704437 - Test issue: run-time vs. compile-time evaluation in Float and Float4 tests

Status: RESOLVED WONTFIX

(Tamarin Graveyard :: Baseline JIT (CodegenLIR), defect, P3)

Description

[William Maddox]

The following test case revealed an error in constant folding for unary
negation of float4 values.  The error was actually found by inspection,
and it was with some fussing around that I produced the test case.  In
particular, using -Ojit did not allow folding of the negations in the
top-level script.  Examination of the acceptance suite suggests that we have
not systematically attempted to control for whether compile-time evaluation
would take place, meaning that in some cases, we are likely testing either
the run-time or the compile-time case and leaving a hole in coverage.

// Test sign of unary negation of Float4

var flt4:float4;

function neg1():float4 { 
   var f4:float4 = new float4(0f);
   return -f4; // folded
}
trace(1/neg1());

flt4 = new float4(0f);
flt4 = -flt4;  // not folded
trace(1/flt4);

function neg2():float4 { 
   var f4:float4 = new float4(-0f);
   return -f4;  // folded
}
trace(1/neg2());

flt4 = new float4(-0f);
flt4 = -flt4; // not folded
trace(1/flt4);

Comment 1

[Lars T Hansen]

Out of curiosity, does the "new" make any difference here?  It should not - float4(x) is the same as new float4(x) in all contexts; the canonical expression is the former.

Also, if x is constant then ASC is required to turn either expression into a pushfloat4, basically, though it will not perform the folding of negate for "var" bindings, only for "const" bindings.

Comment 2

[William Maddox]

(In reply to Lars T Hansen from comment #1)
> Out of curiosity, does the "new" make any difference here?  It should not -
> float4(x) is the same as new float4(x) in all contexts; the canonical
> expression is the former.

I used the 'new float4()' form simply because the test case was derived from
one of the acceptance tests that did so: as3/Types/Float4/op_negate.as .