coalescing as a separate pass rather than inside of
LiveIntervalAnalysis.

llvm-svn: 59146

no longer records a unique defining instruction, and virtual
registers may have multiple kills while still being defined
and killed in the same block.

llvm-svn: 59145

llvm-svn: 59143

llvm-svn: 59142

When constructing std::strings from C strings, we should check the input
value to be not NULL so that the std::string constructor does not
segfault.
Fixes #3047.

llvm-svn: 59131

llvm-svn: 59130

llvm-svn: 59129

what "cond.?" means, and this avoids quoting).

llvm-svn: 59128

llvm-svn: 59127

llvm-svn: 59126

This pushes it a lot closer to being able to deal with most of the stuff 
CodeGen's constant expression evaluator knows how to deal with.  This 
also fixes PR3003.

The test could possibly use some improvement, but this'll work for now.  
Test 6 is inspired by PR3003; the other tests are mostly just designed
to exercise the new code.  The reason for the funny structure of the 
tests is that type fixing for arrays inside of structs is the only place 
in Sema that calls tryEvaluate, at least for the moment.

llvm-svn: 59125

This cuts another 200 lines off expr.ll, forming 23 selects.

llvm-svn: 59124

This happens for stuff like this:

x = cond1 || cond2 || cond3 || cond4;

llvm-svn: 59123

that it no longer handles non-power-of-two vectors.
However it previously only handled them sometimes,
depending on obscure numerical relationships between
the index and vector type.  For example, for a vector
of length 6, it would succeed if and only if the
index was an even multiple of 6.  I consider this
more confusing than useful.

llvm-svn: 59122

constant folding.

llvm-svn: 59121

when the target does not support ADDC/SUBC.  This
fixes PR3044.

llvm-svn: 59120

llvm-svn: 59119

 - Split out "simple" statements which can easily handle IR generation
   when there is no insert point. These are generally statements which
   start by emitting a new block or are only containers for other
   statements.

 - This fixes a regression in emitting dummy blocks, notably for case
   statements.

 - This also fixes spurious emission of a number of debug stoppoint
   intrinsic instructions.

Remove unneeded sw.body block, just clear the insertion point.

Lift out CodeGenFunction::EmitStopPoint which calls into the
CGDebugInfo class when generating debug info.

Normalize definitions of Emit{Break,Continue}Stmt and usage of
ErrorUnsupported.

llvm-svn: 59118

llvm-svn: 59117

llvm-svn: 59116

codegen stuff like "if (!(X && Y))"

llvm-svn: 59115

shrinks code yet again by a bit.

llvm-svn: 59114

CodeGenFunction.cpp.  Change VisitConditionalOperator to use
constant fold instead of codegen'ing a constant conditional.

Change ForStmt to use EmitBranchOnBoolExpr, this shrinks
expr.c very slightly to 40239 lines.

llvm-svn: 59113

llvm-svn: 59112

have a condition that is an &&/||.  Before we used to compile things like this:

int test() {
  if (x && y) foo(); else bar();
}

into:

	%0 = load i32* @x		; <i32> [#uses=1]
	%1 = icmp ne i32 %0, 0		; <i1> [#uses=1]
	br i1 %1, label %land_rhs, label %land_cont

land_rhs:		; preds = %entry
	%2 = load i32* @y		; <i32> [#uses=1]
	%3 = icmp ne i32 %2, 0		; <i1> [#uses=1]
	br label %land_cont

land_cont:		; preds = %land_rhs, %entry
	%4 = phi i1 [ false, %entry ], [ %3, %land_rhs ]		; <i1> [#uses=1]
	br i1 %4, label %ifthen, label %ifelse

ifthen:		; preds = %land_cont
	%call = call i32 (...)* @foo()		; <i32> [#uses=0]
	br label %ifend

ifelse:		; preds = %land_cont
	%call1 = call i32 (...)* @bar()		; <i32> [#uses=0]
	br label %ifend

ifend:		; preds = %ifelse, %ifthen


Now we turn it into the much more svelte code:

        %0 = load i32* @x               ; <i32> [#uses=1]
        %1 = icmp ne i32 %0, 0          ; <i1> [#uses=1]
        br i1 %1, label %land_lhs_true, label %ifelse

land_lhs_true:          ; preds = %entry
        %2 = load i32* @y               ; <i32> [#uses=1]
        %3 = icmp ne i32 %2, 0          ; <i1> [#uses=1]
        br i1 %3, label %ifthen, label %ifelse

ifthen:         ; preds = %land_lhs_true
        %call = call i32 (...)* @foo()          ; <i32> [#uses=0]
        br label %ifend

ifelse:         ; preds = %land_lhs_true, %entry
        %call1 = call i32 (...)* @bar()         ; <i32> [#uses=0]
        br label %ifend

ifend:          ; preds = %ifelse, %ifthen

Note the lack of a phi node.

This shrinks the -O0 .ll file for 176.gcc/expr.c from 43176 to 40267 lines.

llvm-svn: 59111

crashing because we errors are ignored in subexpressions that are not evaluated,
but we still evaluate the result of parents.  This would cause an assertion 
because the erroneous subexpr didn't have its result value set to the right type.

llvm-svn: 59110

llvm-svn: 59109

llvm-svn: 59107

llvm-svn: 59106

llvm-svn: 59105

llvm-svn: 59104

llvm-svn: 59103

llvm-svn: 59102

Reenable test.

llvm-svn: 59101

llvm-svn: 59100

llvm-svn: 59099

llvm-svn: 59098

Only the tools that are mentioned in the compilation graph definition
are now inserted by PopulateCompilationGraph(). This should cut down
plugin loading time a little.

llvm-svn: 59097

llvm-svn: 59096

llvm-svn: 59095