Enhance the "compare with shift" and "compare with div"

optimizations to be much more aggressive in the face of
exact/nsw/nuw div and shifts.  For example, these (which
are the same except the first is 'exact' sdiv:

define i1 @sdiv_icmp4_exact(i64 %X) nounwind {
  %A = sdiv exact i64 %X, -5   ; X/-5 == 0 --> x == 0
  %B = icmp eq i64 %A, 0
  ret i1 %B
}

define i1 @sdiv_icmp4(i64 %X) nounwind {
  %A = sdiv i64 %X, -5   ; X/-5 == 0 --> x == 0
  %B = icmp eq i64 %A, 0
  ret i1 %B
}

compile down to:

define i1 @sdiv_icmp4_exact(i64 %X) nounwind {
  %1 = icmp eq i64 %X, 0
  ret i1 %1
}

define i1 @sdiv_icmp4(i64 %X) nounwind {
  %X.off = add i64 %X, 4
  %1 = icmp ult i64 %X.off, 9
  ret i1 %1
}

This happens when you do something like:
  (ptr1-ptr2) == 42

where the pointers are pointers to non-unit types.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@125266 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 99 insertions, 30 deletions

diff --git a/test/Transforms/InstCombine/exact.ll b/test/Transforms/InstCombine/exact.ll
index 940469e..2add717 100644
--- a/test/Transforms/InstCombine/exact.ll
+++ b/test/Transforms/InstCombine/exact.ll
@@ -1,60 +1,119 @@
 ; RUN: opt < %s -instcombine -S | FileCheck %s
 
-; CHECK: define i32 @foo
+; CHECK: @sdiv1
 ; CHECK: sdiv i32 %x, 8
-define i32 @foo(i32 %x) {
+define i32 @sdiv1(i32 %x) {
   %y = sdiv i32 %x, 8
   ret i32 %y
 }
 
-; CHECK: define i32 @bar
-; CHECK: ashr i32 %x, 3
-define i32 @bar(i32 %x) {
-  %y = sdiv exact i32 %x, 8
-  ret i32 %y
-}
-
-; CHECK: i32 @a0
+; CHECK: @sdiv3
 ; CHECK: %y = srem i32 %x, 3
 ; CHECK: %z = sub i32 %x, %y
 ; CHECK: ret i32 %z
-define i32 @a0(i32 %x) {
+define i32 @sdiv3(i32 %x) {
   %y = sdiv i32 %x, 3
   %z = mul i32 %y, 3
   ret i32 %z
 }
 
-; CHECK: i32 @b0
+; CHECK: @sdiv4
 ; CHECK: ret i32 %x
-define i32 @b0(i32 %x) {
+define i32 @sdiv4(i32 %x) {
   %y = sdiv exact i32 %x, 3
   %z = mul i32 %y, 3
   ret i32 %z
 }
 
-; CHECK: i32 @a1
+; CHECK: i32 @sdiv5
 ; CHECK: %y = srem i32 %x, 3
 ; CHECK: %z = sub i32 %y, %x
 ; CHECK: ret i32 %z
-define i32 @a1(i32 %x) {
+define i32 @sdiv5(i32 %x) {
   %y = sdiv i32 %x, 3
   %z = mul i32 %y, -3
   ret i32 %z
 }
 
-; CHECK: i32 @b1
+; CHECK: @sdiv6
 ; CHECK: %z = sub i32 0, %x
 ; CHECK: ret i32 %z
-define i32 @b1(i32 %x) {
+define i32 @sdiv6(i32 %x) {
   %y = sdiv exact i32 %x, 3
   %z = mul i32 %y, -3
   ret i32 %z
 }
 
-; CHECK: i32 @b2
+; CHECK: @udiv1
 ; CHECK: ret i32 %x
-define i32 @b2(i32 %x, i32 %w) {
+define i32 @udiv1(i32 %x, i32 %w) {
   %y = udiv exact i32 %x, %w
   %z = mul i32 %y, %w
   ret i32 %z
 }
+
+; CHECK: @ashr_icmp
+; CHECK: %B = icmp eq i64 %X, 0
+; CHECK: ret i1 %B
+define i1 @ashr_icmp(i64 %X) nounwind {
+  %A = ashr exact i64 %X, 2   ; X/4
+  %B = icmp eq i64 %A, 0
+  ret i1 %B
+}
+
+; CHECK: @udiv_icmp1
+; CHECK: icmp ne i64 %X, 0
+define i1 @udiv_icmp1(i64 %X) nounwind {
+  %A = udiv exact i64 %X, 5   ; X/5
+  %B = icmp ne i64 %A, 0
+  ret i1 %B
+}
+
+; CHECK: @sdiv_icmp1
+; CHECK: icmp eq i64 %X, 0
+define i1 @sdiv_icmp1(i64 %X) nounwind {
+  %A = sdiv exact i64 %X, 5   ; X/5 == 0 --> x == 0
+  %B = icmp eq i64 %A, 0
+  ret i1 %B
+}
+
+; CHECK: @sdiv_icmp2
+; CHECK: icmp eq i64 %X, 5
+define i1 @sdiv_icmp2(i64 %X) nounwind {
+  %A = sdiv exact i64 %X, 5   ; X/5 == 1 --> x == 5
+  %B = icmp eq i64 %A, 1
+  ret i1 %B
+}
+
+; CHECK: @sdiv_icmp3
+; CHECK: icmp eq i64 %X, -5
+define i1 @sdiv_icmp3(i64 %X) nounwind {
+  %A = sdiv exact i64 %X, 5   ; X/5 == -1 --> x == -5
+  %B = icmp eq i64 %A, -1
+  ret i1 %B
+}
+
+; CHECK: @sdiv_icmp4
+; CHECK: icmp eq i64 %X, 0
+define i1 @sdiv_icmp4(i64 %X) nounwind {
+  %A = sdiv exact i64 %X, -5   ; X/-5 == 0 --> x == 0
+  %B = icmp eq i64 %A, 0
+  ret i1 %B
+}
+
+; CHECK: @sdiv_icmp5
+; CHECK: icmp eq i64 %X, -5
+define i1 @sdiv_icmp5(i64 %X) nounwind {
+  %A = sdiv exact i64 %X, -5   ; X/-5 == 1 --> x == -5
+  %B = icmp eq i64 %A, 1
+  ret i1 %B
+}
+
+; CHECK: @sdiv_icmp6
+; CHECK: icmp eq i64 %X, 5
+define i1 @sdiv_icmp6(i64 %X) nounwind {
+  %A = sdiv exact i64 %X, -5   ; X/-5 == 1 --> x == 5
+  %B = icmp eq i64 %A, -1
+  ret i1 %B
+}
+
diff --git a/test/Transforms/InstCombine/nsw.ll b/test/Transforms/InstCombine/nsw.ll
index 821cebe..efe2391 100644
--- a/test/Transforms/InstCombine/nsw.ll
+++ b/test/Transforms/InstCombine/nsw.ll
@@ -1,20 +1,30 @@
 ; RUN: opt < %s -instcombine -S | FileCheck %s
 
-; CHECK: define i32 @foo
-; %y = sub i32 0, %x
-; %z = sdiv i32 %y, 337
-; ret i32 %y
-define i32 @foo(i32 %x) {
+; CHECK: @sub1
+; CHECK: %y = sub i32 0, %x
+; CHECK: %z = sdiv i32 %y, 337
+; CHECK: ret i32 %z
+define i32 @sub1(i32 %x) {
   %y = sub i32 0, %x
   %z = sdiv i32 %y, 337
-  ret i32 %y
+  ret i32 %z
 }
 
-; CHECK: define i32 @bar
-; %y = sdiv i32 %x, -337
-; ret i32 %y
-define i32 @bar(i32 %x) {
+; CHECK: @sub2
+; CHECK: %z = sdiv i32 %x, -337
+; CHECK: ret i32 %z
+define i32 @sub2(i32 %x) {
   %y = sub nsw i32 0, %x
   %z = sdiv i32 %y, 337
-  ret i32 %y
+  ret i32 %z
 }
+
+; CHECK: @shl_icmp
+; CHECK: %B = icmp eq i64 %X, 0
+; CHECK: ret i1 %B
+define i1 @shl_icmp(i64 %X) nounwind {
+  %A = shl nuw i64 %X, 2   ; X/4
+  %B = icmp eq i64 %A, 0
+  ret i1 %B
+}
+