Improve varags handling, with testcases. Patch by Sasa Stankovic

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

diff --git a/lib/Target/README.txt b/lib/Target/README.txt
index 1a33aa8ed8faca8289d52c5abe8eac78e44810fd..45da3ddb607a1532a3b24faeed0549895271b308 100644 (file)
--- a/lib/Target/README.txt
+++ b/lib/Target/README.txt
@@ -254,6 +254,20 @@ unsigned long reverse(unsigned v) {
 
 //===---------------------------------------------------------------------===//
 
+[LOOP DELETION]
+
+We don't delete this output free loop, because trip count analysis doesn't
+realize that it is finite (if it were infinite, it would be undefined).  Not
+having this blocks Loop Idiom from matching strlen and friends.  
+
+void foo(char *C) {
+  int x = 0;
+  while (*C)
+    ++x,++C;
+}
+
+//===---------------------------------------------------------------------===//
+
 [LOOP RECOGNITION]
 
 These idioms should be recognized as popcount (see PR1488):
@@ -287,6 +301,16 @@ unsigned int popcount(unsigned int input) {
   return count;
 }
 
+This should be recognized as CLZ:  rdar://8459039
+
+unsigned clz_a(unsigned a) {
+  int i;
+  for (i=0;i<32;i++)
+    if (a & (1<<(31-i)))
+      return i;
+  return 32;
+}
+
 This sort of thing should be added to the loop idiom pass.
 
 //===---------------------------------------------------------------------===//
@@ -1301,6 +1325,21 @@ codegen.
 
 //===---------------------------------------------------------------------===//
 
+simplifylibcalls should turn these snprintf idioms into memcpy (GCC PR47917)
+
+char buf1[6], buf2[6], buf3[4], buf4[4];
+int i;
+
+int foo (void) {
+  int ret = snprintf (buf1, sizeof buf1, "abcde");
+  ret += snprintf (buf2, sizeof buf2, "abcdef") * 16;
+  ret += snprintf (buf3, sizeof buf3, "%s", i++ < 6 ? "abc" : "def") * 256;
+  ret += snprintf (buf4, sizeof buf4, "%s", i++ > 10 ? "abcde" : "defgh")*4096;
+  return ret;
+}
+
+//===---------------------------------------------------------------------===//
+
 "gas" uses this idiom:
   else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
 ..
@@ -1756,43 +1795,6 @@ case it choses instead to keep the max operation obvious.
 
 //===---------------------------------------------------------------------===//
 
-Take the following testcase on x86-64 (similar testcases exist for all targets
-with addc/adde):
-
-define void @a(i64* nocapture %s, i64* nocapture %t, i64 %a, i64 %b,
-i64 %c) nounwind {
-entry:
- %0 = zext i64 %a to i128                        ; <i128> [#uses=1]
- %1 = zext i64 %b to i128                        ; <i128> [#uses=1]
- %2 = add i128 %1, %0                            ; <i128> [#uses=2]
- %3 = zext i64 %c to i128                        ; <i128> [#uses=1]
- %4 = shl i128 %3, 64                            ; <i128> [#uses=1]
- %5 = add i128 %4, %2                            ; <i128> [#uses=1]
- %6 = lshr i128 %5, 64                           ; <i128> [#uses=1]
- %7 = trunc i128 %6 to i64                       ; <i64> [#uses=1]
- store i64 %7, i64* %s, align 8
- %8 = trunc i128 %2 to i64                       ; <i64> [#uses=1]
- store i64 %8, i64* %t, align 8
- ret void
-}
-
-Generated code:
-        addq   %rcx, %rdx
-        sbbq   %rax, %rax
-        subq   %rax, %r8
-        movq   %r8, (%rdi)
-        movq   %rdx, (%rsi)
-        ret
-
-Expected code:
-       addq    %rcx, %rdx
-       adcq    $0, %r8
-       movq    %r8, (%rdi)
-       movq    %rdx, (%rsi)
-       ret
-
-//===---------------------------------------------------------------------===//
-
 Switch lowering generates less than ideal code for the following switch:
 define void @a(i32 %x) nounwind {
 entry:
@@ -2225,16 +2227,6 @@ avoids partial register stalls in some important cases.
 
 //===---------------------------------------------------------------------===//
 
-We should fold compares like this:
-
- %1266 = add nsw i32 %.84.i.i.i, 1
- %560 = add nsw i32 %556, 1
- %1267 = icmp slt i32 %1266, %560
-
-to a single 'icmp slt' when the add's have a single use, since they are NSW.
-
-//===---------------------------------------------------------------------===//
-
 We don't fold (icmp (add) (add)) unless the two adds only have a single use.
 There are a lot of cases that we're refusing to fold in (e.g.) 256.bzip2, for
 example: