Subzero: Fix floating-point constant pooling.

tests_lit/llvm2ice_tests/fp_const_pool.ll

+; This tests that different floating point constants (such as 0.0 and -0.0)
+; remain distinct even when they sort of look equal, and also that different
+; instances of the same floating point constant (such as NaN and NaN) get the
+; same constant pool entry even when "a==a" would suggest they are different.
+
+; REQUIRES: allow_dump
+
+define void @consume_float(float %f) {
+  ret void
+}
+
+define void @consume_double(double %d) {
+  ret void
+}
+
+define void @test_zeros() {
+entry:
+  call void @consume_float(float 0.0)
+  call void @consume_float(float -0.0)
+  call void @consume_double(double 0.0)
+  call void @consume_double(double -0.0)
+  ret void
+}
+; RUN: %p2i -i %s --insts | FileCheck --check-prefix=ZERO %s

[JF, 2015/03/19 16:33:09]

Why different RUN lines? It seems like you could just use the one below, not
this one, and encode zero as 0x0000000000000000 and minus zero as
0x8000000000000000.

[Jim Stichnoth, 2015/03/19 18:07:04]

I tried to clarify this in comments, ptal.  The ZERO test style is preferable
because it very precisely tests whether +/-0.0 have been incorrectly merged. 
For the NANS tests, though, we can't tell from the dump output whether two equal
constant values share a unique pool entry, so we have to rely on the downstream
translation output.

The ZERO test just checks that 

+; ZERO: test_zeros
+; ZERO-NEXT: entry:
+; ZERO-NEXT: call void @consume_float(float 0.0
+; ZERO-NEXT: call void @consume_float(float -0.0
+; ZERO-NEXT: call void @consume_double(double 0.0
+; ZERO-NEXT: call void @consume_double(double -0.0
+
+
+define void @test_nans() {
+entry:
+  call void @consume_float(float 0x7FF8000000000000)
+  call void @consume_float(float 0x7FF8000000000000)
+  call void @consume_float(float 0xFFF8000000000000)
+  call void @consume_float(float 0xFFF8000000000000)
+  call void @consume_double(double 0x7FF8000000000000)
+  call void @consume_double(double 0x7FF8000000000000)
+  call void @consume_double(double 0xFFF8000000000000)
+  call void @consume_double(double 0xFFF8000000000000)
+  ret void
+}
+; RUN: %p2i -i %s --filetype=asm --llvm-source \
+; RUN:   | FileCheck --check-prefix=NANS1 %s
+; NANS1: float nan
+; NANS1-NOT: float nan
+
+; RUN: %p2i -i %s --filetype=asm --llvm-source \
+; RUN:   | FileCheck --check-prefix=NANS2 %s
+; NANS2: float -nan
+; NANS2-NOT: float -nan
+
+; RUN: %p2i -i %s --filetype=asm --llvm-source \
+; RUN:   | FileCheck --check-prefix=NANS3 %s
+; NANS3: double nan
+; NANS3-NOT: double nan
+
+; RUN: %p2i -i %s --filetype=asm --llvm-source \
+; RUN:   | FileCheck --check-prefix=NANS4 %s
+; NANS4: double -nan
+; NANS4-NOT: double -nan