Skip to content

Optimize/streamline fill operations #2395

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Apr 12, 2025
Merged

Optimize/streamline fill operations #2395

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3a056c7
Simplify fill.
ncruces Apr 11, 2025
2584cce
Faster fill.
ncruces Apr 11, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 11 additions & 6 deletions internal/engine/interpreter/interpreter.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1725,12 +1725,17 @@ func (ce *callEngine) callNativeFunc(ctx context.Context, m *wasm.ModuleInstance
if fillSize+offset > uint64(len(memoryInst.Buffer)) {
panic(wasmruntime.ErrRuntimeOutOfBoundsMemoryAccess)
} else if fillSize != 0 {
// Uses the copy trick for faster filling buffer.
// https://gist.github.com/taylorza/df2f89d5f9ab3ffd06865062a4cf015d
// Uses the copy trick for faster filling the buffer with the value.
// https://github.com/golang/go/blob/go1.24.0/src/bytes/bytes.go#L664-L673
buf := memoryInst.Buffer[offset : offset+fillSize]
buf[0] = value
for i := 1; i < len(buf); i *= 2 {
copy(buf[i:], buf[:i])
if value == 0 {
clear(buf)
} else {
buf[0] = value
for i := 1; i < len(buf); {
chunk := min(i, 8192)
i += copy(buf[i:], buf[:chunk])
}
}
}
frame.pc++
Expand Down Expand Up @@ -1804,7 +1809,7 @@ func (ce *callEngine) callNativeFunc(ctx context.Context, m *wasm.ModuleInstance
panic(wasmruntime.ErrRuntimeInvalidTableAccess)
} else if num > 0 {
// Uses the copy trick for faster filling the region with the value.
// https://gist.github.com/taylorza/df2f89d5f9ab3ffd06865062a4cf015d
// https://github.com/golang/go/blob/go1.24.0/src/slices/slices.go#L514-L517
targetRegion := table.References[offset : offset+num]
targetRegion[0] = ref
for i := 1; i < len(targetRegion); i *= 2 {
Expand Down
88 changes: 43 additions & 45 deletions internal/engine/wazevo/frontend/lower.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -665,19 +665,22 @@ func (c *Compiler) lowerCurrentOpcode() {
tableBaseAddr := c.loadTableBaseAddr(tableInstancePtr)
addr := builder.AllocateInstruction().AsIadd(tableBaseAddr, offsetInBytes).Insert(builder).Return()

// Prepare the loop and following block.
beforeLoop := builder.AllocateBasicBlock()
loopBlk := builder.AllocateBasicBlock()
loopVar := loopBlk.AddParam(builder, ssa.TypeI64)
followingBlk := builder.AllocateBasicBlock()

// Uses the copy trick for faster filling buffer like memory.fill, but in this case we copy 8 bytes at a time.
// Tables are rarely huge, so ignore the 8KB maximum.
// https://github.com/golang/go/blob/go1.24.0/src/slices/slices.go#L514-L517
//
// buf := memoryInst.Buffer[offset : offset+fillSize]
// buf[0:8] = value
// for i := 8; i < fillSize; i *= 2 { Begin with 8 bytes.
// copy(buf[i:], buf[:i])
// }

// Prepare the loop and following block.
beforeLoop := builder.AllocateBasicBlock()
loopBlk := builder.AllocateBasicBlock()
loopVar := loopBlk.AddParam(builder, ssa.TypeI64)
followingBlk := builder.AllocateBasicBlock()

// Insert the jump to the beforeLoop block; If the fillSize is zero, then jump to the following block to skip entire logics.
zero := builder.AllocateInstruction().AsIconst64(0).Insert(builder).Return()
ifFillSizeZero := builder.AllocateInstruction().AsIcmp(fillSizeExt, zero, ssa.IntegerCmpCondEqual).
Expand All @@ -688,32 +691,24 @@ func (c *Compiler) lowerCurrentOpcode() {
// buf[0:8] = value
builder.SetCurrentBlock(beforeLoop)
builder.AllocateInstruction().AsStore(ssa.OpcodeStore, value, addr, 0).Insert(builder)
initValue := builder.AllocateInstruction().AsIconst64(8).Insert(builder).Return()
c.insertJumpToBlock(c.allocateVarLengthValues(1, initValue), loopBlk)
eight := builder.AllocateInstruction().AsIconst64(8).Insert(builder).Return()
c.insertJumpToBlock(c.allocateVarLengthValues(1, eight), loopBlk)

builder.SetCurrentBlock(loopBlk)
dstAddr := builder.AllocateInstruction().AsIadd(addr, loopVar).Insert(builder).Return()

// If loopVar*2 > fillSizeInBytes, then count must be fillSizeInBytes-loopVar.
var count ssa.Value
{
loopVarDoubled := builder.AllocateInstruction().AsIadd(loopVar, loopVar).Insert(builder).Return()
loopVarDoubledLargerThanFillSize := builder.
AllocateInstruction().AsIcmp(loopVarDoubled, fillSizeInBytes, ssa.IntegerCmpCondUnsignedGreaterThanOrEqual).
Insert(builder).Return()
diff := builder.AllocateInstruction().AsIsub(fillSizeInBytes, loopVar).Insert(builder).Return()
count = builder.AllocateInstruction().AsSelect(loopVarDoubledLargerThanFillSize, diff, loopVar).Insert(builder).Return()
}
newLoopVar := builder.AllocateInstruction().AsIadd(loopVar, loopVar).Insert(builder).Return()
newLoopVarLessThanFillSize := builder.AllocateInstruction().
AsIcmp(newLoopVar, fillSizeInBytes, ssa.IntegerCmpCondUnsignedLessThan).Insert(builder).Return()

c.callMemmove(dstAddr, addr, count)
// On the last iteration, count must be fillSizeInBytes-loopVar.
diff := builder.AllocateInstruction().AsIsub(fillSizeInBytes, loopVar).Insert(builder).Return()
count := builder.AllocateInstruction().AsSelect(newLoopVarLessThanFillSize, loopVar, diff).Insert(builder).Return()

shiftAmount := builder.AllocateInstruction().AsIconst64(1).Insert(builder).Return()
newLoopVar := builder.AllocateInstruction().AsIshl(loopVar, shiftAmount).Insert(builder).Return()
loopVarLessThanFillSize := builder.AllocateInstruction().
AsIcmp(newLoopVar, fillSizeInBytes, ssa.IntegerCmpCondUnsignedLessThan).Insert(builder).Return()
c.callMemmove(dstAddr, addr, count)

builder.AllocateInstruction().
AsBrnz(loopVarLessThanFillSize, c.allocateVarLengthValues(1, newLoopVar), loopBlk).
AsBrnz(newLoopVarLessThanFillSize, c.allocateVarLengthValues(1, newLoopVar), loopBlk).
Insert(builder)

c.insertJumpToBlock(ssa.ValuesNil, followingBlk)
Expand Down Expand Up @@ -741,11 +736,15 @@ func (c *Compiler) lowerCurrentOpcode() {
// Calculate the base address:
addr := builder.AllocateInstruction().AsIadd(c.getMemoryBaseValue(false), offset).Insert(builder).Return()

// Uses the copy trick for faster filling buffer: https://gist.github.com/taylorza/df2f89d5f9ab3ffd06865062a4cf015d
// Uses the copy trick for faster filling buffer, with a maximum chunk size of 8KB.
// https://github.com/golang/go/blob/go1.24.0/src/bytes/bytes.go#L664-L673
//
// buf := memoryInst.Buffer[offset : offset+fillSize]
// buf[0] = value
// for i := 1; i < fillSize; i *= 2 {
// copy(buf[i:], buf[:i])
// for i := 1; i < fillSize; {
// chunk := ((i - 1) & 8191) + 1
// copy(buf[i:], buf[:chunk])
// i += chunk
// }

// Prepare the loop and following block.
Expand All @@ -764,32 +763,31 @@ func (c *Compiler) lowerCurrentOpcode() {
// buf[0] = value
builder.SetCurrentBlock(beforeLoop)
builder.AllocateInstruction().AsStore(ssa.OpcodeIstore8, value, addr, 0).Insert(builder)
initValue := builder.AllocateInstruction().AsIconst64(1).Insert(builder).Return()
c.insertJumpToBlock(c.allocateVarLengthValues(1, initValue), loopBlk)
one := builder.AllocateInstruction().AsIconst64(1).Insert(builder).Return()
c.insertJumpToBlock(c.allocateVarLengthValues(1, one), loopBlk)

builder.SetCurrentBlock(loopBlk)
dstAddr := builder.AllocateInstruction().AsIadd(addr, loopVar).Insert(builder).Return()

// If loopVar*2 > fillSizeExt, then count must be fillSizeExt-loopVar.
var count ssa.Value
{
loopVarDoubled := builder.AllocateInstruction().AsIadd(loopVar, loopVar).Insert(builder).Return()
loopVarDoubledLargerThanFillSize := builder.
AllocateInstruction().AsIcmp(loopVarDoubled, fillSize, ssa.IntegerCmpCondUnsignedGreaterThanOrEqual).
Insert(builder).Return()
diff := builder.AllocateInstruction().AsIsub(fillSize, loopVar).Insert(builder).Return()
count = builder.AllocateInstruction().AsSelect(loopVarDoubledLargerThanFillSize, diff, loopVar).Insert(builder).Return()
}

c.callMemmove(dstAddr, addr, count)
// chunk := ((i - 1) & 8191) + 1
mask := builder.AllocateInstruction().AsIconst64(16383).Insert(builder).Return()
tmp1 := builder.AllocateInstruction().AsIsub(loopVar, one).Insert(builder).Return()
tmp2 := builder.AllocateInstruction().AsBand(tmp1, mask).Insert(builder).Return()
chunk := builder.AllocateInstruction().AsIadd(tmp2, one).Insert(builder).Return()

shiftAmount := builder.AllocateInstruction().AsIconst64(1).Insert(builder).Return()
newLoopVar := builder.AllocateInstruction().AsIshl(loopVar, shiftAmount).Insert(builder).Return()
loopVarLessThanFillSize := builder.AllocateInstruction().
// i += chunk
newLoopVar := builder.AllocateInstruction().AsIadd(loopVar, chunk).Insert(builder).Return()
newLoopVarLessThanFillSize := builder.AllocateInstruction().
AsIcmp(newLoopVar, fillSize, ssa.IntegerCmpCondUnsignedLessThan).Insert(builder).Return()

// count = min(chunk, fillSize-loopVar)
diff := builder.AllocateInstruction().AsIsub(fillSize, loopVar).Insert(builder).Return()
count := builder.AllocateInstruction().AsSelect(newLoopVarLessThanFillSize, chunk, diff).Insert(builder).Return()

c.callMemmove(dstAddr, addr, count)

builder.AllocateInstruction().
AsBrnz(loopVarLessThanFillSize, c.allocateVarLengthValues(1, newLoopVar), loopBlk).
AsBrnz(newLoopVarLessThanFillSize, c.allocateVarLengthValues(1, newLoopVar), loopBlk).
Insert(builder)

c.insertJumpToBlock(ssa.ValuesNil, followingBlk)
Expand Down
6 changes: 5 additions & 1 deletion internal/wasm/table.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -326,10 +326,14 @@ func (t *TableInstance) Grow(delta uint32, initialRef Reference) (currentLen uin
newLen >= math.MaxUint32 || (t.Max != nil && newLen > int64(*t.Max)) {
return 0xffffffff // = -1 in signed 32-bit integer.
}

t.References = append(t.References, make([]uintptr, delta)...)
if initialRef == 0 {
return
}

// Uses the copy trick for faster filling the new region with the initial value.
// https://gist.github.com/taylorza/df2f89d5f9ab3ffd06865062a4cf015d
// https://github.com/golang/go/blob/go1.24.0/src/slices/slices.go#L514-L517
newRegion := t.References[currentLen:]
newRegion[0] = initialRef
for i := 1; i < len(newRegion); i *= 2 {
Expand Down
Loading