cmd/compile: don't allow NaNs in floating-point constant ops

We store 32-bit floating point constants in a 64-bit field, by
converting that 32-bit float to 64-bit float to store it, and convert
it back to use it.

That works for *almost* all floating-point constants. The exception is
signaling NaNs. The round trip described above means we can't represent
a 32-bit signaling NaN, because conversions strip the signaling bit.

To fix this issue, just forbid NaNs as floating-point constants in SSA
form. This shouldn't affect any real-world code, as people seldom
constant-propagate NaNs (except in test code).

Additionally, NaNs are somewhat underspecified (which of the many NaNs
do you get when dividing 0/0?), so when cross-compiling there's a
danger of using the compiler machine's NaN regime for some math, and
the target machine's NaN regime for other math. Better to use the
target machine's NaN regime always.

This has been a bug since 1.10, and there's an easy workaround
(declare a global varaible containing the signaling NaN pattern, and
use that as the argument to math.Float32frombits) so we'll fix it in
1.15.

Fixes #36400
Update #36399

Change-Id: Icf155e743281560eda2eed953d19a829552ccfda
Reviewed-on: https://go-review.googlesource.com/c/go/+/213477
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>

1 files changed, 31 insertions, 2 deletions

diff --git a/test/codegen/math.go b/test/codegen/math.go
index 80e5d60d96..1ebfda0405 100644
--- a/test/codegen/math.go
+++ b/test/codegen/math.go
@@ -151,13 +151,13 @@ func toFloat32(u32 uint32) float32 {
 func constantCheck64() bool {
 	// amd64:"MOVB\t[$]0",-"FCMP",-"MOVB\t[$]1"
 	// s390x:"MOV(B|BZ|D)\t[$]0,",-"FCMPU",-"MOV(B|BZ|D)\t[$]1,"
-	return 0.5 == float64(uint32(1)) || 1.5 > float64(uint64(1<<63)) || math.NaN() == math.NaN()
+	return 0.5 == float64(uint32(1)) || 1.5 > float64(uint64(1<<63))
 }
 
 func constantCheck32() bool {
 	// amd64:"MOVB\t[$]1",-"FCMP",-"MOVB\t[$]0"
 	// s390x:"MOV(B|BZ|D)\t[$]1,",-"FCMPU",-"MOV(B|BZ|D)\t[$]0,"
-	return float32(0.5) <= float32(int64(1)) && float32(1.5) >= float32(int32(-1<<31)) && float32(math.NaN()) != float32(math.NaN())
+	return float32(0.5) <= float32(int64(1)) && float32(1.5) >= float32(int32(-1<<31))
 }
 
 // Test that integer constants are converted to floating point constants
@@ -186,3 +186,32 @@ func constantConvertInt32(x uint32) uint32 {
 	}
 	return x
 }
+
+func nanGenerate64() float64 {
+	// Test to make sure we don't generate a NaN while constant propagating.
+	// See issue 36400.
+	zero := 0.0
+	// amd64:-"DIVSD"
+	inf := 1 / zero // +inf. We can constant propagate this one.
+	negone := -1.0
+
+	// amd64:"DIVSD"
+	z0 := zero / zero
+	// amd64:"MULSD"
+	z1 := zero * inf
+	// amd64:"SQRTSD"
+	z2 := math.Sqrt(negone)
+	return z0 + z1 + z2
+}
+
+func nanGenerate32() float32 {
+	zero := float32(0.0)
+	// amd64:-"DIVSS"
+	inf := 1 / zero // +inf. We can constant propagate this one.
+
+	// amd64:"DIVSS"
+	z0 := zero / zero
+	// amd64:"MULSS"
+	z1 := zero * inf
+	return z0 + z1
+}