| ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py |
| ; RUN: opt < %s -passes=correlated-propagation -S | FileCheck %s |
| |
| target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128" |
| |
| define void @test0(i32 %n) { |
| ; CHECK-LABEL: @test0( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: br label [[FOR_COND:%.*]] |
| ; CHECK: for.cond: |
| ; CHECK-NEXT: [[J_0:%.*]] = phi i32 [ [[N:%.*]], [[ENTRY:%.*]] ], [ [[DIV1:%.*]], [[FOR_BODY:%.*]] ] |
| ; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[J_0]], 1 |
| ; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END:%.*]] |
| ; CHECK: for.body: |
| ; CHECK-NEXT: [[DIV1]] = udiv i32 [[J_0]], 2 |
| ; CHECK-NEXT: br label [[FOR_COND]] |
| ; CHECK: for.end: |
| ; CHECK-NEXT: ret void |
| ; |
| entry: |
| br label %for.cond |
| |
| for.cond: ; preds = %for.body, %entry |
| %j.0 = phi i32 [ %n, %entry ], [ %div, %for.body ] |
| %cmp = icmp sgt i32 %j.0, 1 |
| br i1 %cmp, label %for.body, label %for.end |
| |
| for.body: ; preds = %for.cond |
| %div = sdiv i32 %j.0, 2 |
| br label %for.cond |
| |
| for.end: ; preds = %for.cond |
| ret void |
| } |
| |
| define void @test1(i32 %n) { |
| ; CHECK-LABEL: @test1( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: br label [[FOR_COND:%.*]] |
| ; CHECK: for.cond: |
| ; CHECK-NEXT: [[J_0:%.*]] = phi i32 [ [[N:%.*]], [[ENTRY:%.*]] ], [ [[DIV:%.*]], [[FOR_BODY:%.*]] ] |
| ; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[J_0]], -2 |
| ; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END:%.*]] |
| ; CHECK: for.body: |
| ; CHECK-NEXT: [[DIV]] = sdiv i32 [[J_0]], 2 |
| ; CHECK-NEXT: br label [[FOR_COND]] |
| ; CHECK: for.end: |
| ; CHECK-NEXT: ret void |
| ; |
| entry: |
| br label %for.cond |
| |
| for.cond: ; preds = %for.body, %entry |
| %j.0 = phi i32 [ %n, %entry ], [ %div, %for.body ] |
| %cmp = icmp sgt i32 %j.0, -2 |
| br i1 %cmp, label %for.body, label %for.end |
| |
| for.body: ; preds = %for.cond |
| %div = sdiv i32 %j.0, 2 |
| br label %for.cond |
| |
| for.end: ; preds = %for.cond |
| ret void |
| } |
| |
| define void @test2(i32 %n) { |
| ; CHECK-LABEL: @test2( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[N:%.*]], 1 |
| ; CHECK-NEXT: br i1 [[CMP]], label [[BB:%.*]], label [[EXIT:%.*]] |
| ; CHECK: bb: |
| ; CHECK-NEXT: [[DIV1:%.*]] = udiv i32 [[N]], 2 |
| ; CHECK-NEXT: br label [[EXIT]] |
| ; CHECK: exit: |
| ; CHECK-NEXT: ret void |
| ; |
| entry: |
| %cmp = icmp sgt i32 %n, 1 |
| br i1 %cmp, label %bb, label %exit |
| |
| bb: |
| %div = sdiv i32 %n, 2 |
| br label %exit |
| |
| exit: |
| ret void |
| } |
| |
| ; looping case where loop has exactly one block |
| ; at the point of sdiv, we know that %a is always greater than 0, |
| ; because of the guard before it, so we can transform it to udiv. |
| declare void @llvm.experimental.guard(i1,...) |
| define void @test4(i32 %n) { |
| ; CHECK-LABEL: @test4( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[N:%.*]], 0 |
| ; CHECK-NEXT: br i1 [[CMP]], label [[LOOP:%.*]], label [[EXIT:%.*]] |
| ; CHECK: loop: |
| ; CHECK-NEXT: [[A:%.*]] = phi i32 [ [[N]], [[ENTRY:%.*]] ], [ [[DIV1:%.*]], [[LOOP]] ] |
| ; CHECK-NEXT: [[COND:%.*]] = icmp sgt i32 [[A]], 4 |
| ; CHECK-NEXT: call void (i1, ...) @llvm.experimental.guard(i1 [[COND]]) [ "deopt"() ] |
| ; CHECK-NEXT: [[DIV1]] = udiv i32 [[A]], 6 |
| ; CHECK-NEXT: br i1 [[COND]], label [[LOOP]], label [[EXIT]] |
| ; CHECK: exit: |
| ; CHECK-NEXT: ret void |
| ; |
| entry: |
| %cmp = icmp sgt i32 %n, 0 |
| br i1 %cmp, label %loop, label %exit |
| |
| loop: |
| %a = phi i32 [ %n, %entry ], [ %div, %loop ] |
| %cond = icmp sgt i32 %a, 4 |
| call void(i1,...) @llvm.experimental.guard(i1 %cond) [ "deopt"() ] |
| %div = sdiv i32 %a, 6 |
| br i1 %cond, label %loop, label %exit |
| |
| exit: |
| ret void |
| } |
| |
| ; same test as above with assume instead of guard. |
| declare void @llvm.assume(i1) |
| define void @test5(i32 %n) { |
| ; CHECK-LABEL: @test5( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[N:%.*]], 0 |
| ; CHECK-NEXT: br i1 [[CMP]], label [[LOOP:%.*]], label [[EXIT:%.*]] |
| ; CHECK: loop: |
| ; CHECK-NEXT: [[A:%.*]] = phi i32 [ [[N]], [[ENTRY:%.*]] ], [ [[DIV1:%.*]], [[LOOP]] ] |
| ; CHECK-NEXT: [[COND:%.*]] = icmp ugt i32 [[A]], 4 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[COND]]) |
| ; CHECK-NEXT: [[DIV1]] = udiv i32 [[A]], 6 |
| ; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp ugt i32 [[DIV1]], 8 |
| ; CHECK-NEXT: br i1 [[LOOPCOND]], label [[LOOP]], label [[EXIT]] |
| ; CHECK: exit: |
| ; CHECK-NEXT: ret void |
| ; |
| entry: |
| %cmp = icmp sgt i32 %n, 0 |
| br i1 %cmp, label %loop, label %exit |
| |
| loop: |
| %a = phi i32 [ %n, %entry ], [ %div, %loop ] |
| %cond = icmp sgt i32 %a, 4 |
| call void @llvm.assume(i1 %cond) |
| %div = sdiv i32 %a, 6 |
| %loopcond = icmp sgt i32 %div, 8 |
| br i1 %loopcond, label %loop, label %exit |
| |
| exit: |
| ret void |
| } |
| |
| ; Now, let's try various domain combinations for operands. |
| |
| define i32 @test6_pos_pos(i32 %x, i32 %y) { |
| ; CHECK-LABEL: @test6_pos_pos( |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sge i32 [[X:%.*]], 0 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sge i32 [[Y:%.*]], 0 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[DIV1:%.*]] = udiv i32 [[X]], [[Y]] |
| ; CHECK-NEXT: ret i32 [[DIV1]] |
| ; |
| %c0 = icmp sge i32 %x, 0 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sge i32 %y, 0 |
| call void @llvm.assume(i1 %c1) |
| |
| %div = sdiv i32 %x, %y |
| ret i32 %div |
| } |
| define i32 @test7_pos_neg(i32 %x, i32 %y) { |
| ; CHECK-LABEL: @test7_pos_neg( |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sge i32 [[X:%.*]], 0 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sle i32 [[Y:%.*]], 0 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[Y_NONNEG:%.*]] = sub i32 0, [[Y]] |
| ; CHECK-NEXT: [[DIV1:%.*]] = udiv i32 [[X]], [[Y_NONNEG]] |
| ; CHECK-NEXT: [[DIV1_NEG:%.*]] = sub i32 0, [[DIV1]] |
| ; CHECK-NEXT: ret i32 [[DIV1_NEG]] |
| ; |
| %c0 = icmp sge i32 %x, 0 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sle i32 %y, 0 |
| call void @llvm.assume(i1 %c1) |
| |
| %div = sdiv i32 %x, %y |
| ret i32 %div |
| } |
| define i32 @test8_neg_pos(i32 %x, i32 %y) { |
| ; CHECK-LABEL: @test8_neg_pos( |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sle i32 [[X:%.*]], 0 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sge i32 [[Y:%.*]], 0 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[X_NONNEG:%.*]] = sub i32 0, [[X]] |
| ; CHECK-NEXT: [[DIV1:%.*]] = udiv i32 [[X_NONNEG]], [[Y]] |
| ; CHECK-NEXT: [[DIV1_NEG:%.*]] = sub i32 0, [[DIV1]] |
| ; CHECK-NEXT: ret i32 [[DIV1_NEG]] |
| ; |
| %c0 = icmp sle i32 %x, 0 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sge i32 %y, 0 |
| call void @llvm.assume(i1 %c1) |
| |
| %div = sdiv i32 %x, %y |
| ret i32 %div |
| } |
| define i32 @test9_neg_neg(i32 %x, i32 %y) { |
| ; CHECK-LABEL: @test9_neg_neg( |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sle i32 [[X:%.*]], 0 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sle i32 [[Y:%.*]], 0 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[X_NONNEG:%.*]] = sub i32 0, [[X]] |
| ; CHECK-NEXT: [[Y_NONNEG:%.*]] = sub i32 0, [[Y]] |
| ; CHECK-NEXT: [[DIV1:%.*]] = udiv i32 [[X_NONNEG]], [[Y_NONNEG]] |
| ; CHECK-NEXT: ret i32 [[DIV1]] |
| ; |
| %c0 = icmp sle i32 %x, 0 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sle i32 %y, 0 |
| call void @llvm.assume(i1 %c1) |
| |
| %div = sdiv i32 %x, %y |
| ret i32 %div |
| } |
| |
| ; After making division unsigned, can we narrow it? |
| define i32 @test10_narrow(i32 %x, i32 %y) { |
| ; CHECK-LABEL: @test10_narrow( |
| ; CHECK-NEXT: [[C0:%.*]] = icmp ult i32 [[X:%.*]], 128 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp ult i32 [[Y:%.*]], 128 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[DIV1_LHS_TRUNC:%.*]] = trunc i32 [[X]] to i8 |
| ; CHECK-NEXT: [[DIV1_RHS_TRUNC:%.*]] = trunc i32 [[Y]] to i8 |
| ; CHECK-NEXT: [[DIV12:%.*]] = udiv i8 [[DIV1_LHS_TRUNC]], [[DIV1_RHS_TRUNC]] |
| ; CHECK-NEXT: [[DIV1_ZEXT:%.*]] = zext i8 [[DIV12]] to i32 |
| ; CHECK-NEXT: ret i32 [[DIV1_ZEXT]] |
| ; |
| %c0 = icmp ult i32 %x, 128 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp ult i32 %y, 128 |
| call void @llvm.assume(i1 %c1) |
| |
| %div = sdiv i32 %x, %y |
| ret i32 %div |
| } |
| |
| ; Ok, but what about narrowing sdiv in general? |
| |
| ; If both operands are i15, it's uncontroversial - we can truncate to i16 |
| define i64 @test11_i15_i15(i64 %x, i64 %y) { |
| ; CHECK-LABEL: @test11_i15_i15( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sle i64 [[X:%.*]], 16383 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sge i64 [[X]], -16384 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[C2:%.*]] = icmp sle i64 [[Y:%.*]], 16383 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C2]]) |
| ; CHECK-NEXT: [[C3:%.*]] = icmp sge i64 [[Y]], -16384 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C3]]) |
| ; CHECK-NEXT: [[DIV_LHS_TRUNC:%.*]] = trunc i64 [[X]] to i16 |
| ; CHECK-NEXT: [[DIV_RHS_TRUNC:%.*]] = trunc i64 [[Y]] to i16 |
| ; CHECK-NEXT: [[DIV1:%.*]] = sdiv i16 [[DIV_LHS_TRUNC]], [[DIV_RHS_TRUNC]] |
| ; CHECK-NEXT: [[DIV_SEXT:%.*]] = sext i16 [[DIV1]] to i64 |
| ; CHECK-NEXT: ret i64 [[DIV_SEXT]] |
| ; |
| entry: |
| %c0 = icmp sle i64 %x, 16383 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sge i64 %x, -16384 |
| call void @llvm.assume(i1 %c1) |
| |
| %c2 = icmp sle i64 %y, 16383 |
| call void @llvm.assume(i1 %c2) |
| %c3 = icmp sge i64 %y, -16384 |
| call void @llvm.assume(i1 %c3) |
| |
| %div = sdiv i64 %x, %y |
| ret i64 %div |
| } |
| |
| ; But if operands are i16, we can only truncate to i32, because we can't |
| ; rule out UB of i16 INT_MIN s/ i16 -1 |
| define i64 @test12_i16_i16(i64 %x, i64 %y) { |
| ; CHECK-LABEL: @test12_i16_i16( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sle i64 [[X:%.*]], 32767 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sge i64 [[X]], -32768 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[C2:%.*]] = icmp sle i64 [[Y:%.*]], 32767 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C2]]) |
| ; CHECK-NEXT: [[C3:%.*]] = icmp sge i64 [[Y]], -32768 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C3]]) |
| ; CHECK-NEXT: [[DIV_LHS_TRUNC:%.*]] = trunc i64 [[X]] to i32 |
| ; CHECK-NEXT: [[DIV_RHS_TRUNC:%.*]] = trunc i64 [[Y]] to i32 |
| ; CHECK-NEXT: [[DIV1:%.*]] = sdiv i32 [[DIV_LHS_TRUNC]], [[DIV_RHS_TRUNC]] |
| ; CHECK-NEXT: [[DIV_SEXT:%.*]] = sext i32 [[DIV1]] to i64 |
| ; CHECK-NEXT: ret i64 [[DIV_SEXT]] |
| ; |
| entry: |
| %c0 = icmp sle i64 %x, 32767 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sge i64 %x, -32768 |
| call void @llvm.assume(i1 %c1) |
| |
| %c2 = icmp sle i64 %y, 32767 |
| call void @llvm.assume(i1 %c2) |
| %c3 = icmp sge i64 %y, -32768 |
| call void @llvm.assume(i1 %c3) |
| |
| %div = sdiv i64 %x, %y |
| ret i64 %div |
| } |
| |
| ; But if divident is i16, and divisor is u15, then we know that i16 is UB-safe. |
| define i64 @test13_i16_u15(i64 %x, i64 %y) { |
| ; CHECK-LABEL: @test13_i16_u15( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sle i64 [[X:%.*]], 32767 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sge i64 [[X]], -32768 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[C2:%.*]] = icmp ule i64 [[Y:%.*]], 32767 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C2]]) |
| ; CHECK-NEXT: [[DIV_LHS_TRUNC:%.*]] = trunc i64 [[X]] to i16 |
| ; CHECK-NEXT: [[DIV_RHS_TRUNC:%.*]] = trunc i64 [[Y]] to i16 |
| ; CHECK-NEXT: [[DIV1:%.*]] = sdiv i16 [[DIV_LHS_TRUNC]], [[DIV_RHS_TRUNC]] |
| ; CHECK-NEXT: [[DIV_SEXT:%.*]] = sext i16 [[DIV1]] to i64 |
| ; CHECK-NEXT: ret i64 [[DIV_SEXT]] |
| ; |
| entry: |
| %c0 = icmp sle i64 %x, 32767 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sge i64 %x, -32768 |
| call void @llvm.assume(i1 %c1) |
| |
| %c2 = icmp ule i64 %y, 32767 |
| call void @llvm.assume(i1 %c2) |
| |
| %div = sdiv i64 %x, %y |
| ret i64 %div |
| } |
| |
| ; And likewise, if we know that if the divident is never i16 INT_MIN, |
| ; we can truncate to i16. |
| define i64 @test14_i16safe_i16(i64 %x, i64 %y) { |
| ; CHECK-LABEL: @test14_i16safe_i16( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sle i64 [[X:%.*]], 32767 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sgt i64 [[X]], -32768 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[C2:%.*]] = icmp sle i64 [[Y:%.*]], 32767 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C2]]) |
| ; CHECK-NEXT: [[C3:%.*]] = icmp sge i64 [[Y]], -32768 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C3]]) |
| ; CHECK-NEXT: [[DIV_LHS_TRUNC:%.*]] = trunc i64 [[X]] to i16 |
| ; CHECK-NEXT: [[DIV_RHS_TRUNC:%.*]] = trunc i64 [[Y]] to i16 |
| ; CHECK-NEXT: [[DIV1:%.*]] = sdiv i16 [[DIV_LHS_TRUNC]], [[DIV_RHS_TRUNC]] |
| ; CHECK-NEXT: [[DIV_SEXT:%.*]] = sext i16 [[DIV1]] to i64 |
| ; CHECK-NEXT: ret i64 [[DIV_SEXT]] |
| ; |
| entry: |
| %c0 = icmp sle i64 %x, 32767 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sgt i64 %x, -32768 |
| call void @llvm.assume(i1 %c1) |
| |
| %c2 = icmp sle i64 %y, 32767 |
| call void @llvm.assume(i1 %c2) |
| %c3 = icmp sge i64 %y, -32768 |
| call void @llvm.assume(i1 %c3) |
| |
| %div = sdiv i64 %x, %y |
| ret i64 %div |
| } |
| |
| ; Of course, both of the conditions can happen at once. |
| define i64 @test15_i16safe_u15(i64 %x, i64 %y) { |
| ; CHECK-LABEL: @test15_i16safe_u15( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sle i64 [[X:%.*]], 32767 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sgt i64 [[X]], -32768 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[C2:%.*]] = icmp ule i64 [[Y:%.*]], 32767 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C2]]) |
| ; CHECK-NEXT: [[DIV_LHS_TRUNC:%.*]] = trunc i64 [[X]] to i16 |
| ; CHECK-NEXT: [[DIV_RHS_TRUNC:%.*]] = trunc i64 [[Y]] to i16 |
| ; CHECK-NEXT: [[DIV1:%.*]] = sdiv i16 [[DIV_LHS_TRUNC]], [[DIV_RHS_TRUNC]] |
| ; CHECK-NEXT: [[DIV_SEXT:%.*]] = sext i16 [[DIV1]] to i64 |
| ; CHECK-NEXT: ret i64 [[DIV_SEXT]] |
| ; |
| entry: |
| %c0 = icmp sle i64 %x, 32767 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sgt i64 %x, -32768 |
| call void @llvm.assume(i1 %c1) |
| |
| %c2 = icmp ule i64 %y, 32767 |
| call void @llvm.assume(i1 %c2) |
| |
| %div = sdiv i64 %x, %y |
| ret i64 %div |
| } |
| |
| ; We at most truncate to i8 |
| define i64 @test16_i4_i4(i64 %x, i64 %y) { |
| ; CHECK-LABEL: @test16_i4_i4( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sle i64 [[X:%.*]], 3 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sge i64 [[X]], -4 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[C2:%.*]] = icmp sle i64 [[Y:%.*]], 3 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C2]]) |
| ; CHECK-NEXT: [[C3:%.*]] = icmp sge i64 [[Y]], -4 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C3]]) |
| ; CHECK-NEXT: [[DIV_LHS_TRUNC:%.*]] = trunc i64 [[X]] to i8 |
| ; CHECK-NEXT: [[DIV_RHS_TRUNC:%.*]] = trunc i64 [[Y]] to i8 |
| ; CHECK-NEXT: [[DIV1:%.*]] = sdiv i8 [[DIV_LHS_TRUNC]], [[DIV_RHS_TRUNC]] |
| ; CHECK-NEXT: [[DIV_SEXT:%.*]] = sext i8 [[DIV1]] to i64 |
| ; CHECK-NEXT: ret i64 [[DIV_SEXT]] |
| ; |
| entry: |
| %c0 = icmp sle i64 %x, 3 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sge i64 %x, -4 |
| call void @llvm.assume(i1 %c1) |
| |
| %c2 = icmp sle i64 %y, 3 |
| call void @llvm.assume(i1 %c2) |
| %c3 = icmp sge i64 %y, -4 |
| call void @llvm.assume(i1 %c3) |
| |
| %div = sdiv i64 %x, %y |
| ret i64 %div |
| } |
| |
| ; And we round up to the powers of two |
| define i64 @test17_i9_i9(i64 %x, i64 %y) { |
| ; CHECK-LABEL: @test17_i9_i9( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sle i64 [[X:%.*]], 255 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sge i64 [[X]], -256 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[C2:%.*]] = icmp sle i64 [[Y:%.*]], 255 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C2]]) |
| ; CHECK-NEXT: [[C3:%.*]] = icmp sge i64 [[Y]], -256 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C3]]) |
| ; CHECK-NEXT: [[DIV_LHS_TRUNC:%.*]] = trunc i64 [[X]] to i16 |
| ; CHECK-NEXT: [[DIV_RHS_TRUNC:%.*]] = trunc i64 [[Y]] to i16 |
| ; CHECK-NEXT: [[DIV1:%.*]] = sdiv i16 [[DIV_LHS_TRUNC]], [[DIV_RHS_TRUNC]] |
| ; CHECK-NEXT: [[DIV_SEXT:%.*]] = sext i16 [[DIV1]] to i64 |
| ; CHECK-NEXT: ret i64 [[DIV_SEXT]] |
| ; |
| entry: |
| %c0 = icmp sle i64 %x, 255 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sge i64 %x, -256 |
| call void @llvm.assume(i1 %c1) |
| |
| %c2 = icmp sle i64 %y, 255 |
| call void @llvm.assume(i1 %c2) |
| %c3 = icmp sge i64 %y, -256 |
| call void @llvm.assume(i1 %c3) |
| |
| %div = sdiv i64 %x, %y |
| ret i64 %div |
| } |
| |
| ; Don't widen the operation to the next power of two if it wasn't a power of two. |
| define i9 @test18_i9_i9(i9 %x, i9 %y) { |
| ; CHECK-LABEL: @test18_i9_i9( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sle i9 [[X:%.*]], 255 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sge i9 [[X]], -256 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[C2:%.*]] = icmp sle i9 [[Y:%.*]], 255 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C2]]) |
| ; CHECK-NEXT: [[C3:%.*]] = icmp sge i9 [[Y]], -256 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C3]]) |
| ; CHECK-NEXT: [[DIV:%.*]] = sdiv i9 [[X]], [[Y]] |
| ; CHECK-NEXT: ret i9 [[DIV]] |
| ; |
| entry: |
| %c0 = icmp sle i9 %x, 255 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sge i9 %x, -256 |
| call void @llvm.assume(i1 %c1) |
| |
| %c2 = icmp sle i9 %y, 255 |
| call void @llvm.assume(i1 %c2) |
| %c3 = icmp sge i9 %y, -256 |
| call void @llvm.assume(i1 %c3) |
| |
| %div = sdiv i9 %x, %y |
| ret i9 %div |
| } |
| define i10 @test19_i10_i10(i10 %x, i10 %y) { |
| ; CHECK-LABEL: @test19_i10_i10( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sle i10 [[X:%.*]], 255 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sge i10 [[X]], -256 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[C2:%.*]] = icmp sle i10 [[Y:%.*]], 255 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C2]]) |
| ; CHECK-NEXT: [[C3:%.*]] = icmp sge i10 [[Y]], -256 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C3]]) |
| ; CHECK-NEXT: [[DIV:%.*]] = sdiv i10 [[X]], [[Y]] |
| ; CHECK-NEXT: ret i10 [[DIV]] |
| ; |
| entry: |
| %c0 = icmp sle i10 %x, 255 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sge i10 %x, -256 |
| call void @llvm.assume(i1 %c1) |
| |
| %c2 = icmp sle i10 %y, 255 |
| call void @llvm.assume(i1 %c2) |
| %c3 = icmp sge i10 %y, -256 |
| call void @llvm.assume(i1 %c3) |
| |
| %div = sdiv i10 %x, %y |
| ret i10 %div |
| } |
| |
| ; Note that we need to take the maximal bitwidth, in which both of the operands are representable! |
| define i64 @test20_i16_i18(i64 %x, i64 %y) { |
| ; CHECK-LABEL: @test20_i16_i18( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sle i64 [[X:%.*]], 16383 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sge i64 [[X]], -16384 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[C2:%.*]] = icmp sle i64 [[Y:%.*]], 65535 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C2]]) |
| ; CHECK-NEXT: [[C3:%.*]] = icmp sge i64 [[Y]], -65536 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C3]]) |
| ; CHECK-NEXT: [[DIV_LHS_TRUNC:%.*]] = trunc i64 [[X]] to i32 |
| ; CHECK-NEXT: [[DIV_RHS_TRUNC:%.*]] = trunc i64 [[Y]] to i32 |
| ; CHECK-NEXT: [[DIV1:%.*]] = sdiv i32 [[DIV_LHS_TRUNC]], [[DIV_RHS_TRUNC]] |
| ; CHECK-NEXT: [[DIV_SEXT:%.*]] = sext i32 [[DIV1]] to i64 |
| ; CHECK-NEXT: ret i64 [[DIV_SEXT]] |
| ; |
| entry: |
| %c0 = icmp sle i64 %x, 16383 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sge i64 %x, -16384 |
| call void @llvm.assume(i1 %c1) |
| |
| %c2 = icmp sle i64 %y, 65535 |
| call void @llvm.assume(i1 %c2) |
| %c3 = icmp sge i64 %y, -65536 |
| call void @llvm.assume(i1 %c3) |
| |
| %div = sdiv i64 %x, %y |
| ret i64 %div |
| } |
| define i64 @test21_i18_i16(i64 %x, i64 %y) { |
| ; CHECK-LABEL: @test21_i18_i16( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sle i64 [[X:%.*]], 65535 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sge i64 [[X]], -65536 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[C2:%.*]] = icmp sle i64 [[Y:%.*]], 16383 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C2]]) |
| ; CHECK-NEXT: [[C3:%.*]] = icmp sge i64 [[Y]], -16384 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C3]]) |
| ; CHECK-NEXT: [[DIV_LHS_TRUNC:%.*]] = trunc i64 [[X]] to i32 |
| ; CHECK-NEXT: [[DIV_RHS_TRUNC:%.*]] = trunc i64 [[Y]] to i32 |
| ; CHECK-NEXT: [[DIV1:%.*]] = sdiv i32 [[DIV_LHS_TRUNC]], [[DIV_RHS_TRUNC]] |
| ; CHECK-NEXT: [[DIV_SEXT:%.*]] = sext i32 [[DIV1]] to i64 |
| ; CHECK-NEXT: ret i64 [[DIV_SEXT]] |
| ; |
| entry: |
| %c0 = icmp sle i64 %x, 65535 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sge i64 %x, -65536 |
| call void @llvm.assume(i1 %c1) |
| |
| %c2 = icmp sle i64 %y, 16383 |
| call void @llvm.assume(i1 %c2) |
| %c3 = icmp sge i64 %y, -16384 |
| call void @llvm.assume(i1 %c3) |
| |
| %div = sdiv i64 %x, %y |
| ret i64 %div |
| } |
| |
| ; Ensure that we preserve exact-ness |
| define i64 @test22_i16_i16(i64 %x, i64 %y) { |
| ; CHECK-LABEL: @test22_i16_i16( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[C0:%.*]] = icmp sle i64 [[X:%.*]], 32767 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C0]]) |
| ; CHECK-NEXT: [[C1:%.*]] = icmp sge i64 [[X]], -32768 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C1]]) |
| ; CHECK-NEXT: [[C2:%.*]] = icmp sle i64 [[Y:%.*]], 32767 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C2]]) |
| ; CHECK-NEXT: [[C3:%.*]] = icmp sge i64 [[Y]], -32768 |
| ; CHECK-NEXT: call void @llvm.assume(i1 [[C3]]) |
| ; CHECK-NEXT: [[DIV_LHS_TRUNC:%.*]] = trunc i64 [[X]] to i32 |
| ; CHECK-NEXT: [[DIV_RHS_TRUNC:%.*]] = trunc i64 [[Y]] to i32 |
| ; CHECK-NEXT: [[DIV1:%.*]] = sdiv exact i32 [[DIV_LHS_TRUNC]], [[DIV_RHS_TRUNC]] |
| ; CHECK-NEXT: [[DIV_SEXT:%.*]] = sext i32 [[DIV1]] to i64 |
| ; CHECK-NEXT: ret i64 [[DIV_SEXT]] |
| ; |
| entry: |
| %c0 = icmp sle i64 %x, 32767 |
| call void @llvm.assume(i1 %c0) |
| %c1 = icmp sge i64 %x, -32768 |
| call void @llvm.assume(i1 %c1) |
| |
| %c2 = icmp sle i64 %y, 32767 |
| call void @llvm.assume(i1 %c2) |
| %c3 = icmp sge i64 %y, -32768 |
| call void @llvm.assume(i1 %c3) |
| |
| %div = sdiv exact i64 %x, %y |
| ret i64 %div |
| } |