clean this up.

[oota-llvm.git] / lib / Target / README.txt

diff --git a/lib/Target/README.txt b/lib/Target/README.txt
index c68ae76a2574033125bb6e2ba940f5fd1f8ac71c..4fd46a8b28ad2fec820098ef8829f5df015dcd57 100644 (file)
--- a/lib/Target/README.txt
+++ b/lib/Target/README.txt
@@ -156,6 +156,45 @@ void f () {  /* this can be optimized to four additions... */
 This requires reassociating to forms of expressions that are already available,
 something that reassoc doesn't think about yet.
 
+
+//===---------------------------------------------------------------------===//
+
+This function: (derived from GCC PR19988)
+double foo(double x, double y) {
+  return ((x + 0.1234 * y) * (x + -0.1234 * y));
+}
+
+compiles to:
+_foo:
+       movapd  %xmm1, %xmm2
+       mulsd   LCPI1_1(%rip), %xmm1
+       mulsd   LCPI1_0(%rip), %xmm2
+       addsd   %xmm0, %xmm1
+       addsd   %xmm0, %xmm2
+       movapd  %xmm1, %xmm0
+       mulsd   %xmm2, %xmm0
+       ret
+
+Reassociate should be able to turn it into:
+
+double foo(double x, double y) {
+  return ((x + 0.1234 * y) * (x - 0.1234 * y));
+}
+
+Which allows the multiply by constant to be CSE'd, producing:
+
+_foo:
+       mulsd   LCPI1_0(%rip), %xmm1
+       movapd  %xmm1, %xmm2
+       addsd   %xmm0, %xmm2
+       subsd   %xmm1, %xmm0
+       mulsd   %xmm2, %xmm0
+       ret
+
+This doesn't need -ffast-math support at all.  This is particularly bad because
+the llvm-gcc frontend is canonicalizing the later into the former, but clang
+doesn't have this problem.
+
 //===---------------------------------------------------------------------===//
 
 These two functions should generate the same code on big-endian systems:
@@ -1712,22 +1751,71 @@ from gcc.
 Missed instcombine transformation:
 define i32 @a(i32 %x) nounwind readnone {
 entry:
-  %shr = lshr i32 %x, 5                           ; <i32> [#uses=1]
-  %xor = xor i32 %shr, 67108864                   ; <i32> [#uses=1]
-  %sub = add i32 %xor, -67108864                  ; <i32> [#uses=1]
+  %rem = srem i32 %x, 32
+  %shl = shl i32 1, %rem
+  ret i32 %shl
+}
+
+The srem can be transformed to an and because if x is negative, the shift is
+undefined. Testcase derived from gcc.
+
+//===---------------------------------------------------------------------===//
+
+Missed instcombine/dagcombine transformation:
+define i32 @a(i32 %x, i32 %y) nounwind readnone {
+entry:
+  %mul = mul i32 %y, -8
+  %sub = sub i32 %x, %mul
   ret i32 %sub
 }
 
-This function is equivalent to "ashr i32 %x, 5".  Testcase derived from gcc.
+Should compile to something like x+y*8, but currently compiles to an
+inefficient result.  Testcase derived from gcc.
 
 //===---------------------------------------------------------------------===//
 
-isSafeToLoadUnconditionally should allow a GEP of a global/alloca with constant
-indicies within the bounds of the allocated object. Reduced example:
+Missed instcombine/dagcombine transformation:
+define void @lshift_lt(i8 zeroext %a) nounwind {
+entry:
+  %conv = zext i8 %a to i32
+  %shl = shl i32 %conv, 3
+  %cmp = icmp ult i32 %shl, 33
+  br i1 %cmp, label %if.then, label %if.end
 
-const int a[] = {3,6};
-int b(int y) { int* x = y ? &a[0] : &a[1]; return *x; }
+if.then:
+  tail call void @bar() nounwind
+  ret void
 
-All the loads should be eliminated.  Testcase derived from gcc.
+if.end:
+  ret void
+}
+declare void @bar() nounwind
+
+The shift should be eliminated.  Testcase derived from gcc.
+
+//===---------------------------------------------------------------------===//
+
+These compile into different code, one gets recognized as a switch and the
+other doesn't due to phase ordering issues (PR6212):
+
+int test1(int mainType, int subType) {
+  if (mainType == 7)
+    subType = 4;
+  else if (mainType == 9)
+    subType = 6;
+  else if (mainType == 11)
+    subType = 9;
+  return subType;
+}
+
+int test2(int mainType, int subType) {
+  if (mainType == 7)
+    subType = 4;
+  if (mainType == 9)
+    subType = 6;
+  if (mainType == 11)
+    subType = 9;
+  return subType;
+}
 
 //===---------------------------------------------------------------------===//