| ; 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 -tail-folding-policy=prefer-fold-tail -force-target-supports-masked-memory-ops -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: br label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 [[N]], 3 |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = and i64 [[N_RND_UP]], 3 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[N]], 1 |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer |
| ; 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 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = icmp ule <4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <4 x i64> @llvm.masked.load.v4i64.p0(ptr align 8 [[TMP1]], <4 x i1> [[TMP0]], <4 x i64> poison) |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <4 x i64> [[WIDE_MASKED_LOAD]], splat (i64 42) |
| ; CHECK-NEXT: [[TMP9:%.*]] = select <4 x i1> [[TMP0]], <4 x i1> [[TMP2]], <4 x i1> zeroinitializer |
| ; CHECK-NEXT: [[TMP3]] = select <4 x i1> [[TMP9]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP5:%.*]] = call i64 @llvm.vector.reduce.smax.v4i64(<4 x i64> [[TMP3]]) |
| ; CHECK-NEXT: [[TMP6:%.*]] = shl i64 [[TMP5]], 2 |
| ; CHECK-NEXT: [[TMP7:%.*]] = icmp ne i64 [[TMP5]], -9223372036854775808 |
| ; CHECK-NEXT: [[TMP8:%.*]] = select i1 [[TMP7]], i64 [[TMP6]], i64 -1 |
| ; CHECK-NEXT: br label %[[DONE:.*]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: ret i64 [[TMP8]] |
| ; |
| 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: br label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 [[N]], 3 |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = and i64 [[N_RND_UP]], 3 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[N]], 1 |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer |
| ; 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 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = icmp ule <4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <4 x i64> @llvm.masked.load.v4i64.p0(ptr align 8 [[TMP1]], <4 x i1> [[TMP0]], <4 x i64> poison) |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <4 x i64> [[WIDE_MASKED_LOAD]], splat (i64 42) |
| ; CHECK-NEXT: [[TMP9:%.*]] = select <4 x i1> [[TMP0]], <4 x i1> [[TMP2]], <4 x i1> zeroinitializer |
| ; CHECK-NEXT: [[TMP3]] = select <4 x i1> [[TMP9]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP5:%.*]] = call i64 @llvm.vector.reduce.smax.v4i64(<4 x i64> [[TMP3]]) |
| ; CHECK-NEXT: [[TMP6:%.*]] = mul i64 [[TMP5]], 3 |
| ; CHECK-NEXT: [[TMP7:%.*]] = icmp ne i64 [[TMP5]], -9223372036854775808 |
| ; CHECK-NEXT: [[TMP8:%.*]] = select i1 [[TMP7]], i64 [[TMP6]], i64 -1 |
| ; CHECK-NEXT: br label %[[DONE:.*]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: ret i64 [[TMP8]] |
| ; |
| 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 |
| } |
| |
| 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: br label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 [[N]], 3 |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = and i64 [[N_RND_UP]], 3 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[N]], 1 |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer |
| ; 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 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = icmp ule <4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <4 x i64> @llvm.masked.load.v4i64.p0(ptr align 8 [[TMP1]], <4 x i1> [[TMP0]], <4 x i64> poison) |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <4 x i64> [[WIDE_MASKED_LOAD]], splat (i64 42) |
| ; CHECK-NEXT: [[TMP9:%.*]] = select <4 x i1> [[TMP0]], <4 x i1> [[TMP2]], <4 x i1> zeroinitializer |
| ; CHECK-NEXT: [[TMP3]] = select <4 x i1> [[TMP9]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP5:%.*]] = call i64 @llvm.vector.reduce.smax.v4i64(<4 x i64> [[TMP3]]) |
| ; CHECK-NEXT: [[TMP6:%.*]] = or i64 [[TMP5]], 1 |
| ; CHECK-NEXT: [[TMP7:%.*]] = icmp ne i64 [[TMP5]], -9223372036854775808 |
| ; CHECK-NEXT: [[TMP8:%.*]] = select i1 [[TMP7]], i64 [[TMP6]], i64 -1 |
| ; CHECK-NEXT: br label %[[DONE:.*]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: ret i64 [[TMP8]] |
| ; |
| 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: br label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 [[TMP2]], 3 |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = and i64 [[N_RND_UP]], 3 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[TMP2]], 1 |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer |
| ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] |
| ; CHECK: [[VECTOR_BODY]]: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_LOAD_CONTINUE8:.*]] ] |
| ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 10, i64 12, i64 14, i64 16>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_LOAD_CONTINUE8]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP36:%.*]], %[[PRED_LOAD_CONTINUE8]] ] |
| ; CHECK-NEXT: [[VEC_IV:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT8:%.*]], %[[PRED_LOAD_CONTINUE8]] ] |
| ; CHECK-NEXT: [[TMP4:%.*]] = icmp ule <4 x i64> [[VEC_IV]], [[BROADCAST_SPLAT]] |
| ; CHECK-NEXT: [[TMP6:%.*]] = shl nuw i64 [[INDEX]], 1 |
| ; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add nuw i64 10, [[TMP6]] |
| ; CHECK-NEXT: [[TMP5:%.*]] = extractelement <4 x i1> [[TMP4]], i64 0 |
| ; CHECK-NEXT: br i1 [[TMP5]], label %[[PRED_LOAD_IF:.*]], label %[[PRED_LOAD_CONTINUE:.*]] |
| ; CHECK: [[PRED_LOAD_IF]]: |
| ; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[OFFSET_IDX]] |
| ; CHECK-NEXT: [[TMP8:%.*]] = load i64, ptr [[TMP7]], align 8 |
| ; CHECK-NEXT: [[TMP9:%.*]] = insertelement <4 x i64> poison, i64 [[TMP8]], i64 0 |
| ; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE]] |
| ; CHECK: [[PRED_LOAD_CONTINUE]]: |
| ; CHECK-NEXT: [[TMP10:%.*]] = phi <4 x i64> [ poison, %[[VECTOR_BODY]] ], [ [[TMP9]], %[[PRED_LOAD_IF]] ] |
| ; CHECK-NEXT: [[TMP11:%.*]] = extractelement <4 x i1> [[TMP4]], i64 1 |
| ; CHECK-NEXT: br i1 [[TMP11]], label %[[PRED_LOAD_IF3:.*]], label %[[PRED_LOAD_CONTINUE4:.*]] |
| ; CHECK: [[PRED_LOAD_IF3]]: |
| ; CHECK-NEXT: [[TMP12:%.*]] = add i64 [[OFFSET_IDX]], 2 |
| ; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP12]] |
| ; CHECK-NEXT: [[TMP14:%.*]] = load i64, ptr [[TMP13]], align 8 |
| ; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x i64> [[TMP10]], i64 [[TMP14]], i64 1 |
| ; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE4]] |
| ; CHECK: [[PRED_LOAD_CONTINUE4]]: |
| ; CHECK-NEXT: [[TMP16:%.*]] = phi <4 x i64> [ [[TMP10]], %[[PRED_LOAD_CONTINUE]] ], [ [[TMP15]], %[[PRED_LOAD_IF3]] ] |
| ; CHECK-NEXT: [[TMP17:%.*]] = extractelement <4 x i1> [[TMP4]], i64 2 |
| ; CHECK-NEXT: br i1 [[TMP17]], label %[[PRED_LOAD_IF5:.*]], label %[[PRED_LOAD_CONTINUE6:.*]] |
| ; CHECK: [[PRED_LOAD_IF5]]: |
| ; CHECK-NEXT: [[TMP18:%.*]] = add i64 [[OFFSET_IDX]], 4 |
| ; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP18]] |
| ; CHECK-NEXT: [[TMP20:%.*]] = load i64, ptr [[TMP19]], align 8 |
| ; CHECK-NEXT: [[TMP21:%.*]] = insertelement <4 x i64> [[TMP16]], i64 [[TMP20]], i64 2 |
| ; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE6]] |
| ; CHECK: [[PRED_LOAD_CONTINUE6]]: |
| ; CHECK-NEXT: [[TMP22:%.*]] = phi <4 x i64> [ [[TMP16]], %[[PRED_LOAD_CONTINUE4]] ], [ [[TMP21]], %[[PRED_LOAD_IF5]] ] |
| ; CHECK-NEXT: [[TMP23:%.*]] = extractelement <4 x i1> [[TMP4]], i64 3 |
| ; CHECK-NEXT: br i1 [[TMP23]], label %[[PRED_LOAD_IF7:.*]], label %[[PRED_LOAD_CONTINUE8]] |
| ; CHECK: [[PRED_LOAD_IF7]]: |
| ; CHECK-NEXT: [[TMP24:%.*]] = add i64 [[OFFSET_IDX]], 6 |
| ; CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP24]] |
| ; CHECK-NEXT: [[TMP26:%.*]] = load i64, ptr [[TMP25]], align 8 |
| ; CHECK-NEXT: [[TMP27:%.*]] = insertelement <4 x i64> [[TMP22]], i64 [[TMP26]], i64 3 |
| ; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE8]] |
| ; CHECK: [[PRED_LOAD_CONTINUE8]]: |
| ; CHECK-NEXT: [[TMP28:%.*]] = phi <4 x i64> [ [[TMP22]], %[[PRED_LOAD_CONTINUE6]] ], [ [[TMP27]], %[[PRED_LOAD_IF7]] ] |
| ; CHECK-NEXT: [[TMP29:%.*]] = icmp eq <4 x i64> [[TMP28]], splat (i64 42) |
| ; CHECK-NEXT: [[TMP30:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 100) |
| ; CHECK-NEXT: [[TMP31:%.*]] = select <4 x i1> [[TMP4]], <4 x i1> [[TMP29]], <4 x i1> zeroinitializer |
| ; CHECK-NEXT: [[TMP36]] = select <4 x i1> [[TMP31]], <4 x i64> [[TMP30]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i64> [[VEC_IND]], splat (i64 8) |
| ; CHECK-NEXT: [[VEC_IND_NEXT8]] = add nuw <4 x i64> [[VEC_IV]], splat (i64 4) |
| ; CHECK-NEXT: [[TMP32:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP32]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP33:%.*]] = call i64 @llvm.vector.reduce.umax.v4i64(<4 x i64> [[TMP36]]) |
| ; CHECK-NEXT: [[TMP34:%.*]] = icmp ne i64 [[TMP33]], 0 |
| ; CHECK-NEXT: [[TMP35:%.*]] = select i1 [[TMP34]], i64 [[TMP33]], i64 -1 |
| ; CHECK-NEXT: br label %[[DONE:.*]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: ret i64 [[TMP35]] |
| ; |
| 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: br label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 [[N]], 3 |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = and i64 [[N_RND_UP]], 3 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[N]], 1 |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer |
| ; 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]] ], [ [[TMP9:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP8:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IND1:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT2:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT3:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP1:%.*]] = icmp ule <4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] |
| ; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <4 x i32> @llvm.masked.load.v4i32.p0(ptr align 4 [[TMP2]], <4 x i1> [[TMP1]], <4 x i32> poison) |
| ; CHECK-NEXT: [[TMP3:%.*]] = icmp eq <4 x i32> [[WIDE_MASKED_LOAD]], splat (i32 42) |
| ; CHECK-NEXT: [[TMP4:%.*]] = shl <4 x i32> [[VEC_IND1]], splat (i32 2) |
| ; CHECK-NEXT: [[TMP5:%.*]] = select <4 x i1> [[TMP1]], <4 x i1> [[TMP3]], <4 x i1> zeroinitializer |
| ; CHECK-NEXT: [[TMP6:%.*]] = freeze <4 x i1> [[TMP5]] |
| ; CHECK-NEXT: [[TMP7:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP6]]) |
| ; CHECK-NEXT: [[TMP8]] = select i1 [[TMP7]], <4 x i1> [[TMP6]], <4 x i1> [[TMP0]] |
| ; CHECK-NEXT: [[TMP9]] = select i1 [[TMP7]], <4 x i32> [[TMP4]], <4 x i32> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT2]] = add <4 x i32> [[VEC_IND1]], splat (i32 4) |
| ; CHECK-NEXT: [[VEC_IND_NEXT3]] = add nuw <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP11:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP9]], <4 x i1> [[TMP8]], i32 -1) |
| ; CHECK-NEXT: br label %[[DONE:.*]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: ret i32 [[TMP11]] |
| ; |
| 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: br 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_RND_UP:%.*]] = add i64 [[N]], 3 |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = and i64 [[N_RND_UP]], 3 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[N]], 1 |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer |
| ; 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]] ], [ [[TMP13:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP2:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP12:%.*]], %[[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:%.*]] = icmp ule <4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] |
| ; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <4 x i32> @llvm.masked.load.v4i32.p0(ptr align 4 [[TMP4]], <4 x i1> [[TMP3]], <4 x i32> poison), !alias.scope [[META7:![0-9]+]] |
| ; CHECK-NEXT: [[TMP5:%.*]] = icmp eq <4 x i32> [[WIDE_MASKED_LOAD]], splat (i32 42) |
| ; CHECK-NEXT: [[TMP6:%.*]] = mul <4 x i64> [[VEC_IND]], splat (i64 3) |
| ; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[INDEX]] |
| ; CHECK-NEXT: call void @llvm.masked.store.v4i64.p0(<4 x i64> [[TMP6]], ptr align 8 [[TMP7]], <4 x i1> [[TMP3]]), !alias.scope [[META10:![0-9]+]], !noalias [[META7]] |
| ; CHECK-NEXT: [[TMP8:%.*]] = add <4 x i32> [[VEC_IND2]], splat (i32 100) |
| ; CHECK-NEXT: [[TMP9:%.*]] = select <4 x i1> [[TMP3]], <4 x i1> [[TMP5]], <4 x i1> zeroinitializer |
| ; CHECK-NEXT: [[TMP10:%.*]] = freeze <4 x i1> [[TMP9]] |
| ; CHECK-NEXT: [[TMP11:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP10]]) |
| ; CHECK-NEXT: [[TMP12]] = select i1 [[TMP11]], <4 x i1> [[TMP10]], <4 x i1> [[TMP2]] |
| ; CHECK-NEXT: [[TMP13]] = select i1 [[TMP11]], <4 x i32> [[TMP8]], <4 x i32> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[VEC_IND_NEXT3]] = add <4 x i32> [[VEC_IND2]], splat (i32 4) |
| ; CHECK-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP14]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP15:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP13]], <4 x i1> [[TMP12]], i32 -1) |
| ; CHECK-NEXT: br label %[[DONE:.*]] |
| ; CHECK: [[SCALAR_PH]]: |
| ; CHECK-NEXT: br label %[[LOOP:.*]] |
| ; CHECK: [[LOOP]]: |
| ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] |
| ; CHECK-NEXT: [[RDX:%.*]] = phi i32 [ -1, %[[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 [[LOOP13:![0-9]+]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: [[SEL_LCSSA:%.*]] = phi i32 [ [[SEL]], %[[LOOP]] ], [ [[TMP15]], %[[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 |
| } |
| |
| ; Test sdiv where the IV is the second operand (divisor). Verifies that |
| ; operand order is preserved when sinking the expression to the middle block. |
| ; IV starts at 1 to avoid division by zero. |
| define i64 @findlast_sdiv_iv_as_divisor(ptr %a, i64 %n) { |
| ; CHECK-LABEL: define i64 @findlast_sdiv_iv_as_divisor( |
| ; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*:]] |
| ; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N]], -1 |
| ; CHECK-NEXT: br label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 [[TMP0]], 3 |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = and i64 [[N_RND_UP]], 3 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[TMP0]], 1 |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer |
| ; 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 1, i64 2, i64 3, i64 4>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -1), %[[VECTOR_PH]] ], [ [[TMP11:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP1:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP10:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IV:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT2:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp ule <4 x i64> [[VEC_IV]], [[BROADCAST_SPLAT]] |
| ; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add nuw i64 1, [[INDEX]] |
| ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[OFFSET_IDX]] |
| ; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <4 x i64> @llvm.masked.load.v4i64.p0(ptr align 8 [[TMP3]], <4 x i1> [[TMP2]], <4 x i64> poison) |
| ; CHECK-NEXT: [[TMP4:%.*]] = icmp eq <4 x i64> [[WIDE_MASKED_LOAD]], splat (i64 42) |
| ; CHECK-NEXT: [[TMP6:%.*]] = call <4 x i64> @llvm.masked.sdiv.v4i64(<4 x i64> splat (i64 100), <4 x i64> [[VEC_IND]], <4 x i1> [[TMP2]]) |
| ; CHECK-NEXT: [[TMP7:%.*]] = select <4 x i1> [[TMP2]], <4 x i1> [[TMP4]], <4 x i1> zeroinitializer |
| ; CHECK-NEXT: [[TMP8:%.*]] = freeze <4 x i1> [[TMP7]] |
| ; CHECK-NEXT: [[TMP9:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP8]]) |
| ; CHECK-NEXT: [[TMP10]] = select i1 [[TMP9]], <4 x i1> [[TMP8]], <4 x i1> [[TMP1]] |
| ; CHECK-NEXT: [[TMP11]] = select i1 [[TMP9]], <4 x i64> [[TMP6]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[VEC_IND_NEXT2]] = add nuw <4 x i64> [[VEC_IV]], splat (i64 4) |
| ; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP12]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP14:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP13:%.*]] = call i64 @llvm.experimental.vector.extract.last.active.v4i64(<4 x i64> [[TMP11]], <4 x i1> [[TMP10]], i64 -1) |
| ; CHECK-NEXT: br label %[[DONE:.*]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: ret i64 [[TMP13]] |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %iv = phi i64 [ 1, %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 = sdiv i64 100, %iv |
| %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 expression sinking with flipped select polarity (TrueVal == PhiR) on the |
| ; AnyOf path. Based on @select_decreasing_induction_icmp_non_const_start from |
| ; iv-select-cmp-decreasing.ll but with the select operands swapped to exercise |
| ; the TrueVal == PhiR case, where the condition must be inverted without |
| ; corrupting the original Cond used later by the AnyOf tracking. |
| define i64 @findlast_expr_flipped_select_anyof(ptr %a, ptr %b, i64 %rdx.start, i64 %n) { |
| ; CHECK-LABEL: define i64 @findlast_expr_flipped_select_anyof( |
| ; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]], i64 [[RDX_START:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*:]] |
| ; CHECK-NEXT: [[UMIN:%.*]] = call i64 @llvm.umin.i64(i64 [[N]], i64 1) |
| ; CHECK-NEXT: [[TMP18:%.*]] = add i64 [[N]], 1 |
| ; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[TMP18]], [[UMIN]] |
| ; CHECK-NEXT: br label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 [[TMP1]], 3 |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = and i64 [[N_RND_UP]], 3 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[TMP1]], 1 |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i64> poison, i64 [[N]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT1]], <4 x i64> poison, <4 x i32> zeroinitializer |
| ; CHECK-NEXT: [[TMP17:%.*]] = sub nsw <4 x i64> [[BROADCAST_SPLAT2]], <i64 0, i64 1, i64 2, i64 3> |
| ; 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> [ [[TMP17]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 9223372036854775807), %[[VECTOR_PH]] ], [ [[TMP9:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP10:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IV:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT8:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp ule <4 x i64> [[VEC_IV]], [[BROADCAST_SPLAT]] |
| ; CHECK-NEXT: [[OFFSET_IDX:%.*]] = sub i64 [[N]], [[INDEX]] |
| ; CHECK-NEXT: [[TMP3:%.*]] = add nsw i64 [[OFFSET_IDX]], -1 |
| ; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP3]] |
| ; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i64, ptr [[TMP4]], i64 -3 |
| ; CHECK-NEXT: [[REVERSE:%.*]] = shufflevector <4 x i1> [[TMP2]], <4 x i1> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| ; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <4 x i64> @llvm.masked.load.v4i64.p0(ptr align 8 [[TMP5]], <4 x i1> [[REVERSE]], <4 x i64> poison) |
| ; CHECK-NEXT: [[REVERSE6:%.*]] = shufflevector <4 x i64> [[WIDE_MASKED_LOAD]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| ; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[TMP3]] |
| ; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i64, ptr [[TMP6]], i64 -3 |
| ; CHECK-NEXT: [[WIDE_MASKED_LOAD8:%.*]] = call <4 x i64> @llvm.masked.load.v4i64.p0(ptr align 8 [[TMP7]], <4 x i1> [[REVERSE]], <4 x i64> poison) |
| ; CHECK-NEXT: [[REVERSE9:%.*]] = shufflevector <4 x i64> [[WIDE_MASKED_LOAD8]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| ; CHECK-NEXT: [[TMP8:%.*]] = icmp sle <4 x i64> [[REVERSE6]], [[REVERSE9]] |
| ; CHECK-NEXT: [[TMP16:%.*]] = select <4 x i1> [[TMP2]], <4 x i1> [[TMP8]], <4 x i1> zeroinitializer |
| ; CHECK-NEXT: [[TMP9]] = select <4 x i1> [[TMP16]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[TMP10]] = or <4 x i1> [[VEC_PHI3]], [[TMP16]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 -4) |
| ; CHECK-NEXT: [[VEC_IND_NEXT8]] = add nuw <4 x i64> [[VEC_IV]], splat (i64 4) |
| ; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP11]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP12:%.*]] = call i64 @llvm.vector.reduce.smin.v4i64(<4 x i64> [[TMP9]]) |
| ; CHECK-NEXT: [[TMP13:%.*]] = add nsw i64 [[TMP12]], -1 |
| ; CHECK-NEXT: [[TMP14:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP10]]) |
| ; CHECK-NEXT: [[TMP15:%.*]] = freeze i1 [[TMP14]] |
| ; CHECK-NEXT: [[RDX_SELECT:%.*]] = select i1 [[TMP15]], i64 [[TMP13]], i64 [[RDX_START]] |
| ; CHECK-NEXT: br label %[[EXIT:.*]] |
| ; CHECK: [[EXIT]]: |
| ; CHECK-NEXT: ret i64 [[RDX_SELECT]] |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %iv = phi i64 [ %iv.next, %loop ], [ %n, %entry ] |
| %rdx = phi i64 [ %cond, %loop ], [ %rdx.start, %entry ] |
| %iv.next = add nsw i64 %iv, -1 |
| %gep.a.iv = getelementptr inbounds i64, ptr %a, i64 %iv.next |
| %ld.a = load i64, ptr %gep.a.iv, align 8 |
| %gep.b.iv = getelementptr inbounds i64, ptr %b, i64 %iv.next |
| %ld.b = load i64, ptr %gep.b.iv, align 8 |
| %cmp.a.b = icmp sgt i64 %ld.a, %ld.b |
| ; Flipped: TrueVal is rdx (PhiR), FalseVal is iv.next (expression of IV) |
| %cond = select i1 %cmp.a.b, i64 %rdx, i64 %iv.next |
| %exit.cond = icmp ugt i64 %iv, 1 |
| br i1 %exit.cond, label %loop, label %exit |
| |
| exit: |
| ret i64 %cond |
| } |
| |
| ; Test case where the expression %iv steps by one, but %expr is an AddRec |
| ; stepping by -1. |
| define i64 @findlast_sub_n_expr(ptr %a, i64 %n) { |
| ; CHECK-LABEL: define i64 @findlast_sub_n_expr( |
| ; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*:]] |
| ; CHECK-NEXT: br label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 [[N]], 3 |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = and i64 [[N_RND_UP]], 3 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[N]], 1 |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer |
| ; 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 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = icmp ule <4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <4 x i64> @llvm.masked.load.v4i64.p0(ptr align 8 [[TMP1]], <4 x i1> [[TMP0]], <4 x i64> poison) |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <4 x i64> [[WIDE_MASKED_LOAD]], splat (i64 42) |
| ; CHECK-NEXT: [[TMP9:%.*]] = select <4 x i1> [[TMP0]], <4 x i1> [[TMP2]], <4 x i1> zeroinitializer |
| ; CHECK-NEXT: [[TMP3]] = select <4 x i1> [[TMP9]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP5:%.*]] = call i64 @llvm.vector.reduce.smax.v4i64(<4 x i64> [[TMP3]]) |
| ; CHECK-NEXT: [[TMP6:%.*]] = sub i64 [[N]], [[TMP5]] |
| ; CHECK-NEXT: [[TMP7:%.*]] = icmp ne i64 [[TMP5]], -9223372036854775808 |
| ; CHECK-NEXT: [[TMP8:%.*]] = select i1 [[TMP7]], i64 [[TMP6]], i64 -1 |
| ; CHECK-NEXT: br label %[[EXIT:.*]] |
| ; CHECK: [[EXIT]]: |
| ; CHECK-NEXT: ret i64 [[TMP8]] |
| ; |
| 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 = sub i64 %n, %iv |
| %sel = select i1 %cmp, i64 %expr, i64 %rdx |
| %iv.next = add nuw nsw i64 %iv, 1 |
| %exitcond = icmp eq i64 %iv.next, %n |
| br i1 %exitcond, label %exit, label %loop |
| |
| exit: |
| ret i64 %sel |
| } |
| |
| ; Test sub where IV is the first operand (non-commutative). Verifies that |
| ; operand order is preserved when sinking the expression to the middle block. |
| define i64 @findlast_sub_iv_as_lhs(ptr %a, i64 %n) { |
| ; CHECK-LABEL: define i64 @findlast_sub_iv_as_lhs( |
| ; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*:]] |
| ; CHECK-NEXT: br label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 [[N]], 3 |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = and i64 [[N_RND_UP]], 3 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[N]], 1 |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer |
| ; 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 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = icmp ule <4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <4 x i64> @llvm.masked.load.v4i64.p0(ptr align 8 [[TMP1]], <4 x i1> [[TMP0]], <4 x i64> poison) |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <4 x i64> [[WIDE_MASKED_LOAD]], splat (i64 42) |
| ; CHECK-NEXT: [[TMP9:%.*]] = select <4 x i1> [[TMP0]], <4 x i1> [[TMP2]], <4 x i1> zeroinitializer |
| ; CHECK-NEXT: [[TMP3]] = select <4 x i1> [[TMP9]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP17:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP5:%.*]] = call i64 @llvm.vector.reduce.smax.v4i64(<4 x i64> [[TMP3]]) |
| ; CHECK-NEXT: [[TMP6:%.*]] = sub i64 [[TMP5]], [[N]] |
| ; CHECK-NEXT: [[TMP7:%.*]] = icmp ne i64 [[TMP5]], -9223372036854775808 |
| ; CHECK-NEXT: [[TMP8:%.*]] = select i1 [[TMP7]], i64 [[TMP6]], i64 -1 |
| ; CHECK-NEXT: br label %[[EXIT:.*]] |
| ; CHECK: [[EXIT]]: |
| ; CHECK-NEXT: ret i64 [[TMP8]] |
| ; |
| 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 = sub i64 %iv, %n |
| %sel = select i1 %cmp, i64 %expr, i64 %rdx |
| %iv.next = add nuw nsw i64 %iv, 1 |
| %exitcond = icmp eq i64 %iv.next, %n |
| br i1 %exitcond, label %exit, label %loop |
| |
| exit: |
| ret i64 %sel |
| } |
| |
| ; The IV starts at SMIN, so the signed range includes SMIN; signed sentinel |
| ; fails and an unsigned sentinel = 0 is used. The sunk udiv in the middle block |
| ; would divide by 0 if the condition is never true. |
| define i64 @findlast_udiv_may_trap_due_to_sentinel(ptr %a, i64 %n) { |
| ; CHECK-LABEL: define i64 @findlast_udiv_may_trap_due_to_sentinel( |
| ; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*:]] |
| ; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N]], -9223372036854775808 |
| ; CHECK-NEXT: br label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 [[TMP0]], 3 |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = and i64 [[N_RND_UP]], 3 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[TMP0]], 1 |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer |
| ; 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 -9223372036854775808, i64 -9223372036854775807, i64 -9223372036854775806, i64 -9223372036854775805>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -1), %[[VECTOR_PH]] ], [ [[TMP11:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP1:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP10:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IV:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT2:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp ule <4 x i64> [[VEC_IV]], [[BROADCAST_SPLAT]] |
| ; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add nuw i64 -9223372036854775808, [[INDEX]] |
| ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[OFFSET_IDX]] |
| ; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <4 x i64> @llvm.masked.load.v4i64.p0(ptr align 8 [[TMP3]], <4 x i1> [[TMP2]], <4 x i64> poison) |
| ; CHECK-NEXT: [[TMP4:%.*]] = icmp eq <4 x i64> [[WIDE_MASKED_LOAD]], splat (i64 42) |
| ; CHECK-NEXT: [[TMP6:%.*]] = call <4 x i64> @llvm.masked.udiv.v4i64(<4 x i64> splat (i64 100), <4 x i64> [[VEC_IND]], <4 x i1> [[TMP2]]) |
| ; CHECK-NEXT: [[TMP7:%.*]] = select <4 x i1> [[TMP2]], <4 x i1> [[TMP4]], <4 x i1> zeroinitializer |
| ; CHECK-NEXT: [[TMP8:%.*]] = freeze <4 x i1> [[TMP7]] |
| ; CHECK-NEXT: [[TMP9:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP8]]) |
| ; CHECK-NEXT: [[TMP10]] = select i1 [[TMP9]], <4 x i1> [[TMP8]], <4 x i1> [[TMP1]] |
| ; CHECK-NEXT: [[TMP11]] = select i1 [[TMP9]], <4 x i64> [[TMP6]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[VEC_IND_NEXT2]] = add nuw <4 x i64> [[VEC_IV]], splat (i64 4) |
| ; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP12]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[TMP13:%.*]] = call i64 @llvm.experimental.vector.extract.last.active.v4i64(<4 x i64> [[TMP11]], <4 x i1> [[TMP10]], i64 -1) |
| ; CHECK-NEXT: br label %[[DONE:.*]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: ret i64 [[TMP13]] |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %iv = phi i64 [ -9223372036854775808, %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 = udiv i64 100, %iv |
| %sel = select i1 %cmp, i64 %expr, i64 %rdx |
| %iv.next = add nuw i64 %iv, 1 |
| %exit = icmp eq i64 %iv.next, %n |
| br i1 %exit, label %done, label %loop |
| |
| done: |
| ret i64 %sel |
| } |
| |
| ; srem is not sinkable (may trap), so the expression should not be sunk. |
| define i64 @findlast_srem_iv_as_divisor(ptr %a, i64 %n) { |
| ; CHECK-LABEL: define i64 @findlast_srem_iv_as_divisor( |
| ; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*:]] |
| ; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N]], -1 |
| ; CHECK-NEXT: br label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 [[TMP0]], 3 |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = and i64 [[N_RND_UP]], 3 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[TMP0]], 1 |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer |
| ; 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 1, i64 2, i64 3, i64 4>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -1), %[[VECTOR_PH]] ], [ [[TMP11:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP1:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP10:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IV:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT2:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP2:%.*]] = icmp ule <4 x i64> [[VEC_IV]], [[BROADCAST_SPLAT]] |
| ; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add nuw i64 1, [[INDEX]] |
| ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[OFFSET_IDX]] |
| ; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <4 x i64> @llvm.masked.load.v4i64.p0(ptr align 8 [[TMP3]], <4 x i1> [[TMP2]], <4 x i64> poison) |
| ; CHECK-NEXT: [[TMP4:%.*]] = icmp eq <4 x i64> [[WIDE_MASKED_LOAD]], splat (i64 42) |
| ; CHECK-NEXT: [[TMP6:%.*]] = call <4 x i64> @llvm.masked.srem.v4i64(<4 x i64> splat (i64 100), <4 x i64> [[VEC_IND]], <4 x i1> [[TMP2]]) |
| ; CHECK-NEXT: [[TMP7:%.*]] = select <4 x i1> [[TMP2]], <4 x i1> [[TMP4]], <4 x i1> zeroinitializer |
| ; CHECK-NEXT: [[TMP8:%.*]] = freeze <4 x i1> [[TMP7]] |
| ; CHECK-NEXT: [[TMP9:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP8]]) |
| ; CHECK-NEXT: [[TMP10]] = select i1 [[TMP9]], <4 x i1> [[TMP8]], <4 x i1> [[TMP1]] |
| ; CHECK-NEXT: [[TMP11]] = select i1 [[TMP9]], <4 x i64> [[TMP6]], <4 x i64> [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i64> [[VEC_IND]], splat (i64 4) |
| ; CHECK-NEXT: [[VEC_IND_NEXT2]] = add nuw <4 x i64> [[VEC_IV]], splat (i64 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 i64 @llvm.experimental.vector.extract.last.active.v4i64(<4 x i64> [[TMP11]], <4 x i1> [[TMP10]], i64 -1) |
| ; CHECK-NEXT: br label %[[DONE:.*]] |
| ; CHECK: [[DONE]]: |
| ; CHECK-NEXT: ret i64 [[TMP13]] |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %iv = phi i64 [ 1, %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 = srem i64 100, %iv |
| %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 |
| } |
