Skip to content

Commit

Permalink
cmd/compile: wire up math/bits.Len intrinsics for loong64
Browse files Browse the repository at this point in the history 
For the SubFromLen64 test case to work, we need to fold c-(-(x-d))
into x+(c-d) as well.

benchmark TODO

Updates golang#59120

Change-Id: Icc8f7d8e79c6168aae634f5c36f044f3fd034d89
  • Loading branch information
@xen0n
xen0n committed Mar 27, 2023
1 parent 5873cbd commit 71e8866
Show file tree
Hide file tree
Showing 7 changed files with 114 additions and 5 deletions.
2 changes: 2 additions & 0 deletions src/cmd/compile/internal/loong64/ssa.go
View file
Expand Up @@ -351,6 +351,8 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
ssa.OpLOONG64NEGD,
ssa.OpLOONG64CTZW,
ssa.OpLOONG64CTZV,
ssa.OpLOONG64CLZW,
ssa.OpLOONG64CLZV,
ssa.OpLOONG64SQRTD,
ssa.OpLOONG64SQRTF:
p := s.Prog(v.Op.Asm())
Expand Down
3 changes: 3 additions & 0 deletions src/cmd/compile/internal/ssa/_gen/LOONG64.rules
View file
Expand Up @@ -128,6 +128,8 @@
(Neg(32|64)F ...) => (NEG(F|D) ...)
(Ctz(32|64) ...) => (CTZ(W|V) ...)
(Ctz(8|16|32|64)NonZero ...) => (Ctz(8|16|32|64) ...)
(BitLen64 <t> x) => (NEGV <t> (SUBVconst <t> [64] (CLZV <t> x)))
(BitLen32 <t> x) => (NEGV <t> (SUBVconst <t> [32] (CLZW <t> x)))

(Com(64|32|16|8) x) => (NOR (MOVVconst [0]) x)

Expand Down Expand Up @@ -634,6 +636,7 @@
(SUBVconst [c] (MOVVconst [d])) => (MOVVconst [d-c])
(SUBVconst [c] (SUBVconst [d] x)) && is32Bit(-c-d) => (ADDVconst [-c-d] x)
(SUBVconst [c] (ADDVconst [d] x)) && is32Bit(-c+d) => (ADDVconst [-c+d] x)
(SUBV (MOVVconst [c]) (NEGV (SUBVconst [d] x))) => (ADDVconst [c-d] x)
(SLLVconst [c] (MOVVconst [d])) => (MOVVconst [d<<uint64(c)])
(SRLVconst [c] (MOVVconst [d])) => (MOVVconst [int64(uint64(d)>>uint64(c))])
(SRAVconst [c] (MOVVconst [d])) => (MOVVconst [d>>uint64(c)])
Expand Down
2 changes: 2 additions & 0 deletions src/cmd/compile/internal/ssa/_gen/LOONG64Ops.go
View file
Expand Up @@ -194,6 +194,8 @@ func init() {
{name: "SQRTF", argLength: 1, reg: fp11, asm: "SQRTF"}, // sqrt(arg0), float32
{name: "CTZW", argLength: 1, reg: gp11, asm: "CTZW"}, // Count trailing (low order) zeroes (returns 0-32)
{name: "CTZV", argLength: 1, reg: gp11, asm: "CTZV"}, // Count trailing (low order) zeroes (returns 0-64)
{name: "CLZW", argLength: 1, reg: gp11, asm: "CLZW"}, // Count leading (high order) zeroes (returns 0-32)
{name: "CLZV", argLength: 1, reg: gp11, asm: "CLZV"}, // Count leading (high order) zeroes (returns 0-64)

{name: "MASKEQZ", argLength: 2, reg: gp21, asm: "MASKEQZ"}, // returns 0 if arg1 == 0, otherwise returns arg0
{name: "MASKNEZ", argLength: 2, reg: gp21, asm: "MASKNEZ"}, // returns 0 if arg1 != 0, otherwise returns arg0
Expand Down
28 changes: 28 additions & 0 deletions src/cmd/compile/internal/ssa/opGen.go
View file

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

63 changes: 63 additions & 0 deletions src/cmd/compile/internal/ssa/rewriteLOONG64.go
View file

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

10 changes: 5 additions & 5 deletions src/cmd/compile/internal/ssagen/ssa.go
View file
Expand Up @@ -4557,12 +4557,12 @@ func InitTables() {
func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
return s.newValue1(ssa.OpBitLen64, types.Types[types.TINT], args[0])
},
sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS, sys.PPC64, sys.Wasm)
sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.Loong64, sys.MIPS, sys.PPC64, sys.Wasm)
addF("math/bits", "Len32",
func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
return s.newValue1(ssa.OpBitLen32, types.Types[types.TINT], args[0])
},
sys.AMD64, sys.ARM64, sys.PPC64)
sys.AMD64, sys.ARM64, sys.Loong64, sys.PPC64)
addF("math/bits", "Len32",
func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
if s.config.PtrSize == 4 {
Expand All @@ -4581,7 +4581,7 @@ func InitTables() {
x := s.newValue1(ssa.OpZeroExt16to64, types.Types[types.TUINT64], args[0])
return s.newValue1(ssa.OpBitLen64, types.Types[types.TINT], x)
},
sys.ARM64, sys.ARM, sys.S390X, sys.MIPS, sys.PPC64, sys.Wasm)
sys.ARM64, sys.ARM, sys.S390X, sys.Loong64, sys.MIPS, sys.PPC64, sys.Wasm)
addF("math/bits", "Len16",
func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
return s.newValue1(ssa.OpBitLen16, types.Types[types.TINT], args[0])
Expand All @@ -4596,7 +4596,7 @@ func InitTables() {
x := s.newValue1(ssa.OpZeroExt8to64, types.Types[types.TUINT64], args[0])
return s.newValue1(ssa.OpBitLen64, types.Types[types.TINT], x)
},
sys.ARM64, sys.ARM, sys.S390X, sys.MIPS, sys.PPC64, sys.Wasm)
sys.ARM64, sys.ARM, sys.S390X, sys.Loong64, sys.MIPS, sys.PPC64, sys.Wasm)
addF("math/bits", "Len8",
func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
return s.newValue1(ssa.OpBitLen8, types.Types[types.TINT], args[0])
Expand All @@ -4609,7 +4609,7 @@ func InitTables() {
}
return s.newValue1(ssa.OpBitLen64, types.Types[types.TINT], args[0])
},
sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS, sys.PPC64, sys.Wasm)
sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.Loong64, sys.MIPS, sys.PPC64, sys.Wasm)
// LeadingZeros is handled because it trivially calls Len.
addF("math/bits", "Reverse64",
func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
Expand Down
11 changes: 11 additions & 0 deletions test/codegen/mathbits.go
View file
Expand Up @@ -17,6 +17,7 @@ func LeadingZeros(n uint) int {
// amd64/v3:"LZCNTQ", -"BSRQ"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// loong64:"CLZV"
// mips:"CLZ"
// wasm:"I64Clz"
// ppc64x:"CNTLZD"
Expand All @@ -28,6 +29,7 @@ func LeadingZeros64(n uint64) int {
// amd64/v3:"LZCNTQ", -"BSRQ"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// loong64:"CLZV"
// mips:"CLZ"
// wasm:"I64Clz"
// ppc64x:"CNTLZD"
Expand All @@ -39,6 +41,7 @@ func LeadingZeros32(n uint32) int {
// amd64/v3: "LZCNTL",- "BSRL"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZW"
// loong64:"CLZW"
// mips:"CLZ"
// wasm:"I64Clz"
// ppc64x:"CNTLZW"
Expand All @@ -50,6 +53,7 @@ func LeadingZeros16(n uint16) int {
// amd64/v3: "LZCNTL",- "BSRL"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// loong64:"CLZV"
// mips:"CLZ"
// wasm:"I64Clz"
// ppc64x:"CNTLZD"
Expand All @@ -61,6 +65,7 @@ func LeadingZeros8(n uint8) int {
// amd64/v3: "LZCNTL",- "BSRL"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// loong64:"CLZV"
// mips:"CLZ"
// wasm:"I64Clz"
// ppc64x:"CNTLZD"
Expand All @@ -76,6 +81,7 @@ func Len(n uint) int {
// amd64/v3: "LZCNTQ"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// loong64:"CLZV"
// mips:"CLZ"
// wasm:"I64Clz"
// ppc64x:"SUBC","CNTLZD"
Expand All @@ -87,13 +93,15 @@ func Len64(n uint64) int {
// amd64/v3: "LZCNTQ"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// loong64:"CLZV"
// mips:"CLZ"
// wasm:"I64Clz"
// ppc64x:"SUBC","CNTLZD"
return bits.Len64(n)
}

func SubFromLen64(n uint64) int {
// loong64:"CLZV",-"ADD"
// ppc64x:"CNTLZD",-"SUBC"
return 64 - bits.Len64(n)
}
Expand All @@ -103,6 +111,7 @@ func Len32(n uint32) int {
// amd64/v3: "LZCNTL"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// loong64:"CLZW"
// mips:"CLZ"
// wasm:"I64Clz"
// ppc64x: "CNTLZW"
Expand All @@ -114,6 +123,7 @@ func Len16(n uint16) int {
// amd64/v3: "LZCNTL"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// loong64:"CLZV"
// mips:"CLZ"
// wasm:"I64Clz"
// ppc64x:"SUBC","CNTLZD"
Expand All @@ -125,6 +135,7 @@ func Len8(n uint8) int {
// amd64/v3: "LZCNTL"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// loong64:"CLZV"
// mips:"CLZ"
// wasm:"I64Clz"
// ppc64x:"SUBC","CNTLZD"
Expand Down

0 comments on commit 71e8866

Please sign in to comment.