CHECK-NEXT: [[EXPR:%.*]] = mul <4 x i64> [[VEC_IND]], [[WIDE_LOAD1]] |
| ; CHECK-NEXT: [[TMP4:%.*]] = freeze <4 x i1> [[TMP2]] |
| ; CHECK-NEXT: [[TMP5:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP4]]) |
| ; CHECK-NEXT: [[TMP6]] = select i1 [[TMP5]], <4 x i1> [[TMP2]], <4 x i1> [[TMP0]] |
| ; CHECK-NEXT: [[TMP7]] = select i1 [[TMP5]], <4 x i64> [[EXPR]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP9:%.*]] = call i64 @llvm.experimental.vector.extract.last.active.v4i64(<4 x i64> [[TMP7]], <4 x i1> [[TMP6]], i64 -1) |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label %[[DONE:.*]], label %[[SCALAR_PH]] |
| ; CHECK: [[SCALAR_PH]]: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP9]], %[[MIDDLE_BLOCK]] ], [ -1, %[[ENTRY]] ] |
| ; CHECK-NEXT: br label %[[LOOP:.*]] |
| ; CHECK: [[LOOP]]: |
| ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[RDX:%.*]] = phi i64 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[SEL:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[GEP_A:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] |
| ; CHECK-NEXT: [[LOAD_A:%.*]] = load i64, ptr [[GEP_A]], align 8 |
| ; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[LOAD_A]], 42 |
| ; CHECK-NEXT: [[GEP_B:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[IV]] |
| ; CHECK-NEXT: [[LOAD_B:%.*]] = load i64, ptr [[GEP_B]], align 8 |
| ; CHECK-NEXT: [[EXPR1:%.*]] = mul i64 [[IV]], [[LOAD_B]] |
| ; CHECK-NEXT: [[SEL]] = select i1 [[CMP]], i64 [[EXPR1]], i64 [[RDX]] |
| ; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 |
| ; CHECK-NEXT: [[EXIT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[EXIT]], label %[[DONE]], label %[[LOOP]], !llvm.loop [[LOOP7:![0-9]+]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: [[SEL_LCSSA:%.*]] = phi i64 [ [[SEL]], %[[LOOP]] ], [ [[TMP9]], %[[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret i64 [[SEL_LCSSA]] |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] |
| %rdx = phi i64 [ -1, %entry ], [ %sel, %loop ] |
| %gep.a = getelementptr inbounds i64, ptr %a, i64 %iv |
| %load.a = load i64, ptr %gep.a, align 8 |
| %cmp = icmp eq i64 %load.a, 42 |
| %gep.b = getelementptr inbounds i64, ptr %b, i64 %iv |
| %load.b = load i64, ptr %gep.b, align 8 |
| ; IV multiplied by a loop-varying value (load) - should NOT sink expression |
| %expr = mul i64 %iv, %load.b |
| %sel = select i1 %cmp, i64 %expr, i64 %rdx |
| %iv.next = add nuw nsw i64 %iv, 1 |
| %exit = icmp eq i64 %iv.next, %n |
| br i1 %exit, label %done, label %loop |
| |
| done: |
| ret i64 %sel |
| } |
| |
| define i64 @findlast_or_expr(ptr %a, i64 %n) { |
| ; CHECK-LABEL: define i64 @findlast_or_expr( |
| ; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*]]: |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] |
| ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] |
| ; CHECK: [[VECTOR_BODY]]: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -1), %[[VECTOR_PH]] ], [ [[TMP7:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 8 |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD]], splat (i64 42) |
| ; CHECK-NEXT: [[TMP3:%.*]] = or <4 x i64> [[VEC_IND]], splat (i64 1) |
| ; CHECK-NEXT: [[TMP4:%.*]] = freeze <4 x i1> [[TMP2]] |
| ; CHECK-NEXT: [[TMP5:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP4]]) |
| ; CHECK-NEXT: [[TMP6]] = select i1 [[TMP5]], <4 x i1> [[TMP2]], <4 x i1> [[TMP0]] |
| ; CHECK-NEXT: [[TMP7]] = select i1 [[TMP5]], <4 x i64> [[TMP3]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP9:%.*]] = call i64 @llvm.experimental.vector.extract.last.active.v4i64(<4 x i64> [[TMP7]], <4 x i1> [[TMP6]], i64 -1) |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label %[[DONE:.*]], label %[[SCALAR_PH]] |
| ; CHECK: [[SCALAR_PH]]: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP9]], %[[MIDDLE_BLOCK]] ], [ -1, %[[ENTRY]] ] |
| ; CHECK-NEXT: br label %[[LOOP:.*]] |
| ; CHECK: [[LOOP]]: |
| ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[RDX:%.*]] = phi i64 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[SEL:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] |
| ; CHECK-NEXT: [[LOAD:%.*]] = load i64, ptr [[GEP]], align 8 |
| ; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[LOAD]], 42 |
| ; CHECK-NEXT: [[EXPR:%.*]] = or i64 [[IV]], 1 |
| ; CHECK-NEXT: [[SEL]] = select i1 [[CMP]], i64 [[EXPR]], i64 [[RDX]] |
| ; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 |
| ; CHECK-NEXT: [[EXIT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[EXIT]], label %[[DONE]], label %[[LOOP]], !llvm.loop [[LOOP9:![0-9]+]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: [[SEL_LCSSA:%.*]] = phi i64 [ [[SEL]], %[[LOOP]] ], [ [[TMP9]], %[[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret i64 [[SEL_LCSSA]] |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] |
| %rdx = phi i64 [ -1, %entry ], [ %sel, %loop ] |
| %gep = getelementptr inbounds i64, ptr %a, i64 %iv |
| %load = load i64, ptr %gep, align 8 |
| %cmp = icmp eq i64 %load, 42 |
| %expr = or i64 %iv, 1 |
| %sel = select i1 %cmp, i64 %expr, i64 %rdx |
| %iv.next = add nuw nsw i64 %iv, 1 |
| %exit = icmp eq i64 %iv.next, %n |
| br i1 %exit, label %done, label %loop |
| |
| done: |
| ret i64 %sel |
| } |
| |
| ; Test non-canonical IV (starting at 10, step 2) with expression sinking. |
| ; The IV is non-canonical but expression sinking should still apply. |
| define i64 @findlast_non_canonical_iv_with_expr(ptr %a, i64 %n) { |
| ; CHECK-LABEL: define i64 @findlast_non_canonical_iv_with_expr( |
| ; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*]]: |
| ; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N]], -12 |
| ; CHECK-NEXT: [[TMP1:%.*]] = lshr i64 [[TMP0]], 1 |
| ; CHECK-NEXT: [[TMP2:%.*]] = add nuw i64 [[TMP1]], 1 |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[N_VEC]], 2 |
| ; CHECK-NEXT: [[TMP4:%.*]] = add i64 10, [[TMP3]] |
| ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] |
| ; CHECK: [[VECTOR_BODY]]: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IND1:%.*]] = phi <4 x i64> [ <i64 10, i64 12, i64 14, i64 16>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP23:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP5:%.*]] = mul i64 [[INDEX]], 2 |
| ; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i64 10, [[TMP5]] |
| ; CHECK-NEXT: [[TMP7:%.*]] = add i64 [[OFFSET_IDX]], 2 |
| ; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[OFFSET_IDX]], 4 |
| ; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[OFFSET_IDX]], 6 |
| ; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[OFFSET_IDX]] |
| ; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP7]] |
| ; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP8]] |
| ; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP9]] |
| ; CHECK-NEXT: [[TMP14:%.*]] = load i64, ptr [[TMP10]], align 8 |
| ; CHECK-NEXT: [[TMP15:%.*]] = load i64, ptr [[TMP11]], align 8 |
| ; CHECK-NEXT: [[TMP16:%.*]] = load i64, ptr [[TMP12]], align 8 |
| ; CHECK-NEXT: [[TMP17:%.*]] = load i64, ptr [[TMP13]], align 8 |
| ; CHECK-NEXT: [[TMP18:%.*]] = insertelement <4 x i64> poison, i64 [[TMP14]], i32 0 |
| ; CHECK-NEXT: [[TMP19:%.*]] = insertelement <4 x i64> [[TMP18]], i64 [[TMP15]], i32 1 |
| ; CHECK-NEXT: [[TMP20:%.*]] = insertelement <4 x i64> [[TMP19]], i64 [[TMP16]], i32 2 |
| ; CHECK-NEXT: [[TMP21:%.*]] = insertelement <4 x i64> [[TMP20]], i64 [[TMP17]], i32 3 |
| ; CHECK-NEXT: [[TMP22:%.*]] = icmp eq <4 x i64> [[TMP21]], splat (i64 42) |
| ; CHECK-NEXT: [[VEC_IND:%.*]] = add <4 x i64> [[VEC_IND1]], splat (i64 100) |
| ; CHECK-NEXT: [[TMP23]] = select <4 x i1> [[TMP22]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i64> [[VEC_IND1]], splat (i64 8) |
| ; CHECK-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP24]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP27:%.*]] = call i64 @llvm.vector.reduce.umax.v4i64(<4 x i64> [[TMP23]]) |
| ; CHECK-NEXT: [[TMP26:%.*]] = icmp ne i64 [[TMP27]], 0 |
| ; CHECK-NEXT: [[TMP28:%.*]] = select i1 [[TMP26]], i64 [[TMP27]], i64 -1 |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label %[[DONE:.*]], label %[[SCALAR_PH]] |
| ; CHECK: [[SCALAR_PH]]: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP4]], %[[MIDDLE_BLOCK]] ], [ 10, %[[ENTRY]] ] |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP28]], %[[MIDDLE_BLOCK]] ], [ -1, %[[ENTRY]] ] |
| ; CHECK-NEXT: br label %[[LOOP:.*]] |
| ; CHECK: [[LOOP]]: |
| ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[RDX:%.*]] = phi i64 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[SEL:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] |
| ; CHECK-NEXT: [[LOAD:%.*]] = load i64, ptr [[GEP]], align 8 |
| ; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[LOAD]], 42 |
| ; CHECK-NEXT: [[EXPR:%.*]] = add i64 [[IV]], 100 |
| ; CHECK-NEXT: [[SEL]] = select i1 [[CMP]], i64 [[EXPR]], i64 [[RDX]] |
| ; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 2 |
| ; CHECK-NEXT: [[EXIT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[EXIT]], label %[[DONE]], label %[[LOOP]], !llvm.loop [[LOOP11:![0-9]+]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: [[SEL_LCSSA:%.*]] = phi i64 [ [[SEL]], %[[LOOP]] ], [ [[TMP28]], %[[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret i64 [[SEL_LCSSA]] |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %iv = phi i64 [ 10, %entry ], [ %iv.next, %loop ] |
| %rdx = phi i64 [ -1, %entry ], [ %sel, %loop ] |
| %gep = getelementptr inbounds i64, ptr %a, i64 %iv |
| %load = load i64, ptr %gep, align 8 |
| %cmp = icmp eq i64 %load, 42 |
| %expr = add i64 %iv, 100 |
| %sel = select i1 %cmp, i64 %expr, i64 %rdx |
| %iv.next = add nuw nsw i64 %iv, 2 |
| %exit = icmp eq i64 %iv.next, %n |
| br i1 %exit, label %done, label %loop |
| |
| done: |
| ret i64 %sel |
| } |
| |
| ; Test wider IV (i64) with narrower reduction (i32), only scalar users. |
| ; Expression sinking should NOT apply because introducing wider computations |
| ; in the middle block would be slower when the wide IV can be eliminated. |
| define i32 @findlast_wider_iv_only_scalar_users(ptr %a, i64 %n) { |
| ; CHECK-LABEL: define i32 @findlast_wider_iv_only_scalar_users( |
| ; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*]]: |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] |
| ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] |
| ; CHECK: [[VECTOR_BODY]]: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ splat (i32 -1), %[[VECTOR_PH]] ], [ [[TMP7:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP1]], align 4 |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <4 x i32> [[WIDE_LOAD]], splat (i32 42) |
| ; CHECK-NEXT: [[TMP3:%.*]] = shl <4 x i32> [[VEC_IND]], splat (i32 2) |
| ; CHECK-NEXT: [[TMP4:%.*]] = freeze <4 x i1> [[TMP2]] |
| ; CHECK-NEXT: [[TMP5:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP4]]) |
| ; CHECK-NEXT: [[TMP6]] = select i1 [[TMP5]], <4 x i1> [[TMP2]], <4 x i1> [[TMP0]] |
| ; CHECK-NEXT: [[TMP7]] = select i1 [[TMP5]], <4 x i32> [[TMP3]], <4 x i32> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], splat (i32 4) |
| ; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP9:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP7]], <4 x i1> [[TMP6]], i32 -1) |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label %[[DONE:.*]], label %[[SCALAR_PH]] |
| ; CHECK: [[SCALAR_PH]]: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP9]], %[[MIDDLE_BLOCK]] ], [ -1, %[[ENTRY]] ] |
| ; CHECK-NEXT: br label %[[LOOP:.*]] |
| ; CHECK: [[LOOP]]: |
| ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[RDX:%.*]] = phi i32 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[SEL:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV]] |
| ; CHECK-NEXT: [[LOAD:%.*]] = load i32, ptr [[GEP]], align 4 |
| ; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[LOAD]], 42 |
| ; CHECK-NEXT: [[IV_TRUNC:%.*]] = trunc i64 [[IV]] to i32 |
| ; CHECK-NEXT: [[EXPR:%.*]] = shl i32 [[IV_TRUNC]], 2 |
| ; CHECK-NEXT: [[SEL]] = select i1 [[CMP]], i32 [[EXPR]], i32 [[RDX]] |
| ; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 |
| ; CHECK-NEXT: [[EXIT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[EXIT]], label %[[DONE]], label %[[LOOP]], !llvm.loop [[LOOP13:![0-9]+]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: [[SEL_LCSSA:%.*]] = phi i32 [ [[SEL]], %[[LOOP]] ], [ [[TMP9]], %[[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret i32 [[SEL_LCSSA]] |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] |
| %rdx = phi i32 [ -1, %entry ], [ %sel, %loop ] |
| %gep = getelementptr inbounds i32, ptr %a, i64 %iv |
| %load = load i32, ptr %gep, align 4 |
| %cmp = icmp eq i32 %load, 42 |
| %iv.trunc = trunc i64 %iv to i32 |
| %expr = shl i32 %iv.trunc, 2 |
| %sel = select i1 %cmp, i32 %expr, i32 %rdx |
| %iv.next = add nuw nsw i64 %iv, 1 |
| %exit = icmp eq i64 %iv.next, %n |
| br i1 %exit, label %done, label %loop |
| |
| done: |
| ret i32 %sel |
| } |
| |
| ; Test wider IV (i64) with narrower reduction (i32), with vector users. |
| ; Expression sinking SHOULD apply because the wide IV has vector users |
| ; (the mul that feeds into both the store and the reduction), so it will |
| ; stay wide anyway. |
| define i32 @findlast_wider_iv_with_vector_users(ptr %a, ptr %b, i64 %n) { |
| ; CHECK-LABEL: define i32 @findlast_wider_iv_with_vector_users( |
| ; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*]]: |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK:.*]] |
| ; CHECK: [[VECTOR_MEMCHECK]]: |
| ; CHECK-NEXT: [[TMP0:%.*]] = shl i64 [[N]], 3 |
| ; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP0]] |
| ; CHECK-NEXT: [[TMP1:%.*]] = shl i64 [[N]], 2 |
| ; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP1]] |
| ; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[B]], [[SCEVGEP1]] |
| ; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[A]], [[SCEVGEP]] |
| ; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] |
| ; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] |
| ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] |
| ; CHECK: [[VECTOR_BODY]]: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ splat (i32 -1), %[[VECTOR_PH]] ], [ [[TMP11:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP2:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP10:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IND2:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT3:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP3]], align 4, !alias.scope [[META14:![0-9]+]] |
| ; CHECK-NEXT: [[TMP4:%.*]] = icmp eq <4 x i32> [[WIDE_LOAD]], splat (i32 42) |
| ; CHECK-NEXT: [[TMP5:%.*]] = mul <4 x i64> [[VEC_IND]], splat (i64 3) |
| ; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[INDEX]] |
| ; CHECK-NEXT: store <4 x i64> [[TMP5]], ptr [[TMP6]], align 8, !alias.scope [[META17:![0-9]+]], !noalias [[META14]] |
| ; CHECK-NEXT: [[TMP7:%.*]] = add <4 x i32> [[VEC_IND2]], splat (i32 100) |
| ; CHECK-NEXT: [[TMP8:%.*]] = freeze <4 x i1> [[TMP4]] |
| ; CHECK-NEXT: [[TMP9:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP8]]) |
| ; CHECK-NEXT: [[TMP10]] = select i1 [[TMP9]], <4 x i1> [[TMP4]], <4 x i1> [[TMP2]] |
| ; CHECK-NEXT: [[TMP11]] = select i1 [[TMP9]], <4 x i32> [[TMP7]], <4 x i32> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[VEC_IND_NEXT3]] = add <4 x i32> [[VEC_IND2]], splat (i32 4) |
| ; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP12]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP19:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP13:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP11]], <4 x i1> [[TMP10]], i32 -1) |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label %[[DONE:.*]], label %[[SCALAR_PH]] |
| ; CHECK: [[SCALAR_PH]]: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ], [ 0, %[[VECTOR_MEMCHECK]] ] |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP13]], %[[MIDDLE_BLOCK]] ], [ -1, %[[ENTRY]] ], [ -1, %[[VECTOR_MEMCHECK]] ] |
| ; CHECK-NEXT: br label %[[LOOP:.*]] |
| ; CHECK: [[LOOP]]: |
| ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[RDX:%.*]] = phi i32 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[SEL:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[GEP_A:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV]] |
| ; CHECK-NEXT: [[LOAD:%.*]] = load i32, ptr [[GEP_A]], align 4 |
| ; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[LOAD]], 42 |
| ; CHECK-NEXT: [[IV_MUL:%.*]] = mul i64 [[IV]], 3 |
| ; CHECK-NEXT: [[GEP_B:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[IV]] |
| ; CHECK-NEXT: store i64 [[IV_MUL]], ptr [[GEP_B]], align 8 |
| ; CHECK-NEXT: [[IV_TRUNC:%.*]] = trunc i64 [[IV]] to i32 |
| ; CHECK-NEXT: [[EXPR:%.*]] = add i32 [[IV_TRUNC]], 100 |
| ; CHECK-NEXT: [[SEL]] = select i1 [[CMP]], i32 [[EXPR]], i32 [[RDX]] |
| ; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 |
| ; CHECK-NEXT: [[EXIT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[EXIT]], label %[[DONE]], label %[[LOOP]], !llvm.loop [[LOOP20:![0-9]+]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: [[SEL_LCSSA:%.*]] = phi i32 [ [[SEL]], %[[LOOP]] ], [ [[TMP13]], %[[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret i32 [[SEL_LCSSA]] |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] |
| %rdx = phi i32 [ -1, %entry ], [ %sel, %loop ] |
| %gep.a = getelementptr inbounds i32, ptr %a, i64 %iv |
| %load = load i32, ptr %gep.a, align 4 |
| %cmp = icmp eq i32 %load, 42 |
| ; This mul uses the wide IV and feeds into a store (vector user) |
| %iv.mul = mul i64 %iv, 3 |
| %gep.b = getelementptr inbounds i64, ptr %b, i64 %iv |
| store i64 %iv.mul, ptr %gep.b, align 8 |
| %iv.trunc = trunc i64 %iv to i32 |
| %expr = add i32 %iv.trunc, 100 |
| %sel = select i1 %cmp, i32 %expr, i32 %rdx |
| %iv.next = add nuw nsw i64 %iv, 1 |
| %exit = icmp eq i64 %iv.next, %n |
| br i1 %exit, label %done, label %loop |
| |
| done: |
| ret i32 %sel |
| } |
| |
| ; Negative test: i8 reduction with IV expression. Should NOT create a wide (i64) |
| ; IV in the vectorized code - the IV should be narrowed to i8 since only scalar |
| ; i8 values are needed from it. |
| define i8 @findlast_narrow_i8_reduction(ptr %a, i64 %n) { |
| ; CHECK-LABEL: define i8 @findlast_narrow_i8_reduction( |
| ; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*]]: |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] |
| ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] |
| ; CHECK: [[VECTOR_BODY]]: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i8> [ splat (i8 -1), %[[VECTOR_PH]] ], [ [[TMP7:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i8> [ <i8 0, i8 1, i8 2, i8 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, ptr [[TMP1]], align 1 |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <4 x i8> [[WIDE_LOAD]], splat (i8 42) |
| ; CHECK-NEXT: [[TMP3:%.*]] = add <4 x i8> [[VEC_IND]], splat (i8 10) |
| ; CHECK-NEXT: [[TMP4:%.*]] = freeze <4 x i1> [[TMP2]] |
| ; CHECK-NEXT: [[TMP5:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP4]]) |
| ; CHECK-NEXT: [[TMP6]] = select i1 [[TMP5]], <4 x i1> [[TMP2]], <4 x i1> [[TMP0]] |
| ; CHECK-NEXT: [[TMP7]] = select i1 [[TMP5]], <4 x i8> [[TMP3]], <4 x i8> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i8> [[VEC_IND]], splat (i8 4) |
| ; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP21:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP9:%.*]] = call i8 @llvm.experimental.vector.extract.last.active.v4i8(<4 x i8> [[TMP7]], <4 x i1> [[TMP6]], i8 -1) |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label %[[DONE:.*]], label %[[SCALAR_PH]] |
| ; CHECK: [[SCALAR_PH]]: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i8 [ [[TMP9]], %[[MIDDLE_BLOCK]] ], [ -1, %[[ENTRY]] ] |
| ; CHECK-NEXT: br label %[[LOOP:.*]] |
| ; CHECK: [[LOOP]]: |
| ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[RDX:%.*]] = phi i8 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[SEL:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[IV]] |
| ; CHECK-NEXT: [[LOAD:%.*]] = load i8, ptr [[GEP]], align 1 |
| ; CHECK-NEXT: [[CMP:%.*]] = icmp eq i8 [[LOAD]], 42 |
| ; CHECK-NEXT: [[IV_TRUNC:%.*]] = trunc i64 [[IV]] to i8 |
| ; CHECK-NEXT: [[EXPR:%.*]] = add i8 [[IV_TRUNC]], 10 |
| ; CHECK-NEXT: [[SEL]] = select i1 [[CMP]], i8 [[EXPR]], i8 [[RDX]] |
| ; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 |
| ; CHECK-NEXT: [[EXIT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[EXIT]], label %[[DONE]], label %[[LOOP]], !llvm.loop [[LOOP22:![0-9]+]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: [[SEL_LCSSA:%.*]] = phi i8 [ [[SEL]], %[[LOOP]] ], [ [[TMP9]], %[[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret i8 [[SEL_LCSSA]] |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] |
| %rdx = phi i8 [ -1, %entry ], [ %sel, %loop ] |
| %gep = getelementptr inbounds i8, ptr %a, i64 %iv |
| %load = load i8, ptr %gep, align 1 |
| %cmp = icmp eq i8 %load, 42 |
| %iv.trunc = trunc i64 %iv to i8 |
| %expr = add i8 %iv.trunc, 10 |
| %sel = select i1 %cmp, i8 %expr, i8 %rdx |
| %iv.next = add nuw nsw i64 %iv, 1 |
| %exit = icmp eq i64 %iv.next, %n |
| br i1 %exit, label %done, label %loop |
| |
| done: |
| ret i8 %sel |
| } |
