Stowage/go
2
0
Fork 0
mirror of https://github.com/golang/go.git synced 2025-12-08 06:10:04 +00:00
Code Issues Projects Releases Packages Wiki Activity

[dev.simd] cmd/compile: fix SIMD const rematerialization condition

Browse source 
This CL fixes a condition for the previous fix CL 704056.

Change-Id: I1f1f8c6f72870403cb3dff14755c43385dc0c933
Reviewed-on: https://go-review.googlesource.com/c/go/+/705499
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
This commit is contained in:
Junyang Shao 2025-09-19 18:38:25 +00:00
parent 2ca96d218d
commit 63a09d6d3d
2 changed files with 53 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

17
src/cmd/compile/internal/ssa/regalloc.go
View file

@ -2576,22 +2576,25 @@ func (e *edgeState) processDest(loc Location, vid ID, splice **Value, pos src.XP
e.s.f.Fatalf("can't find source for %s->%s: %s\n", e.p, e.b, v.LongString())
}
if dstReg {
// Handle incompatible registers.
// We want to rematerialize v into a register that is incompatible with v's op's register mask.
// Instead of setting the wrong register for the rematerialized v, we should find the right register
// for it and emit an additional copy to move to the desired register.
// For #70451.
if e.s.regspec(v).outputs[0].regs&regMask(1<<register(loc.(*Register).num)) == 0 && c != nil {
if e.s.regspec(v).outputs[0].regs&regMask(1<<register(loc.(*Register).num)) == 0 {
_, srcReg := src.(*Register)
if !srcReg {
if srcReg {
// It exists in a valid register already, so just copy it to the desired register
// If src is a Register, c must have already been set.
x = e.p.NewValue1(pos, OpCopy, c.Type, c)
} else {
// We need a tmp register
x = v.copyInto(e.p)
r := e.findRegFor(x.Type)
e.erase(r)
// Rematerialize to a tmp register
// Rematerialize to the tmp register
e.set(r, vid, x, false, pos)
// Copy from tmp to the desired register
x = e.p.NewValue1(pos, OpCopy, x.Type, x)
} else {
// It exist in a valid register already, so just copy it to the desired register
x = e.p.NewValue1(pos, OpCopy, c.Type, c)
}
} else {
x = v.copyInto(e.p)

43
test/simd/bug2.go
View file

@ -24,3 +24,46 @@ func PackComplex(b bool) {
}
}
}
func PackComplex2(x0 uint16, src [][4]float32, b, b2 bool) {
var out [][4]byte
if b2 {
for y := range x0 {
row := out[:x0]
for x := range row {
px := &src[y]
if b {
var indices [4]uint32
fu := simd.LoadFloat32x4(px).AsUint32x4()
fu.ShiftAllRight(0).Store(nil)
entry := simd.LoadUint32x4(&[4]uint32{
toSrgbTable[indices[0]],
})
var res [4]uint32
entry.ShiftAllRight(19).Store(nil)
row[x] = [4]uint8{
uint8(res[0]),
uint8(res[1]),
uint8(res[2]),
}
} else {
row[x] = [4]uint8{
float32ToSrgb8(0),
float32ToSrgb8(1),
float32ToSrgb8(2),
}
}
}
out = out[len(out):]
}
}
}
var toSrgbTable = [4]uint32{}
func float32ToSrgb8(f float32) uint8 {
f = min(0, f)
fu := uint32(f)
entry := toSrgbTable[fu]
return uint8(entry * fu)
}