Skip to content

Commit

Permalink
draw: avoid FMA (Fused Multiply Add)
Browse files Browse the repository at this point in the history 
Changing "expr" to "float64(expr)" disables FMA on that expression, even
when expr's type is already nominally float64.

The output of this package should now be independent of GOARCH.

See the "FMA (Fused Multiply Add) detector" discussion at
https://groups.google.com/g/golang-dev/c/oZv0PaZmUGs

Fixes golang/go#67029

Change-Id: I43cecee0e9c8156eae39f52b5a2338c34d31a65a
Reviewed-on: https://go-review.googlesource.com/c/image/+/596816
Reviewed-by: Nigel Tao (INACTIVE; USE @golang.org INSTEAD) <[email protected]>
LUCI-TryBot-Result: Go LUCI <[email protected]>
Reviewed-by: Ian Lance Taylor <[email protected]>
  • Loading branch information
@nigeltao
nigeltao committed Jul 8, 2024
1 parent 3bbf4a6 commit 9abbe10
Show file tree
Hide file tree
Showing 3 changed files with 680 additions and 675 deletions.
77 changes: 41 additions & 36 deletions draw/gen.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -283,20 +283,20 @@ func expnDollar(prefix, dollar, suffix string, d *data) string {
switch d.sType {
default:
return argf(args, ""+
"$3r = $0*$1r + $2*$3r\n"+
"$3g = $0*$1g + $2*$3g\n"+
"$3b = $0*$1b + $2*$3b\n"+
"$3a = $0*$1a + $2*$3a",
"$3r = float64($0*$1r) + float64($2*$3r)\n"+
"$3g = float64($0*$1g) + float64($2*$3g)\n"+
"$3b = float64($0*$1b) + float64($2*$3b)\n"+
"$3a = float64($0*$1a) + float64($2*$3a)",
)
case "*image.Gray":
return argf(args, ""+
"$3r = $0*$1r + $2*$3r",
"$3r = float64($0*$1r) + float64($2*$3r)",
)
case "*image.YCbCr":
return argf(args, ""+
"$3r = $0*$1r + $2*$3r\n"+
"$3g = $0*$1g + $2*$3g\n"+
"$3b = $0*$1b + $2*$3b",
"$3r = float64($0*$1r) + float64($2*$3r)\n"+
"$3g = float64($0*$1g) + float64($2*$3g)\n"+
"$3b = float64($0*$1b) + float64($2*$3b)",
)
}

Expand Down Expand Up @@ -783,34 +783,39 @@ func expnDollar(prefix, dollar, suffix string, d *data) string {
}

if dollar == "srcf" {
avoidFMA0, avoidFMA1 := "", "" // FMA is Fused Multiply Add.
if extra != "" {
avoidFMA0, avoidFMA1 = "float64(", ")"
}

switch d.sType {
default:
fmt.Fprintf(buf, ""+
"%[1]sr %[2]s float64(%[1]sru)%[3]s\n"+
"%[1]sg %[2]s float64(%[1]sgu)%[3]s\n"+
"%[1]sb %[2]s float64(%[1]sbu)%[3]s\n"+
"%[1]sa %[2]s float64(%[1]sau)%[3]s\n",
lhs, eqOp, extra,
"%[1]sr %[2]s %[4]sfloat64(%[1]sru)%[3]s%[5]s\n"+
"%[1]sg %[2]s %[4]sfloat64(%[1]sgu)%[3]s%[5]s\n"+
"%[1]sb %[2]s %[4]sfloat64(%[1]sbu)%[3]s%[5]s\n"+
"%[1]sa %[2]s %[4]sfloat64(%[1]sau)%[3]s%[5]s\n",
lhs, eqOp, extra, avoidFMA0, avoidFMA1,
)
case "*image.Gray":
fmt.Fprintf(buf, ""+
"%[1]sr %[2]s float64(%[1]sru)%[3]s\n",
lhs, eqOp, extra,
"%[1]sr %[2]s %[4]sfloat64(%[1]sru)%[3]s%[5]s\n",
lhs, eqOp, extra, avoidFMA0, avoidFMA1,
)
case "*image.YCbCr":
fmt.Fprintf(buf, ""+
"%[1]sr %[2]s float64(%[1]sru)%[3]s\n"+
"%[1]sg %[2]s float64(%[1]sgu)%[3]s\n"+
"%[1]sb %[2]s float64(%[1]sbu)%[3]s\n",
lhs, eqOp, extra,
"%[1]sr %[2]s %[4]sfloat64(%[1]sru)%[3]s%[5]s\n"+
"%[1]sg %[2]s %[4]sfloat64(%[1]sgu)%[3]s%[5]s\n"+
"%[1]sb %[2]s %[4]sfloat64(%[1]sbu)%[3]s%[5]s\n",
lhs, eqOp, extra, avoidFMA0, avoidFMA1,
)
case "image.RGBA64Image":
fmt.Fprintf(buf, ""+
"%[1]sr %[2]s float64(%[1]su.R)%[3]s\n"+
"%[1]sg %[2]s float64(%[1]su.G)%[3]s\n"+
"%[1]sb %[2]s float64(%[1]su.B)%[3]s\n"+
"%[1]sa %[2]s float64(%[1]su.A)%[3]s\n",
lhs, eqOp, extra,
"%[1]sr %[2]s %[4]sfloat64(%[1]su.R)%[3]s%[5]s\n"+
"%[1]sg %[2]s %[4]sfloat64(%[1]su.G)%[3]s%[5]s\n"+
"%[1]sb %[2]s %[4]sfloat64(%[1]su.B)%[3]s%[5]s\n"+
"%[1]sa %[2]s %[4]sfloat64(%[1]su.A)%[3]s%[5]s\n",
lhs, eqOp, extra, avoidFMA0, avoidFMA1,
)
}
}
Expand Down Expand Up @@ -1175,8 +1180,8 @@ const (
$preInner
for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ { $tweakDx
dxf := float64(dr.Min.X + int(dx)) + 0.5
sx0 := int(d2s[0]*dxf + d2s[1]*dyf + d2s[2]) + bias.X
sy0 := int(d2s[3]*dxf + d2s[4]*dyf + d2s[5]) + bias.Y
sx0 := int(float64(d2s[0]*dxf) + float64(d2s[1]*dyf) + d2s[2]) + bias.X
sy0 := int(float64(d2s[3]*dxf) + float64(d2s[4]*dyf) + d2s[5]) + bias.Y
if !(image.Point{sx0, sy0}).In(sr) {
continue
}
Expand All @@ -1197,7 +1202,7 @@ const (
$preOuter
for dy := int32(adr.Min.Y); dy < int32(adr.Max.Y); dy++ {
sy := (float64(dy)+0.5)*yscale - 0.5
sy := float64((float64(dy)+0.5)*yscale) - 0.5
// If sy < 0, we will clamp sy0 to 0 anyway, so it doesn't matter if
// we say int32(sy) instead of int32(math.Floor(sy)). Similarly for
// sx, below.
Expand All @@ -1215,7 +1220,7 @@ const (
$preInner
for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ { $tweakDx
sx := (float64(dx)+0.5)*xscale - 0.5
sx := float64((float64(dx)+0.5)*xscale) - 0.5
sx0 := int32(sx)
xFrac0 := sx - float64(sx0)
xFrac1 := 1 - xFrac0
Expand Down Expand Up @@ -1250,8 +1255,8 @@ const (
$preInner
for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ { $tweakDx
dxf := float64(dr.Min.X + int(dx)) + 0.5
sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
sx := float64(d2s[0]*dxf) + float64(d2s[1]*dyf) + d2s[2]
sy := float64(d2s[3]*dxf) + float64(d2s[4]*dyf) + d2s[5]
if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
continue
}
Expand Down Expand Up @@ -1458,10 +1463,10 @@ const (
var pr, pg, pb, pa float64
for _, c := range z.vertical.contribs[s.i:s.j] {
p := &tmp[c.coord*z.dw+dx]
pr += p[0] * c.weight
pg += p[1] * c.weight
pb += p[2] * c.weight
pa += p[3] * c.weight
pr += float64(p[0] * c.weight)
pg += float64(p[1] * c.weight)
pb += float64(p[2] * c.weight)
pa += float64(p[3] * c.weight)
}
$clampToAlpha
$outputf[dr.Min.X + int(dx), dr.Min.Y + int(adr.Min.Y + dy), ftou, p, s.invTotalWeight]
Expand Down Expand Up @@ -1495,8 +1500,8 @@ const (
$preInner
for dx := int32(adr.Min.X); dx < int32(adr.Max.X); dx++ { $tweakDx
dxf := float64(dr.Min.X + int(dx)) + 0.5
sx := d2s[0]*dxf + d2s[1]*dyf + d2s[2]
sy := d2s[3]*dxf + d2s[4]*dyf + d2s[5]
sx := float64(d2s[0]*dxf) + float64(d2s[1]*dyf) + d2s[2]
sy := float64(d2s[3]*dxf) + float64(d2s[4]*dyf) + d2s[5]
if !(image.Point{int(sx) + bias.X, int(sy) + bias.Y}).In(sr) {
continue
}
Expand Down
Loading

0 comments on commit 9abbe10

Please sign in to comment.