[SimplifyLibCalls] Turn memchr(const, C, const) into a bitfield check.

strchr("123!", C) != nullptr is a common pattern to check if C is one
of 1, 2, 3 or !. If the largest element of the string is smaller than
the target's register size we can easily create a bitfield and just
do a simple test for set membership.

int foo(char C) { return strchr("123!", C) != nullptr; } now becomes

	cmpl	$64, %edi ## range check
	sbbb	%al, %al
	movabsq	$0xE000200000001, %rcx
	btq	%rdi, %rcx ## bit test
	sbbb	%cl, %cl
	andb	%al, %cl ## and the two conditions
	andb	$1, %cl
	movzbl	%cl, %eax ## returning an int
	ret

(imho the backend should expand this into a series of branches, but
that's a different story)

The code is currently limited to bit fields that fit in a register, so
usually 64 or 32 bits. Sadly, this misses anything using alpha chars
or {}. This could be fixed by just emitting a i128 bit field, but that
can generate really ugly code so we have to find a better way. To some
degree this is also recreating switch lowering logic, but we can't
simply emit a switch instruction and thus change the CFG within
instcombine.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@232902 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/README.txt b/lib/Target/README.txt
index 0fa56e6..52a334a 100644 (file)
--- a/lib/Target/README.txt
+++ b/lib/Target/README.txt
@@ -1268,7 +1268,8 @@ int foo (void) {
 ..
   else if (strchr ("<>", *intel_parser.op_string)
 
-Those should be turned into a switch.
+Those should be turned into a switch.  SimplifyLibCalls only gets the second
+case.
 
 //===---------------------------------------------------------------------===//
 

diff --git a/lib/Transforms/Utils/SimplifyLibCalls.cpp b/lib/Transforms/Utils/SimplifyLibCalls.cpp
index 0d0b77a..f6cc431 100644 (file)
--- a/lib/Transforms/Utils/SimplifyLibCalls.cpp
+++ b/lib/Transforms/Utils/SimplifyLibCalls.cpp
@@ -762,11 +762,9 @@ Value *LibCallSimplifier::optimizeMemChr(CallInst *CI, IRBuilder<> &B) {
   if (LenC && LenC->isNullValue())
     return Constant::getNullValue(CI->getType());
 
-  // Check if all arguments are constants.  If so, we can constant fold.
+  // From now on we need at least constant length and string.
   StringRef Str;
-  if (!CharC || !LenC ||
-      !getConstantStringInfo(SrcStr, Str, /*Offset=*/0,
-                             /*TrimAtNul=*/false))
+  if (!LenC || !getConstantStringInfo(SrcStr, Str, 0, /*TrimAtNul=*/false))
     return nullptr;
 
   // Truncate the string to LenC. If Str is smaller than LenC we will still only
@@ -774,6 +772,54 @@ Value *LibCallSimplifier::optimizeMemChr(CallInst *CI, IRBuilder<> &B) {
   // return null if we don't find the char.
   Str = Str.substr(0, LenC->getZExtValue());
 
+  // If the char is variable but the input str and length are not we can turn
+  // this memchr call into a simple bit field test. Of course this only works
+  // when the return value is only checked against null.
+  //
+  // It would be really nice to reuse switch lowering here but we can't change
+  // the CFG at this point.
+  //
+  // memchr("\r\n", C, 2) != nullptr -> (C & ((1 << '\r') | (1 << '\n'))) != 0
+  //   after bounds check.
+  if (!CharC && !Str.empty() && isOnlyUsedInZeroEqualityComparison(CI)) {
+    unsigned char Max = *std::max_element(Str.begin(), Str.end());
+
+    // Make sure the bit field we're about to create fits in a register on the
+    // target.
+    // FIXME: On a 64 bit architecture this prevents us from using the
+    // interesting range of alpha ascii chars. We could do better by emitting
+    // two bitfields or shifting the range by 64 if no lower chars are used.
+    if (!DL.fitsInLegalInteger(Max + 1))
+      return nullptr;
+
+    // For the bit field use a power-of-2 type with at least 8 bits to avoid
+    // creating unnecessary illegal types.
+    unsigned char Width = NextPowerOf2(std::max((unsigned char)7, Max));
+
+    // Now build the bit field.
+    APInt Bitfield(Width, 0);
+    for (char C : Str)
+      Bitfield.setBit((unsigned char)C);
+    Value *BitfieldC = B.getInt(Bitfield);
+
+    // First check that the bit field access is within bounds.
+    Value *C = B.CreateZExtOrTrunc(CI->getArgOperand(1), BitfieldC->getType());
+    Value *Bounds = B.CreateICmp(ICmpInst::ICMP_ULT, C, B.getIntN(Width, Width),
+                                 "memchr.bounds");
+
+    // Create code that checks if the given bit is set in the field.
+    Value *Shl = B.CreateShl(B.getIntN(Width, 1ULL), C);
+    Value *Bits = B.CreateIsNotNull(B.CreateAnd(Shl, BitfieldC), "memchr.bits");
+
+    // Finally merge both checks and cast to pointer type. The inttoptr
+    // implicitly zexts the i1 to intptr type.
+    return B.CreateIntToPtr(B.CreateAnd(Bounds, Bits, "memchr"), CI->getType());
+  }
+
+  // Check if all arguments are constants.  If so, we can constant fold.
+  if (!CharC)
+    return nullptr;
+
   // Compute the offset.
   size_t I = Str.find(CharC->getSExtValue() & 0xFF);
   if (I == StringRef::npos) // Didn't find the char.  memchr returns null.

diff --git a/test/Transforms/InstCombine/memchr.ll b/test/Transforms/InstCombine/memchr.ll
index 4cadbfa..6783249 100644 (file)
--- a/test/Transforms/InstCombine/memchr.ll
+++ b/test/Transforms/InstCombine/memchr.ll
@@ -1,11 +1,14 @@
 ; Test that the memchr library call simplifier works correctly.
 ; RUN: opt < %s -instcombine -S | FileCheck %s
 
-target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32"
 
 @hello = constant [14 x i8] c"hello world\5Cn\00"
 @hellonull = constant [14 x i8] c"hello\00world\5Cn\00"
 @null = constant [1 x i8] zeroinitializer
+@newlines = constant [3 x i8] c"\0D\0A\00"
+@single = constant [2 x i8] c"\1F\00"
+@spaces = constant [4 x i8] c" \0D\0A\00"
 @chp = global i8* zeroinitializer
 
 declare i8* @memchr(i8*, i32, i32)
@@ -119,3 +122,67 @@ define void @test9() {
   store i8* %dst, i8** @chp
   ret void
 }
+
+define void @test10() {
+; CHECK-LABEL: @test10
+; CHECK: store i8* null, i8** @chp, align 4
+; CHECK-NOT: call i8* @memchr
+; CHECK: ret void
+
+  %str = getelementptr [14 x i8], [14 x i8]* @hello, i32 0, i32 0
+  %dst = call i8* @memchr(i8* %str, i32 119, i32 6)
+  store i8* %dst, i8** @chp
+  ret void
+}
+
+; Check transformation memchr("\r\n", C, 2) != nullptr -> (C & 9216) != 0
+define i1 @test11(i32 %C) {
+; CHECK-LABEL: @test11
+; CHECK-NEXT: [[TRUNC:%.*]] = trunc i32 %C to i16
+; CHECK-NEXT: %memchr.bounds = icmp ult i16 [[TRUNC]], 16
+; CHECK-NEXT: [[SHL:%.*]] = shl i16 1, [[TRUNC]]
+; CHECK-NEXT: [[AND:%.*]] = and i16 [[SHL]], 9216
+; CHECK-NEXT: %memchr.bits = icmp ne i16 [[AND]], 0
+; CHECK-NEXT: %memchr = and i1 %memchr.bounds, %memchr.bits
+; CHECK-NEXT: ret i1 %memchr
+
+  %dst = call i8* @memchr(i8* getelementptr inbounds ([3 x i8], [3 x i8]* @newlines, i64 0, i64 0), i32 %C, i32 2)
+  %cmp = icmp ne i8* %dst, null
+  ret i1 %cmp
+}
+
+; No 64 bits here
+define i1 @test12(i32 %C) {
+; CHECK-LABEL: @test12
+; CHECK-NEXT: %dst = call i8* @memchr(i8* getelementptr inbounds ([4 x i8], [4 x i8]* @spaces, i32 0, i32 0), i32 %C, i32 3)
+; CHECK-NEXT: %cmp = icmp ne i8* %dst, null
+; CHECK-NEXT: ret i1 %cmp
+
+  %dst = call i8* @memchr(i8* getelementptr inbounds ([4 x i8], [4 x i8]* @spaces, i64 0, i64 0), i32 %C, i32 3)
+  %cmp = icmp ne i8* %dst, null
+  ret i1 %cmp
+}
+
+define i1 @test13(i32 %C) {
+; CHECK-LABEL: @test13
+; CHECK-NEXT: %memchr.bounds = icmp ult i32 %C, 32
+; CHECK-NEXT: [[SHL:%.*]] = shl i32 1, %C
+; CHECK-NEXT: [[AND:%.*]] = and i32 [[SHL]], -2147483647
+; CHECK-NEXT: %memchr.bits = icmp ne i32 [[AND]], 0
+; CHECK-NEXT: %memchr = and i1 %memchr.bounds, %memchr.bits
+; CHECK-NEXT: ret i1 %memchr
+
+  %dst = call i8* @memchr(i8* getelementptr inbounds ([2 x i8], [2 x i8]* @single, i64 0, i64 0), i32 %C, i32 2)
+  %cmp = icmp ne i8* %dst, null
+  ret i1 %cmp
+}
+
+define i1 @test14(i32 %C) {
+; CHECK-LABEL: @test14
+; CHECK-NEXT: icmp eq i32 %C, 31
+; CHECK-NEXT: ret
+
+  %dst = call i8* @memchr(i8* getelementptr inbounds ([2 x i8], [2 x i8]* @single, i64 0, i64 0), i32 %C, i32 1)
+  %cmp = icmp ne i8* %dst, null
+  ret i1 %cmp
+}

diff --git a/test/Transforms/InstCombine/strchr-1.ll b/test/Transforms/InstCombine/strchr-1.ll
index 27c9a59..6c10ebd 100644 (file)
--- a/test/Transforms/InstCombine/strchr-1.ll
+++ b/test/Transforms/InstCombine/strchr-1.ll
@@ -1,10 +1,11 @@
 ; Test that the strchr library call simplifier works correctly.
 ; RUN: opt < %s -instcombine -S | FileCheck %s
 
-target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32"
 
 @hello = constant [14 x i8] c"hello world\5Cn\00"
 @null = constant [1 x i8] zeroinitializer
+@newlines = constant [3 x i8] c"\0D\0A\00"
 @chp = global i8* zeroinitializer
 
 declare i8* @strchr(i8*, i32)
@@ -76,3 +77,19 @@ define void @test_simplify6(i8* %str) {
   store i8* %dst, i8** @chp
   ret void
 }
+
+; Check transformation strchr("\r\n", C) != nullptr -> (C & 9217) != 0
+define i1 @test_simplify7(i32 %C) {
+; CHECK-LABEL: @test_simplify7
+; CHECK-NEXT: [[TRUNC:%.*]] = trunc i32 %C to i16
+; CHECK-NEXT: %memchr.bounds = icmp ult i16 [[TRUNC]], 16
+; CHECK-NEXT: [[SHL:%.*]] = shl i16 1, [[TRUNC]]
+; CHECK-NEXT: [[AND:%.*]] = and i16 [[SHL]], 9217
+; CHECK-NEXT: %memchr.bits = icmp ne i16 [[AND]], 0
+; CHECK-NEXT: %memchr1 = and i1 %memchr.bounds, %memchr.bits
+; CHECK-NEXT: ret i1 %memchr1
+
+  %dst = call i8* @strchr(i8* getelementptr inbounds ([3 x i8], [3 x i8]* @newlines, i64 0, i64 0), i32 %C)
+  %cmp = icmp ne i8* %dst, null
+  ret i1 %cmp
+}