AMDPGU/SI: Use AssertZext node to mask high bit for scratch offsets

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

diff --git a/lib/Target/README.txt b/lib/Target/README.txt
index 33941f451a0cab6389313149c34c8f9d1ea24155..282d9234c1a5a5db8d96e5292b690b28911e92ab 100644 (file)
--- a/lib/Target/README.txt
+++ b/lib/Target/README.txt
@@ -2,22 +2,6 @@ Target Independent Opportunities:
 
 //===---------------------------------------------------------------------===//
 
 
 //===---------------------------------------------------------------------===//
 

-With the recent changes to make the implicit def/use set explicit in
-machineinstrs, we should change the target descriptions for 'call' instructions
-so that the .td files don't list all the call-clobbered registers as implicit
-defs.  Instead, these should be added by the code generator (e.g. on the dag).
-
-This has a number of uses:
-
-1. PPC32/64 and X86 32/64 can avoid having multiple copies of call instructions
-   for their different impdef sets.
-2. Targets with multiple calling convs (e.g. x86) which have different clobber
-   sets don't need copies of call instructions.
-3. 'Interprocedural register allocation' can be done to reduce the clobber sets
-   of calls.
-
-//===---------------------------------------------------------------------===//
-

 We should recognized various "overflow detection" idioms and translate them into
 llvm.uadd.with.overflow and similar intrinsics.  Here is a multiply idiom:
 
 We should recognized various "overflow detection" idioms and translate them into
 llvm.uadd.with.overflow and similar intrinsics.  Here is a multiply idiom:
 


@@ -109,44 +93,6 @@ This requires reassociating to forms of expressions that are already available,
 something that reassoc doesn't think about yet.
 
 
 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:
 //===---------------------------------------------------------------------===//
 
 These two functions should generate the same code on big-endian systems:


@@ -168,7 +114,7 @@ stuff too.
 
 //===---------------------------------------------------------------------===//
 
 
 //===---------------------------------------------------------------------===//
 

-For vector types, TargetData.cpp::getTypeInfo() returns alignment that is equal
+For vector types, DataLayout.cpp::getTypeInfo() returns alignment that is equal

 to the type size. It works but can be overly conservative as the alignment of
 specific vector types are target dependent.
 
 to the type size. It works but can be overly conservative as the alignment of
 specific vector types are target dependent.
 


@@ -278,22 +224,7 @@ unsigned countbits_slow(unsigned v) {
     c += v & 1;
   return c;
 }
     c += v & 1;
   return c;
 }

-unsigned countbits_fast(unsigned v){
-  unsigned c;
-  for (c = 0; v; c++)
-    v &= v - 1; // clear the least significant bit set
-  return c;
-}

 
 

-BITBOARD = unsigned long long
-int PopCnt(register BITBOARD a) {
-  register int c=0;
-  while(a) {
-    c++;
-    a &= a - 1;
-  }
-  return c;
-}

 unsigned int popcount(unsigned int input) {
   unsigned int count = 0;
   for (unsigned int i =  0; i < 4 * 8; i++)
 unsigned int popcount(unsigned int input) {
   unsigned int count = 0;
   for (unsigned int i =  0; i < 4 * 8; i++)


@@ -802,7 +733,7 @@ f (unsigned long a, unsigned long b, unsigned long c)
   return ((a & (c - 1)) != 0) | ((b & (c - 1)) != 0);
 }
 Both should combine to ((a|b) & (c-1)) != 0.  Currently not optimized with
   return ((a & (c - 1)) != 0) | ((b & (c - 1)) != 0);
 }
 Both should combine to ((a|b) & (c-1)) != 0.  Currently not optimized with

-"clang -emit-llvm-bc | opt -std-compile-opts".
+"clang -emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 
 //===---------------------------------------------------------------------===//
 


@@ -815,7 +746,7 @@ void clear_pmd_range(unsigned long start, unsigned long end)
 }
 The expression should optimize to something like
 "!((start|end)&~PMD_MASK). Currently not optimized with "clang
 }
 The expression should optimize to something like
 "!((start|end)&~PMD_MASK). Currently not optimized with "clang

--emit-llvm-bc | opt -std-compile-opts".
+-emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 
 //===---------------------------------------------------------------------===//
 


@@ -834,7 +765,7 @@ int f(int x, int y)
  return (abs(x)) >= 0;
 }
 This should optimize to x == INT_MIN. (With -fwrapv.)  Currently not
  return (abs(x)) >= 0;
 }
 This should optimize to x == INT_MIN. (With -fwrapv.)  Currently not

-optimized with "clang -emit-llvm-bc | opt -std-compile-opts".
+optimized with "clang -emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 
 //===---------------------------------------------------------------------===//
 


@@ -870,99 +801,119 @@ rshift_gt (unsigned int a)
    bar ();
 }
 
    bar ();
 }
 

-void neg_eq_cst(unsigned int a) {
-if (-a == 123)
-bar();
-}
-

 All should simplify to a single comparison.  All of these are
 currently not optimized with "clang -emit-llvm-bc | opt
 All should simplify to a single comparison.  All of these are
 currently not optimized with "clang -emit-llvm-bc | opt

--std-compile-opts".
+-O3".

 
 //===---------------------------------------------------------------------===//
 
 From GCC Bug 32605:
 int c(int* x) {return (char*)x+2 == (char*)x;}
 Should combine to 0.  Currently not optimized with "clang
 
 //===---------------------------------------------------------------------===//
 
 From GCC Bug 32605:
 int c(int* x) {return (char*)x+2 == (char*)x;}
 Should combine to 0.  Currently not optimized with "clang

--emit-llvm-bc | opt -std-compile-opts" (although llc can optimize it).
+-emit-llvm-bc | opt -O3" (although llc can optimize it).

 
 //===---------------------------------------------------------------------===//
 
 int a(unsigned b) {return ((b << 31) | (b << 30)) >> 31;}
 Should be combined to  "((b >> 1) | b) & 1".  Currently not optimized
 
 //===---------------------------------------------------------------------===//
 
 int a(unsigned b) {return ((b << 31) | (b << 30)) >> 31;}
 Should be combined to  "((b >> 1) | b) & 1".  Currently not optimized

-with "clang -emit-llvm-bc | opt -std-compile-opts".
+with "clang -emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 unsigned a(unsigned x, unsigned y) { return x | (y & 1) | (y & 2);}
 Should combine to "x | (y & 3)".  Currently not optimized with "clang
 
 //===---------------------------------------------------------------------===//
 
 unsigned a(unsigned x, unsigned y) { return x | (y & 1) | (y & 2);}
 Should combine to "x | (y & 3)".  Currently not optimized with "clang

--emit-llvm-bc | opt -std-compile-opts".
+-emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 int a(int a, int b, int c) {return (~a & c) | ((c|a) & b);}
 Should fold to "(~a & c) | (a & b)".  Currently not optimized with
 
 //===---------------------------------------------------------------------===//
 
 int a(int a, int b, int c) {return (~a & c) | ((c|a) & b);}
 Should fold to "(~a & c) | (a & b)".  Currently not optimized with

-"clang -emit-llvm-bc | opt -std-compile-opts".
+"clang -emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 int a(int a,int b) {return (~(a|b))|a;}
 Should fold to "a|~b".  Currently not optimized with "clang
 
 //===---------------------------------------------------------------------===//
 
 int a(int a,int b) {return (~(a|b))|a;}
 Should fold to "a|~b".  Currently not optimized with "clang

--emit-llvm-bc | opt -std-compile-opts".
+-emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 int a(int a, int b) {return (a&&b) || (a&&!b);}
 Should fold to "a".  Currently not optimized with "clang -emit-llvm-bc
 
 //===---------------------------------------------------------------------===//
 
 int a(int a, int b) {return (a&&b) || (a&&!b);}
 Should fold to "a".  Currently not optimized with "clang -emit-llvm-bc

-| opt -std-compile-opts".
+| opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 int a(int a, int b, int c) {return (a&&b) || (!a&&c);}
 Should fold to "a ? b : c", or at least something sane.  Currently not
 
 //===---------------------------------------------------------------------===//
 
 int a(int a, int b, int c) {return (a&&b) || (!a&&c);}
 Should fold to "a ? b : c", or at least something sane.  Currently not

-optimized with "clang -emit-llvm-bc | opt -std-compile-opts".
+optimized with "clang -emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 int a(int a, int b, int c) {return (a&&b) || (a&&c) || (a&&b&&c);}
 Should fold to a && (b || c).  Currently not optimized with "clang
 
 //===---------------------------------------------------------------------===//
 
 int a(int a, int b, int c) {return (a&&b) || (a&&c) || (a&&b&&c);}
 Should fold to a && (b || c).  Currently not optimized with "clang

--emit-llvm-bc | opt -std-compile-opts".
+-emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 int a(int x) {return x | ((x & 8) ^ 8);}
 Should combine to x | 8.  Currently not optimized with "clang
 
 //===---------------------------------------------------------------------===//
 
 int a(int x) {return x | ((x & 8) ^ 8);}
 Should combine to x | 8.  Currently not optimized with "clang

--emit-llvm-bc | opt -std-compile-opts".
+-emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 int a(int x) {return x ^ ((x & 8) ^ 8);}
 Should also combine to x | 8.  Currently not optimized with "clang
 
 //===---------------------------------------------------------------------===//
 
 int a(int x) {return x ^ ((x & 8) ^ 8);}
 Should also combine to x | 8.  Currently not optimized with "clang

--emit-llvm-bc | opt -std-compile-opts".
+-emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 int a(int x) {return ((x | -9) ^ 8) & x;}
 Should combine to x & -9.  Currently not optimized with "clang
 
 //===---------------------------------------------------------------------===//
 
 int a(int x) {return ((x | -9) ^ 8) & x;}
 Should combine to x & -9.  Currently not optimized with "clang

--emit-llvm-bc | opt -std-compile-opts".
+-emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 unsigned a(unsigned a) {return a * 0x11111111 >> 28 & 1;}
 Should combine to "a * 0x88888888 >> 31".  Currently not optimized
 
 //===---------------------------------------------------------------------===//
 
 unsigned a(unsigned a) {return a * 0x11111111 >> 28 & 1;}
 Should combine to "a * 0x88888888 >> 31".  Currently not optimized

-with "clang -emit-llvm-bc | opt -std-compile-opts".
+with "clang -emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 unsigned a(char* x) {if ((*x & 32) == 0) return b();}
 There's an unnecessary zext in the generated code with "clang
 
 //===---------------------------------------------------------------------===//
 
 unsigned a(char* x) {if ((*x & 32) == 0) return b();}
 There's an unnecessary zext in the generated code with "clang

--emit-llvm-bc | opt -std-compile-opts".
+-emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 unsigned a(unsigned long long x) {return 40 * (x >> 1);}
 Should combine to "20 * (((unsigned)x) & -2)".  Currently not
 
 //===---------------------------------------------------------------------===//
 
 unsigned a(unsigned long long x) {return 40 * (x >> 1);}
 Should combine to "20 * (((unsigned)x) & -2)".  Currently not

-optimized with "clang -emit-llvm-bc | opt -std-compile-opts".
+optimized with "clang -emit-llvm-bc | opt -O3".
+
+//===---------------------------------------------------------------------===//
+
+int g(int x) { return (x - 10) < 0; }
+Should combine to "x <= 9" (the sub has nsw).  Currently not
+optimized with "clang -emit-llvm-bc | opt -O3".
+
+//===---------------------------------------------------------------------===//
+
+int g(int x) { return (x + 10) < 0; }
+Should combine to "x < -10" (the add has nsw).  Currently not
+optimized with "clang -emit-llvm-bc | opt -O3".
+
+//===---------------------------------------------------------------------===//
+
+int f(int i, int j) { return i < j + 1; }
+int g(int i, int j) { return j > i - 1; }
+Should combine to "i <= j" (the add/sub has nsw).  Currently not
+optimized with "clang -emit-llvm-bc | opt -O3".
+
+//===---------------------------------------------------------------------===//
+
+unsigned f(unsigned x) { return ((x & 7) + 1) & 15; }
+The & 15 part should be optimized away, it doesn't change the result. Currently
+not optimized with "clang -emit-llvm-bc | opt -O3".

 
 //===---------------------------------------------------------------------===//
 
 
 //===---------------------------------------------------------------------===//
 


@@ -1174,7 +1125,7 @@ There are many load PRE testcases in testsuite/gcc.dg/tree-ssa/loadpre* in the
 GCC testsuite, ones we don't get yet are (checked through loadpre25):
 
 [CRIT EDGE BREAKING]
 GCC testsuite, ones we don't get yet are (checked through loadpre25):
 
 [CRIT EDGE BREAKING]

-loadpre3.c predcom-4.c
+predcom-4.c

 
 [PRE OF READONLY CALL]
 loadpre5.c
 
 [PRE OF READONLY CALL]
 loadpre5.c


@@ -1317,7 +1268,8 @@ int foo (void) {
 ..
   else if (strchr ("<>", *intel_parser.op_string)
 
 ..
   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.

 
 //===---------------------------------------------------------------------===//
 
 
 //===---------------------------------------------------------------------===//
 


@@ -1767,7 +1719,6 @@ case it choses instead to keep the max operation obvious.
 
 //===---------------------------------------------------------------------===//
 
 
 //===---------------------------------------------------------------------===//
 

-Switch lowering generates less than ideal code for the following switch:

 define void @a(i32 %x) nounwind {
 entry:
   switch i32 %x, label %if.end [
 define void @a(i32 %x) nounwind {
 entry:
   switch i32 %x, label %if.end [


@@ -1788,19 +1739,15 @@ declare void @foo()
 Generated code on x86-64 (other platforms give similar results):
 a:
        cmpl    $5, %edi
 Generated code on x86-64 (other platforms give similar results):
 a:
        cmpl    $5, %edi

-       ja      .LBB0_2
-       movl    %edi, %eax
-       movl    $47, %ecx
-       btq     %rax, %rcx
-       jb      .LBB0_3
+       ja      LBB2_2
+       cmpl    $4, %edi
+       jne     LBB2_3

 .LBB0_2:
        ret
 .LBB0_3:
        jmp     foo  # TAILCALL
 
 .LBB0_2:
        ret
 .LBB0_3:
        jmp     foo  # TAILCALL
 

-The movl+movl+btq+jb could be simplified to a cmpl+jne.
-
-Or, if we wanted to be really clever, we could simplify the whole thing to
+If we wanted to be really clever, we could simplify the whole thing to

 something like the following, which eliminates a branch:
        xorl    $1, %edi
        cmpl    $4, %edi
 something like the following, which eliminates a branch:
        xorl    $1, %edi
        cmpl    $4, %edi


@@ -1897,44 +1844,6 @@ we remove checking in code like
 
 //===---------------------------------------------------------------------===//
 
 
 //===---------------------------------------------------------------------===//
 

-This code (from Benchmarks/Dhrystone/dry.c):
-
-define i32 @Func1(i32, i32) nounwind readnone optsize ssp {
-entry:
-  %sext = shl i32 %0, 24
-  %conv = ashr i32 %sext, 24
-  %sext6 = shl i32 %1, 24
-  %conv4 = ashr i32 %sext6, 24
-  %cmp = icmp eq i32 %conv, %conv4
-  %. = select i1 %cmp, i32 10000, i32 0
-  ret i32 %.
-}
-
-Should be simplified into something like:
-
-define i32 @Func1(i32, i32) nounwind readnone optsize ssp {
-entry:
-  %sext = shl i32 %0, 24
-  %conv = and i32 %sext, 0xFF000000
-  %sext6 = shl i32 %1, 24
-  %conv4 = and i32 %sext6, 0xFF000000
-  %cmp = icmp eq i32 %conv, %conv4
-  %. = select i1 %cmp, i32 10000, i32 0
-  ret i32 %.
-}
-
-and then to:
-
-define i32 @Func1(i32, i32) nounwind readnone optsize ssp {
-entry:
-  %conv = and i32 %0, 0xFF
-  %conv4 = and i32 %1, 0xFF
-  %cmp = icmp eq i32 %conv, %conv4
-  %. = select i1 %cmp, i32 10000, i32 0
-  ret i32 %.
-}
-//===---------------------------------------------------------------------===//
-

 clang -O3 currently compiles this code
 
 int g(unsigned int a) {
 clang -O3 currently compiles this code
 
 int g(unsigned int a) {


@@ -2238,4 +2147,133 @@ missed cases:
 
 //===---------------------------------------------------------------------===//
 
 
 //===---------------------------------------------------------------------===//
 

+define i1 @test1(i32 %x) nounwind {
+  %and = and i32 %x, 3
+  %cmp = icmp ult i32 %and, 2
+  ret i1 %cmp
+}
+
+Can be folded to (x & 2) == 0.
+
+define i1 @test2(i32 %x) nounwind {
+  %and = and i32 %x, 3
+  %cmp = icmp ugt i32 %and, 1
+  ret i1 %cmp
+}
+
+Can be folded to (x & 2) != 0.
+
+SimplifyDemandedBits shrinks the "and" constant to 2 but instcombine misses the
+icmp transform.
+
+//===---------------------------------------------------------------------===//
+
+This code:
+
+typedef struct {
+int f1:1;
+int f2:1;
+int f3:1;
+int f4:29;
+} t1;
+
+typedef struct {
+int f1:1;
+int f2:1;
+int f3:30;
+} t2;
+
+t1 s1;
+t2 s2;
+
+void func1(void)
+{
+s1.f1 = s2.f1;
+s1.f2 = s2.f2;
+}
+
+Compiles into this IR (on x86-64 at least):
+
+%struct.t1 = type { i8, [3 x i8] }
+@s2 = global %struct.t1 zeroinitializer, align 4
+@s1 = global %struct.t1 zeroinitializer, align 4
+define void @func1() nounwind ssp noredzone {
+entry:
+  %0 = load i32* bitcast (%struct.t1* @s2 to i32*), align 4
+  %bf.val.sext5 = and i32 %0, 1
+  %1 = load i32* bitcast (%struct.t1* @s1 to i32*), align 4
+  %2 = and i32 %1, -4
+  %3 = or i32 %2, %bf.val.sext5
+  %bf.val.sext26 = and i32 %0, 2
+  %4 = or i32 %3, %bf.val.sext26
+  store i32 %4, i32* bitcast (%struct.t1* @s1 to i32*), align 4
+  ret void
+}
+
+The two or/and's should be merged into one each.
+
+//===---------------------------------------------------------------------===//
+
+Machine level code hoisting can be useful in some cases.  For example, PR9408
+is about:
+
+typedef union {
+ void (*f1)(int);
+ void (*f2)(long);
+} funcs;
+
+void foo(funcs f, int which) {
+ int a = 5;
+ if (which) {
+   f.f1(a);
+ } else {
+   f.f2(a);
+ }
+}
+
+which we compile to:
+
+foo:                                    # @foo
+# BB#0:                                 # %entry
+       pushq   %rbp
+       movq    %rsp, %rbp
+       testl   %esi, %esi
+       movq    %rdi, %rax
+       je      .LBB0_2
+# BB#1:                                 # %if.then
+       movl    $5, %edi
+       callq   *%rax
+       popq    %rbp
+       ret
+.LBB0_2:                                # %if.else
+       movl    $5, %edi
+       callq   *%rax
+       popq    %rbp
+       ret
+
+Note that bb1 and bb2 are the same.  This doesn't happen at the IR level
+because one call is passing an i32 and the other is passing an i64.

 
 

+//===---------------------------------------------------------------------===//
+
+I see this sort of pattern in 176.gcc in a few places (e.g. the start of
+store_bit_field).  The rem should be replaced with a multiply and subtract:
+
+  %3 = sdiv i32 %A, %B
+  %4 = srem i32 %A, %B
+
+Similarly for udiv/urem.  Note that this shouldn't be done on X86 or ARM,
+which can do this in a single operation (instruction or libcall).  It is
+probably best to do this in the code generator.
+
+//===---------------------------------------------------------------------===//
+
+unsigned foo(unsigned x, unsigned y) { return (x & y) == 0 || x == 0; }
+should fold to (x & y) == 0.
+
+//===---------------------------------------------------------------------===//
+
+unsigned foo(unsigned x, unsigned y) { return x > y && x != 0; }
+should fold to x > y.
+
+//===---------------------------------------------------------------------===//