cmd/compile: optimize single-precision floating point square root

Add generic rule to rewrite the single-precision square root expression
with one single-precision instruction. The optimization will reduce two
times of precision converting between double-precision and single-precision.

On arm64 flatform.

previous:
  FCVTSD F0, F0
  FSQRTD F0, F0
  FCVTDS F0, F0

optimized:
  FSQRTS S0, S0

And this patch adds the test case to check the correctness.

This patch refers to CL 241877, contributed by Alice Xu
(dianhong.xu@arm.com)

Change-Id: I6de5d02281c693017ac4bd4c10963dd55989bd7e
Reviewed-on: https://go-review.googlesource.com/c/go/+/276873
Trust: fannie zhang <Fannie.Zhang@arm.com>
Run-TryBot: fannie zhang <Fannie.Zhang@arm.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

41 files changed, 255 insertions, 28 deletions

diff --git a/src/cmd/compile/internal/amd64/ssa.go b/src/cmd/compile/internal/amd64/ssa.go
index 32cb0a9368..d83d78f080 100644
--- a/src/cmd/compile/internal/amd64/ssa.go
+++ b/src/cmd/compile/internal/amd64/ssa.go
@@ -1053,7 +1053,7 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = v.Reg0()
 
-	case ssa.OpAMD64BSFQ, ssa.OpAMD64BSRQ, ssa.OpAMD64BSFL, ssa.OpAMD64BSRL, ssa.OpAMD64SQRTSD:
+	case ssa.OpAMD64BSFQ, ssa.OpAMD64BSRQ, ssa.OpAMD64BSFL, ssa.OpAMD64BSRL, ssa.OpAMD64SQRTSD, ssa.OpAMD64SQRTSS:
 		p := s.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_REG
 		p.From.Reg = v.Args[0].Reg()
@@ -1061,7 +1061,7 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		switch v.Op {
 		case ssa.OpAMD64BSFQ, ssa.OpAMD64BSRQ:
 			p.To.Reg = v.Reg0()
-		case ssa.OpAMD64BSFL, ssa.OpAMD64BSRL, ssa.OpAMD64SQRTSD:
+		case ssa.OpAMD64BSFL, ssa.OpAMD64BSRL, ssa.OpAMD64SQRTSD, ssa.OpAMD64SQRTSS:
 			p.To.Reg = v.Reg()
 		}
 	case ssa.OpAMD64ROUNDSD:
diff --git a/src/cmd/compile/internal/arm/ssa.go b/src/cmd/compile/internal/arm/ssa.go
index c4d8cbf149..7b2fec3765 100644
--- a/src/cmd/compile/internal/arm/ssa.go
+++ b/src/cmd/compile/internal/arm/ssa.go
@@ -654,6 +654,7 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		ssa.OpARMREV,
 		ssa.OpARMREV16,
 		ssa.OpARMRBIT,
+		ssa.OpARMSQRTF,
 		ssa.OpARMSQRTD,
 		ssa.OpARMNEGF,
 		ssa.OpARMNEGD,
diff --git a/src/cmd/compile/internal/arm64/ssa.go b/src/cmd/compile/internal/arm64/ssa.go
index 5067d92dfe..056a6eb62d 100644
--- a/src/cmd/compile/internal/arm64/ssa.go
+++ b/src/cmd/compile/internal/arm64/ssa.go
@@ -893,6 +893,7 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		ssa.OpARM64FMOVSgpfp,
 		ssa.OpARM64FNEGS,
 		ssa.OpARM64FNEGD,
+		ssa.OpARM64FSQRTS,
 		ssa.OpARM64FSQRTD,
 		ssa.OpARM64FCVTZSSW,
 		ssa.OpARM64FCVTZSDW,
diff --git a/src/cmd/compile/internal/mips/ssa.go b/src/cmd/compile/internal/mips/ssa.go
index 115e3cb8e2..13736d12b4 100644
--- a/src/cmd/compile/internal/mips/ssa.go
+++ b/src/cmd/compile/internal/mips/ssa.go
@@ -363,6 +363,7 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		ssa.OpMIPSMOVDF,
 		ssa.OpMIPSNEGF,
 		ssa.OpMIPSNEGD,
+		ssa.OpMIPSSQRTF,
 		ssa.OpMIPSSQRTD,
 		ssa.OpMIPSCLZ:
 		p := s.Prog(v.Op.Asm())
diff --git a/src/cmd/compile/internal/mips64/ssa.go b/src/cmd/compile/internal/mips64/ssa.go
index d9c47751e1..c5a3ca305a 100644
--- a/src/cmd/compile/internal/mips64/ssa.go
+++ b/src/cmd/compile/internal/mips64/ssa.go
@@ -355,6 +355,7 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		ssa.OpMIPS64MOVDF,
 		ssa.OpMIPS64NEGF,
 		ssa.OpMIPS64NEGD,
+		ssa.OpMIPS64SQRTF,
 		ssa.OpMIPS64SQRTD:
 		p := s.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_REG
diff --git a/src/cmd/compile/internal/s390x/ssa.go b/src/cmd/compile/internal/s390x/ssa.go
index 4830d902c2..ca6720bb33 100644
--- a/src/cmd/compile/internal/s390x/ssa.go
+++ b/src/cmd/compile/internal/s390x/ssa.go
@@ -586,7 +586,7 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		p.Reg = v.Args[1].Reg()
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = v.Reg()
-	case ssa.OpS390XFSQRT:
+	case ssa.OpS390XFSQRTS, ssa.OpS390XFSQRT:
 		p := s.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_REG
 		p.From.Reg = v.Args[0].Reg()
diff --git a/src/cmd/compile/internal/ssa/gen/386.rules b/src/cmd/compile/internal/ssa/gen/386.rules
index df03cb71a6..d6d122dc78 100644
--- a/src/cmd/compile/internal/ssa/gen/386.rules
+++ b/src/cmd/compile/internal/ssa/gen/386.rules
@@ -54,6 +54,7 @@
 (Bswap32 ...) => (BSWAPL ...)
 
 (Sqrt ...) => (SQRTSD ...)
+(Sqrt32 ...) => (SQRTSS ...)
 
 (Ctz16 x) => (BSFL (ORLconst <typ.UInt32> [0x10000] x))
 (Ctz16NonZero ...) => (BSFL ...)
diff --git a/src/cmd/compile/internal/ssa/gen/386Ops.go b/src/cmd/compile/internal/ssa/gen/386Ops.go
index 2b7185e537..c4b49fbb23 100644
--- a/src/cmd/compile/internal/ssa/gen/386Ops.go
+++ b/src/cmd/compile/internal/ssa/gen/386Ops.go
@@ -308,6 +308,7 @@ func init() {
 		{name: "BSWAPL", argLength: 1, reg: gp11, asm: "BSWAPL", resultInArg0: true, clobberFlags: true}, // arg0 swap bytes
 
 		{name: "SQRTSD", argLength: 1, reg: fp11, asm: "SQRTSD"}, // sqrt(arg0)
+		{name: "SQRTSS", argLength: 1, reg: fp11, asm: "SQRTSS"}, // sqrt(arg0), float32
 
 		{name: "SBBLcarrymask", argLength: 1, reg: flagsgp, asm: "SBBL"}, // (int32)(-1) if carry is set, 0 if carry is clear.
 		// Note: SBBW and SBBB are subsumed by SBBL
diff --git a/src/cmd/compile/internal/ssa/gen/AMD64.rules b/src/cmd/compile/internal/ssa/gen/AMD64.rules
index f2bcbd2dfc..bab9cee88c 100644
--- a/src/cmd/compile/internal/ssa/gen/AMD64.rules
+++ b/src/cmd/compile/internal/ssa/gen/AMD64.rules
@@ -104,6 +104,7 @@
 (PopCount8 x) => (POPCNTL (MOVBQZX <typ.UInt32> x))
 
 (Sqrt ...) => (SQRTSD ...)
+(Sqrt32 ...) => (SQRTSS ...)
 
 (RoundToEven x) => (ROUNDSD [0] x)
 (Floor x)       => (ROUNDSD [1] x)
diff --git a/src/cmd/compile/internal/ssa/gen/AMD64Ops.go b/src/cmd/compile/internal/ssa/gen/AMD64Ops.go
index 96475672a8..fd2c2023e6 100644
--- a/src/cmd/compile/internal/ssa/gen/AMD64Ops.go
+++ b/src/cmd/compile/internal/ssa/gen/AMD64Ops.go
@@ -594,6 +594,7 @@ func init() {
 		{name: "POPCNTL", argLength: 1, reg: gp11, asm: "POPCNTL", clobberFlags: true}, // count number of set bits in arg0
 
 		{name: "SQRTSD", argLength: 1, reg: fp11, asm: "SQRTSD"}, // sqrt(arg0)
+		{name: "SQRTSS", argLength: 1, reg: fp11, asm: "SQRTSS"}, // sqrt(arg0), float32
 
 		// ROUNDSD instruction isn't guaranteed to be on the target platform (it is SSE4.1)
 		// Any use must be preceded by a successful check of runtime.x86HasSSE41.
diff --git a/src/cmd/compile/internal/ssa/gen/ARM.rules b/src/cmd/compile/internal/ssa/gen/ARM.rules
index cbafd12a4f..f46f4238f7 100644
--- a/src/cmd/compile/internal/ssa/gen/ARM.rules
+++ b/src/cmd/compile/internal/ssa/gen/ARM.rules
@@ -56,6 +56,7 @@
 (Com(32|16|8) ...) => (MVN ...)
 
 (Sqrt ...) => (SQRTD ...)
+(Sqrt32 ...) => (SQRTF ...)
 (Abs ...) => (ABSD ...)
 
 // TODO: optimize this for ARMv5 and ARMv6
diff --git a/src/cmd/compile/internal/ssa/gen/ARM64.rules b/src/cmd/compile/internal/ssa/gen/ARM64.rules
index 98503748db..ea912f9f97 100644
--- a/src/cmd/compile/internal/ssa/gen/ARM64.rules
+++ b/src/cmd/compile/internal/ssa/gen/ARM64.rules
@@ -60,6 +60,8 @@
 (Trunc ...) => (FRINTZD ...)
 (FMA x y z) => (FMADDD z x y)
 
+(Sqrt32 ...) => (FSQRTS ...)
+
 // lowering rotates
 (RotateLeft8 <t> x (MOVDconst [c])) => (Or8 (Lsh8x64 <t> x (MOVDconst [c&7])) (Rsh8Ux64 <t> x (MOVDconst [-c&7])))
 (RotateLeft16 <t> x (MOVDconst [c])) => (Or16 (Lsh16x64 <t> x (MOVDconst [c&15])) (Rsh16Ux64 <t> x (MOVDconst [-c&15])))
diff --git a/src/cmd/compile/internal/ssa/gen/ARM64Ops.go b/src/cmd/compile/internal/ssa/gen/ARM64Ops.go
index e826e75252..0a4fd14b2b 100644
--- a/src/cmd/compile/internal/ssa/gen/ARM64Ops.go
+++ b/src/cmd/compile/internal/ssa/gen/ARM64Ops.go
@@ -236,6 +236,7 @@ func init() {
 		{name: "FNEGS", argLength: 1, reg: fp11, asm: "FNEGS"},                                // -arg0, float32
 		{name: "FNEGD", argLength: 1, reg: fp11, asm: "FNEGD"},                                // -arg0, float64
 		{name: "FSQRTD", argLength: 1, reg: fp11, asm: "FSQRTD"},                              // sqrt(arg0), float64
+		{name: "FSQRTS", argLength: 1, reg: fp11, asm: "FSQRTS"},                              // sqrt(arg0), float32
 		{name: "REV", argLength: 1, reg: gp11, asm: "REV"},                                    // byte reverse, 64-bit
 		{name: "REVW", argLength: 1, reg: gp11, asm: "REVW"},                                  // byte reverse, 32-bit
 		{name: "REV16W", argLength: 1, reg: gp11, asm: "REV16W"},                              // byte reverse in each 16-bit halfword, 32-bit
diff --git a/src/cmd/compile/internal/ssa/gen/ARMOps.go b/src/cmd/compile/internal/ssa/gen/ARMOps.go
index 70c789937a..253ff573ec 100644
--- a/src/cmd/compile/internal/ssa/gen/ARMOps.go
+++ b/src/cmd/compile/internal/ssa/gen/ARMOps.go
@@ -217,6 +217,7 @@ func init() {
 		{name: "NEGF", argLength: 1, reg: fp11, asm: "NEGF"},   // -arg0, float32
 		{name: "NEGD", argLength: 1, reg: fp11, asm: "NEGD"},   // -arg0, float64
 		{name: "SQRTD", argLength: 1, reg: fp11, asm: "SQRTD"}, // sqrt(arg0), float64
+		{name: "SQRTF", argLength: 1, reg: fp11, asm: "SQRTF"}, // sqrt(arg0), float32
 		{name: "ABSD", argLength: 1, reg: fp11, asm: "ABSD"},   // abs(arg0), float64
 
 		{name: "CLZ", argLength: 1, reg: gp11, asm: "CLZ"},     // count leading zero
diff --git a/src/cmd/compile/internal/ssa/gen/MIPS.rules b/src/cmd/compile/internal/ssa/gen/MIPS.rules
index bc1ce82940..6b59555cbe 100644
--- a/src/cmd/compile/internal/ssa/gen/MIPS.rules
+++ b/src/cmd/compile/internal/ssa/gen/MIPS.rules
@@ -121,6 +121,7 @@
 (Com(32|16|8) x) => (NORconst [0] x)
 
 (Sqrt ...) => (SQRTD ...)
+(Sqrt32 ...) => (SQRTF ...)
 
 // TODO: optimize this case?
 (Ctz32NonZero ...) => (Ctz32 ...)
diff --git a/src/cmd/compile/internal/ssa/gen/MIPS64.rules b/src/cmd/compile/internal/ssa/gen/MIPS64.rules
index e3f7633274..bc51a0d53d 100644
--- a/src/cmd/compile/internal/ssa/gen/MIPS64.rules
+++ b/src/cmd/compile/internal/ssa/gen/MIPS64.rules
@@ -121,6 +121,7 @@
 (Com(64|32|16|8) x) => (NOR (MOVVconst [0]) x)
 
 (Sqrt ...) => (SQRTD ...)
+(Sqrt32 ...) => (SQRTF ...)
 
 // boolean ops -- booleans are represented with 0=false, 1=true
 (AndB ...) => (AND ...)
diff --git a/src/cmd/compile/internal/ssa/gen/MIPS64Ops.go b/src/cmd/compile/internal/ssa/gen/MIPS64Ops.go
index e1e3933502..77f251c0d3 100644
--- a/src/cmd/compile/internal/ssa/gen/MIPS64Ops.go
+++ b/src/cmd/compile/internal/ssa/gen/MIPS64Ops.go
@@ -199,6 +199,7 @@ func init() {
 		{name: "NEGF", argLength: 1, reg: fp11, asm: "NEGF"},   // -arg0, float32
 		{name: "NEGD", argLength: 1, reg: fp11, asm: "NEGD"},   // -arg0, float64
 		{name: "SQRTD", argLength: 1, reg: fp11, asm: "SQRTD"}, // sqrt(arg0), float64
+		{name: "SQRTF", argLength: 1, reg: fp11, asm: "SQRTF"}, // sqrt(arg0), float32
 
 		// shifts
 		{name: "SLLV", argLength: 2, reg: gp21, asm: "SLLV"},                    // arg0 << arg1, shift amount is mod 64
diff --git a/src/cmd/compile/internal/ssa/gen/MIPSOps.go b/src/cmd/compile/internal/ssa/gen/MIPSOps.go
index 75ab99ea26..b92e8cb9f1 100644
--- a/src/cmd/compile/internal/ssa/gen/MIPSOps.go
+++ b/src/cmd/compile/internal/ssa/gen/MIPSOps.go
@@ -182,6 +182,7 @@ func init() {
 		{name: "NEGF", argLength: 1, reg: fp11, asm: "NEGF"},   // -arg0, float32
 		{name: "NEGD", argLength: 1, reg: fp11, asm: "NEGD"},   // -arg0, float64
 		{name: "SQRTD", argLength: 1, reg: fp11, asm: "SQRTD"}, // sqrt(arg0), float64
+		{name: "SQRTF", argLength: 1, reg: fp11, asm: "SQRTF"}, // sqrt(arg0), float32
 
 		// shifts
 		{name: "SLL", argLength: 2, reg: gp21, asm: "SLL"},                    // arg0 << arg1, shift amount is mod 32
diff --git a/src/cmd/compile/internal/ssa/gen/PPC64.rules b/src/cmd/compile/internal/ssa/gen/PPC64.rules
index a762be65d4..85ce9a5b54 100644
--- a/src/cmd/compile/internal/ssa/gen/PPC64.rules
+++ b/src/cmd/compile/internal/ssa/gen/PPC64.rules
@@ -71,6 +71,7 @@
 (Round(32|64)F ...) => (LoweredRound(32|64)F ...)
 
 (Sqrt ...) => (FSQRT ...)
+(Sqrt32 ...) => (FSQRTS ...)
 (Floor ...) => (FFLOOR ...)
 (Ceil ...) => (FCEIL ...)
 (Trunc ...) => (FTRUNC ...)
diff --git a/src/cmd/compile/internal/ssa/gen/RISCV64.rules b/src/cmd/compile/internal/ssa/gen/RISCV64.rules
index 9119ebc0e8..a11d1e6624 100644
--- a/src/cmd/compile/internal/ssa/gen/RISCV64.rules
+++ b/src/cmd/compile/internal/ssa/gen/RISCV64.rules
@@ -92,6 +92,7 @@
 (Com8  ...) => (NOT ...)
 
 (Sqrt ...) => (FSQRTD ...)
+(Sqrt32 ...) => (FSQRTS ...)
 
 // Sign and zero extension.
 
diff --git a/src/cmd/compile/internal/ssa/gen/S390X.rules b/src/cmd/compile/internal/ssa/gen/S390X.rules
index 7111d5e11a..e4a1cd6981 100644
--- a/src/cmd/compile/internal/ssa/gen/S390X.rules
+++ b/src/cmd/compile/internal/ssa/gen/S390X.rules
@@ -142,6 +142,8 @@
 (Round       x) => (FIDBR [1] x)
 (FMA     x y z) => (FMADD z x y)
 
+(Sqrt32    ...) => (FSQRTS ...)
+
 // Atomic loads and stores.
 // The SYNC instruction (fast-BCR-serialization) prevents store-load
 // reordering. Other sequences of memory operations (load-load,
diff --git a/src/cmd/compile/internal/ssa/gen/S390XOps.go b/src/cmd/compile/internal/ssa/gen/S390XOps.go
index b24fd61942..1ddad1febd 100644
--- a/src/cmd/compile/internal/ssa/gen/S390XOps.go
+++ b/src/cmd/compile/internal/ssa/gen/S390XOps.go
@@ -382,6 +382,7 @@ func init() {
 		{name: "NOTW", argLength: 1, reg: gp11, resultInArg0: true, clobberFlags: true}, // ^arg0
 
 		{name: "FSQRT", argLength: 1, reg: fp11, asm: "FSQRT"}, // sqrt(arg0)
+		{name: "FSQRTS", argLength: 1, reg: fp11, asm: "FSQRTS"}, // sqrt(arg0), float32
 
 		// Conditional register-register moves.
 		// The aux for these values is an s390x.CCMask value representing the condition code mask.
diff --git a/src/cmd/compile/internal/ssa/gen/Wasm.rules b/src/cmd/compile/internal/ssa/gen/Wasm.rules
index fc45cd3ed5..7cda16b4b5 100644
--- a/src/cmd/compile/internal/ssa/gen/Wasm.rules
+++ b/src/cmd/compile/internal/ssa/gen/Wasm.rules
@@ -332,6 +332,8 @@
 (Abs ...) => (F64Abs ...)
 (Copysign ...) => (F64Copysign ...)
 
+(Sqrt32 ...) => (F32Sqrt ...)
+
 (Ctz64 ...) => (I64Ctz ...)
 (Ctz32 x) => (I64Ctz (I64Or x (I64Const [0x100000000])))
 (Ctz16 x) => (I64Ctz (I64Or x (I64Const [0x10000])))
diff --git a/src/cmd/compile/internal/ssa/gen/WasmOps.go b/src/cmd/compile/internal/ssa/gen/WasmOps.go
index 36c53bc78c..c92878ca73 100644
--- a/src/cmd/compile/internal/ssa/gen/WasmOps.go
+++ b/src/cmd/compile/internal/ssa/gen/WasmOps.go
@@ -238,13 +238,13 @@ func init() {
 		{name: "I64Extend16S", asm: "I64Extend16S", argLength: 1, reg: gp11, typ: "Int64"}, // sign-extend arg0 from 16 to 64 bit
 		{name: "I64Extend32S", asm: "I64Extend32S", argLength: 1, reg: gp11, typ: "Int64"}, // sign-extend arg0 from 32 to 64 bit
 
-		{name: "F32Sqrt", asm: "F32Sqrt", argLength: 1, reg: fp64_11, typ: "Float32"},         // sqrt(arg0)
-		{name: "F32Trunc", asm: "F32Trunc", argLength: 1, reg: fp64_11, typ: "Float32"},       // trunc(arg0)
-		{name: "F32Ceil", asm: "F32Ceil", argLength: 1, reg: fp64_11, typ: "Float32"},         // ceil(arg0)
-		{name: "F32Floor", asm: "F32Floor", argLength: 1, reg: fp64_11, typ: "Float32"},       // floor(arg0)
-		{name: "F32Nearest", asm: "F32Nearest", argLength: 1, reg: fp64_11, typ: "Float32"},   // round(arg0)
-		{name: "F32Abs", asm: "F32Abs", argLength: 1, reg: fp64_11, typ: "Float32"},           // abs(arg0)
-		{name: "F32Copysign", asm: "F32Copysign", argLength: 2, reg: fp64_21, typ: "Float32"}, // copysign(arg0, arg1)
+		{name: "F32Sqrt", asm: "F32Sqrt", argLength: 1, reg: fp32_11, typ: "Float32"},         // sqrt(arg0)
+		{name: "F32Trunc", asm: "F32Trunc", argLength: 1, reg: fp32_11, typ: "Float32"},       // trunc(arg0)
+		{name: "F32Ceil", asm: "F32Ceil", argLength: 1, reg: fp32_11, typ: "Float32"},         // ceil(arg0)
+		{name: "F32Floor", asm: "F32Floor", argLength: 1, reg: fp32_11, typ: "Float32"},       // floor(arg0)
+		{name: "F32Nearest", asm: "F32Nearest", argLength: 1, reg: fp32_11, typ: "Float32"},   // round(arg0)
+		{name: "F32Abs", asm: "F32Abs", argLength: 1, reg: fp32_11, typ: "Float32"},           // abs(arg0)
+		{name: "F32Copysign", asm: "F32Copysign", argLength: 2, reg: fp32_21, typ: "Float32"}, // copysign(arg0, arg1)
 
 		{name: "F64Sqrt", asm: "F64Sqrt", argLength: 1, reg: fp64_11, typ: "Float64"},         // sqrt(arg0)
 		{name: "F64Trunc", asm: "F64Trunc", argLength: 1, reg: fp64_11, typ: "Float64"},       // trunc(arg0)
diff --git a/src/cmd/compile/internal/ssa/gen/generic.rules b/src/cmd/compile/internal/ssa/gen/generic.rules
index fab45243ed..9dd20a7cfa 100644
--- a/src/cmd/compile/internal/ssa/gen/generic.rules
+++ b/src/cmd/compile/internal/ssa/gen/generic.rules
@@ -1968,6 +1968,9 @@
 (Div32F x (Const32F <t> [c])) && reciprocalExact32(c) => (Mul32F x (Const32F <t> [1/c]))
 (Div64F x (Const64F <t> [c])) && reciprocalExact64(c) => (Mul64F x (Const64F <t> [1/c]))
 
+// rewrite single-precision sqrt expression "float32(math.Sqrt(float64(x)))"
+(Cvt64Fto32F sqrt0:(Sqrt (Cvt32Fto64F x))) && sqrt0.Uses==1 => (Sqrt32 x)
+
 (Sqrt (Const64F [c])) && !math.IsNaN(math.Sqrt(c)) => (Const64F [math.Sqrt(c)])
 
 // for rewriting results of some late-expanded rewrites (below)
diff --git a/src/cmd/compile/internal/ssa/gen/genericOps.go b/src/cmd/compile/internal/ssa/gen/genericOps.go
index 043f445c16..b730c436cf 100644
--- a/src/cmd/compile/internal/ssa/gen/genericOps.go
+++ b/src/cmd/compile/internal/ssa/gen/genericOps.go
@@ -258,13 +258,14 @@ var genericOps = []opData{
 	{name: "RotateLeft32", argLength: 2}, // Rotate bits in arg[0] left by arg[1]
 	{name: "RotateLeft64", argLength: 2}, // Rotate bits in arg[0] left by arg[1]
 
-	// Square root, float64 only.
+	// Square root.
 	// Special cases:
 	//   +∞  → +∞
 	//   ±0  → ±0 (sign preserved)
 	//   x<0 → NaN
 	//   NaN → NaN
-	{name: "Sqrt", argLength: 1}, // √arg0
+	{name: "Sqrt", argLength: 1}, // √arg0   (floating point, double precision)
+	{name: "Sqrt32", argLength: 1}, // √arg0 (floating point, single precision)
 
 	// Round to integer, float64 only.
 	// Special cases:
diff --git a/src/cmd/compile/internal/ssa/opGen.go b/src/cmd/compile/internal/ssa/opGen.go
index 2d37ae4357..bd9741fe3e 100644
--- a/src/cmd/compile/internal/ssa/opGen.go
+++ b/src/cmd/compile/internal/ssa/opGen.go
@@ -432,6 +432,7 @@ const (
 	Op386BSRW
 	Op386BSWAPL
 	Op386SQRTSD
+	Op386SQRTSS
 	Op386SBBLcarrymask
 	Op386SETEQ
 	Op386SETNE
@@ -888,6 +889,7 @@ const (
 	OpAMD64POPCNTQ
 	OpAMD64POPCNTL
 	OpAMD64SQRTSD
+	OpAMD64SQRTSS
 	OpAMD64ROUNDSD
 	OpAMD64VFMADD231SD
 	OpAMD64SBBQcarrymask
@@ -1090,6 +1092,7 @@ const (
 	OpARMNEGF
 	OpARMNEGD
 	OpARMSQRTD
+	OpARMSQRTF
 	OpARMABSD
 	OpARMCLZ
 	OpARMREV
@@ -1358,6 +1361,7 @@ const (
 	OpARM64FNEGS
 	OpARM64FNEGD
 	OpARM64FSQRTD
+	OpARM64FSQRTS
 	OpARM64REV
 	OpARM64REVW
 	OpARM64REV16W
@@ -1641,6 +1645,7 @@ const (
 	OpMIPSNEGF
 	OpMIPSNEGD
 	OpMIPSSQRTD
+	OpMIPSSQRTF
 	OpMIPSSLL
 	OpMIPSSLLconst
 	OpMIPSSRL
@@ -1751,6 +1756,7 @@ const (
 	OpMIPS64NEGF
 	OpMIPS64NEGD
 	OpMIPS64SQRTD
+	OpMIPS64SQRTF
 	OpMIPS64SLLV
 	OpMIPS64SLLVconst
 	OpMIPS64SRLV
@@ -2301,6 +2307,7 @@ const (
 	OpS390XNOT
 	OpS390XNOTW
 	OpS390XFSQRT
+	OpS390XFSQRTS
 	OpS390XLOCGR
 	OpS390XMOVBreg
 	OpS390XMOVBZreg
@@ -2727,6 +2734,7 @@ const (
 	OpRotateLeft32
 	OpRotateLeft64
 	OpSqrt
+	OpSqrt32
 	OpFloor
 	OpCeil
 	OpTrunc
@@ -4779,6 +4787,19 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
+		name:   "SQRTSS",
+		argLen: 1,
+		asm:    x86.ASQRTSS,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 65280}, // X0 X1 X2 X3 X4 X5 X6 X7
+			},
+			outputs: []outputInfo{
+				{0, 65280}, // X0 X1 X2 X3 X4 X5 X6 X7
+			},
+		},
+	},
+	{
 		name:   "SBBLcarrymask",
 		argLen: 1,
 		asm:    x86.ASBBL,
@@ -11631,6 +11652,19 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
+		name:   "SQRTSS",
+		argLen: 1,
+		asm:    x86.ASQRTSS,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14
+			},
+			outputs: []outputInfo{
+				{0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14
+			},
+		},
+	},
+	{
 		name:    "ROUNDSD",
 		auxType: auxInt8,
 		argLen:  1,
@@ -14425,6 +14459,19 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
+		name:   "SQRTF",
+		argLen: 1,
+		asm:    arm.ASQRTF,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
+			},
+			outputs: []outputInfo{
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
+			},
+		},
+	},
+	{
 		name:   "ABSD",
 		argLen: 1,
 		asm:    arm.AABSD,
@@ -18087,6 +18134,19 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
+		name:   "FSQRTS",
+		argLen: 1,
+		asm:    arm64.AFSQRTS,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 9223372034707292160}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+			},
+			outputs: []outputInfo{
+				{0, 9223372034707292160}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+			},
+		},
+	},
+	{
 		name:   "REV",
 		argLen: 1,
 		asm:    arm64.AREV,
@@ -21880,6 +21940,19 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
+		name:   "SQRTF",
+		argLen: 1,
+		asm:    mips.ASQRTF,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 35183835217920}, // F0 F2 F4 F6 F8 F10 F12 F14 F16 F18 F20 F22 F24 F26 F28 F30
+			},
+			outputs: []outputInfo{
+				{0, 35183835217920}, // F0 F2 F4 F6 F8 F10 F12 F14 F16 F18 F20 F22 F24 F26 F28 F30
+			},
+		},
+	},
+	{
 		name:   "SLL",
 		argLen: 2,
 		asm:    mips.ASLL,
@@ -23359,6 +23432,19 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
+		name:   "SQRTF",
+		argLen: 1,
+		asm:    mips.ASQRTF,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 1152921504338411520}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+			},
+			outputs: []outputInfo{
+				{0, 1152921504338411520}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+			},
+		},
+	},
+	{
 		name:   "SLLV",
 		argLen: 2,
 		asm:    mips.ASLLV,
@@ -30943,6 +31029,19 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
+		name:   "FSQRTS",
+		argLen: 1,
+		asm:    s390x.AFSQRTS,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
+			},
+			outputs: []outputInfo{
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
+			},
+		},
+	},
+	{
 		name:         "LOCGR",
 		auxType:      auxS390XCCMask,
 		argLen:       3,
@@ -33876,10 +33975,10 @@ var opcodeTable = [...]opInfo{
 		asm:    wasm.AF32Sqrt,
 		reg: regInfo{
 			inputs: []inputInfo{
-				{0, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
 			outputs: []outputInfo{
-				{0, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
 		},
 	},
@@ -33889,10 +33988,10 @@ var opcodeTable = [...]opInfo{
 		asm:    wasm.AF32Trunc,
 		reg: regInfo{
 			inputs: []inputInfo{
-				{0, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
 			outputs: []outputInfo{
-				{0, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
 		},
 	},
@@ -33902,10 +34001,10 @@ var opcodeTable = [...]opInfo{
 		asm:    wasm.AF32Ceil,
 		reg: regInfo{
 			inputs: []inputInfo{
-				{0, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
 			outputs: []outputInfo{
-				{0, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
 		},
 	},
@@ -33915,10 +34014,10 @@ var opcodeTable = [...]opInfo{
 		asm:    wasm.AF32Floor,
 		reg: regInfo{
 			inputs: []inputInfo{
-				{0, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
 			outputs: []outputInfo{
-				{0, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
 		},
 	},
@@ -33928,10 +34027,10 @@ var opcodeTable = [...]opInfo{
 		asm:    wasm.AF32Nearest,
 		reg: regInfo{
 			inputs: []inputInfo{
-				{0, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
 			outputs: []outputInfo{
-				{0, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
 		},
 	},
@@ -33941,10 +34040,10 @@ var opcodeTable = [...]opInfo{
 		asm:    wasm.AF32Abs,
 		reg: regInfo{
 			inputs: []inputInfo{
-				{0, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
 			outputs: []outputInfo{
-				{0, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
 		},
 	},
@@ -33954,11 +34053,11 @@ var opcodeTable = [...]opInfo{
 		asm:    wasm.AF32Copysign,
 		reg: regInfo{
 			inputs: []inputInfo{
-				{0, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
-				{1, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
+				{1, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
 			outputs: []outputInfo{
-				{0, 281470681743360}, // F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31
+				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
 		},
 	},
@@ -35177,6 +35276,11 @@ var opcodeTable = [...]opInfo{
 		generic: true,
 	},
 	{
+		name:    "Sqrt32",
+		argLen:  1,
+		generic: true,
+	},
+	{
 		name:    "Floor",
 		argLen:  1,
 		generic: true,
diff --git a/src/cmd/compile/internal/ssa/rewrite386.go b/src/cmd/compile/internal/ssa/rewrite386.go
index 4e7fdb9e63..726d68e243 100644
--- a/src/cmd/compile/internal/ssa/rewrite386.go
+++ b/src/cmd/compile/internal/ssa/rewrite386.go
@@ -620,6 +620,9 @@ func rewriteValue386(v *Value) bool {
 	case OpSqrt:
 		v.Op = Op386SQRTSD
 		return true
+	case OpSqrt32:
+		v.Op = Op386SQRTSS
+		return true
 	case OpStaticCall:
 		v.Op = Op386CALLstatic
 		return true
diff --git a/src/cmd/compile/internal/ssa/rewriteAMD64.go b/src/cmd/compile/internal/ssa/rewriteAMD64.go
index 599137c806..52d0fd095d 100644
--- a/src/cmd/compile/internal/ssa/rewriteAMD64.go
+++ b/src/cmd/compile/internal/ssa/rewriteAMD64.go
@@ -1089,6 +1089,9 @@ func rewriteValueAMD64(v *Value) bool {
 	case OpSqrt:
 		v.Op = OpAMD64SQRTSD
 		return true
+	case OpSqrt32:
+		v.Op = OpAMD64SQRTSS
+		return true
 	case OpStaticCall:
 		v.Op = OpAMD64CALLstatic
 		return true
diff --git a/src/cmd/compile/internal/ssa/rewriteARM.go b/src/cmd/compile/internal/ssa/rewriteARM.go
index 1adbceb0ad..ed1f85e340 100644
--- a/src/cmd/compile/internal/ssa/rewriteARM.go
+++ b/src/cmd/compile/internal/ssa/rewriteARM.go
@@ -823,6 +823,9 @@ func rewriteValueARM(v *Value) bool {
 	case OpSqrt:
 		v.Op = OpARMSQRTD
 		return true
+	case OpSqrt32:
+		v.Op = OpARMSQRTF
+		return true
 	case OpStaticCall:
 		v.Op = OpARMCALLstatic
 		return true
diff --git a/src/cmd/compile/internal/ssa/rewriteARM64.go b/src/cmd/compile/internal/ssa/rewriteARM64.go
index ece834f996..55bb486600 100644
--- a/src/cmd/compile/internal/ssa/rewriteARM64.go
+++ b/src/cmd/compile/internal/ssa/rewriteARM64.go
@@ -999,6 +999,9 @@ func rewriteValueARM64(v *Value) bool {
 	case OpSqrt:
 		v.Op = OpARM64FSQRTD
 		return true
+	case OpSqrt32:
+		v.Op = OpARM64FSQRTS
+		return true
 	case OpStaticCall:
 		v.Op = OpARM64CALLstatic
 		return true
diff --git a/src/cmd/compile/internal/ssa/rewriteMIPS.go b/src/cmd/compile/internal/ssa/rewriteMIPS.go
index 0c074364df..fdf329cbd0 100644
--- a/src/cmd/compile/internal/ssa/rewriteMIPS.go
+++ b/src/cmd/compile/internal/ssa/rewriteMIPS.go
@@ -516,6 +516,9 @@ func rewriteValueMIPS(v *Value) bool {
 	case OpSqrt:
 		v.Op = OpMIPSSQRTD
 		return true
+	case OpSqrt32:
+		v.Op = OpMIPSSQRTF
+		return true
 	case OpStaticCall:
 		v.Op = OpMIPSCALLstatic
 		return true
diff --git a/src/cmd/compile/internal/ssa/rewriteMIPS64.go b/src/cmd/compile/internal/ssa/rewriteMIPS64.go
index 073cf8726c..541bdd694a 100644
--- a/src/cmd/compile/internal/ssa/rewriteMIPS64.go
+++ b/src/cmd/compile/internal/ssa/rewriteMIPS64.go
@@ -596,6 +596,9 @@ func rewriteValueMIPS64(v *Value) bool {
 	case OpSqrt:
 		v.Op = OpMIPS64SQRTD
 		return true
+	case OpSqrt32:
+		v.Op = OpMIPS64SQRTF
+		return true
 	case OpStaticCall:
 		v.Op = OpMIPS64CALLstatic
 		return true
diff --git a/src/cmd/compile/internal/ssa/rewritePPC64.go b/src/cmd/compile/internal/ssa/rewritePPC64.go
index 98f748e5fa..3357864291 100644
--- a/src/cmd/compile/internal/ssa/rewritePPC64.go
+++ b/src/cmd/compile/internal/ssa/rewritePPC64.go
@@ -743,6 +743,9 @@ func rewriteValuePPC64(v *Value) bool {
 	case OpSqrt:
 		v.Op = OpPPC64FSQRT
 		return true
+	case OpSqrt32:
+		v.Op = OpPPC64FSQRTS
+		return true
 	case OpStaticCall:
 		v.Op = OpPPC64CALLstatic
 		return true
diff --git a/src/cmd/compile/internal/ssa/rewriteRISCV64.go b/src/cmd/compile/internal/ssa/rewriteRISCV64.go
index bc47d76e87..36e152bd99 100644
--- a/src/cmd/compile/internal/ssa/rewriteRISCV64.go
+++ b/src/cmd/compile/internal/ssa/rewriteRISCV64.go
@@ -582,6 +582,9 @@ func rewriteValueRISCV64(v *Value) bool {
 	case OpSqrt:
 		v.Op = OpRISCV64FSQRTD
 		return true
+	case OpSqrt32:
+		v.Op = OpRISCV64FSQRTS
+		return true
 	case OpStaticCall:
 		v.Op = OpRISCV64CALLstatic
 		return true
diff --git a/src/cmd/compile/internal/ssa/rewriteS390X.go b/src/cmd/compile/internal/ssa/rewriteS390X.go
index 6adae3ff35..e0a5ff4cbb 100644
--- a/src/cmd/compile/internal/ssa/rewriteS390X.go
+++ b/src/cmd/compile/internal/ssa/rewriteS390X.go
@@ -792,6 +792,9 @@ func rewriteValueS390X(v *Value) bool {
 	case OpSqrt:
 		v.Op = OpS390XFSQRT
 		return true
+	case OpSqrt32:
+		v.Op = OpS390XFSQRTS
+		return true
 	case OpStaticCall:
 		v.Op = OpS390XCALLstatic
 		return true
diff --git a/src/cmd/compile/internal/ssa/rewriteWasm.go b/src/cmd/compile/internal/ssa/rewriteWasm.go
index c8ecefc736..7258bc4f8e 100644
--- a/src/cmd/compile/internal/ssa/rewriteWasm.go
+++ b/src/cmd/compile/internal/ssa/rewriteWasm.go
@@ -527,6 +527,9 @@ func rewriteValueWasm(v *Value) bool {
 	case OpSqrt:
 		v.Op = OpWasmF64Sqrt
 		return true
+	case OpSqrt32:
+		v.Op = OpWasmF32Sqrt
+		return true
 	case OpStaticCall:
 		v.Op = OpWasmLoweredStaticCall
 		return true
diff --git a/src/cmd/compile/internal/ssa/rewritegeneric.go b/src/cmd/compile/internal/ssa/rewritegeneric.go
index e5a27199a7..7e7cf458ff 100644
--- a/src/cmd/compile/internal/ssa/rewritegeneric.go
+++ b/src/cmd/compile/internal/ssa/rewritegeneric.go
@@ -4085,6 +4085,26 @@ func rewriteValuegeneric_OpCvt64Fto32F(v *Value) bool {
 		v.AuxInt = float32ToAuxInt(float32(c))
 		return true
 	}
+	// match: (Cvt64Fto32F sqrt0:(Sqrt (Cvt32Fto64F x)))
+	// cond: sqrt0.Uses==1
+	// result: (Sqrt32 x)
+	for {
+		sqrt0 := v_0
+		if sqrt0.Op != OpSqrt {
+			break
+		}
+		sqrt0_0 := sqrt0.Args[0]
+		if sqrt0_0.Op != OpCvt32Fto64F {
+			break
+		}
+		x := sqrt0_0.Args[0]
+		if !(sqrt0.Uses == 1) {
+			break
+		}
+		v.reset(OpSqrt32)
+		v.AddArg(x)
+		return true
+	}
 	return false
 }
 func rewriteValuegeneric_OpCvt64Fto64(v *Value) bool {
diff --git a/src/cmd/compile/internal/x86/ssa.go b/src/cmd/compile/internal/x86/ssa.go
index 4d134c6926..62982f4c6d 100644
--- a/src/cmd/compile/internal/x86/ssa.go
+++ b/src/cmd/compile/internal/x86/ssa.go
@@ -760,7 +760,7 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		p.To.Reg = v.Reg()
 	case ssa.Op386BSFL, ssa.Op386BSFW,
 		ssa.Op386BSRL, ssa.Op386BSRW,
-		ssa.Op386SQRTSD:
+		ssa.Op386SQRTSS, ssa.Op386SQRTSD:
 		p := s.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_REG
 		p.From.Reg = v.Args[0].Reg()
diff --git a/src/math/all_test.go b/src/math/all_test.go
index 3aae0373c7..d154457999 100644
--- a/src/math/all_test.go
+++ b/src/math/all_test.go
@@ -2067,6 +2067,21 @@ var fmaC = []struct{ x, y, z, want float64 }{
 	{-7.751454006381804e-05, 5.588653777189071e-308, -2.2207280111272877e-308, -2.2211612130544025e-308},
 }
 
+var sqrt32 = []float32{
+	0,
+	float32(Copysign(0, -1)),
+	float32(NaN()),
+	float32(Inf(1)),
+	float32(Inf(-1)),
+	1,
+	2,
+	-2,
+	4.9790119248836735e+00,
+	7.7388724745781045e+00,
+	-2.7688005719200159e-01,
+	-5.0106036182710749e+00,
+}
+
 func tolerance(a, b, e float64) bool {
 	// Multiplying by e here can underflow denormal values to zero.
 	// Check a==b so that at least if a and b are small and identical
@@ -3181,6 +3196,25 @@ func TestFloatMinMax(t *testing.T) {
 	}
 }
 
+var indirectSqrt = Sqrt
+
+// TestFloat32Sqrt checks the correctness of the float32 square root optimization result.
+func TestFloat32Sqrt(t *testing.T) {
+	for _, v := range sqrt32 {
+		want := float32(indirectSqrt(float64(v)))
+		got := float32(Sqrt(float64(v)))
+		if IsNaN(float64(want)) {
+			if !IsNaN(float64(got)) {
+				t.Errorf("got=%#v want=NaN, v=%#v", got, v)
+			}
+			continue
+		}
+		if got != want {
+			t.Errorf("got=%#v want=%#v, v=%#v", got, want, v)
+		}
+	}
+}
+
 // Benchmarks
 
 // Global exported variables are used to store the
diff --git a/test/codegen/math.go b/test/codegen/math.go
index ac8071400e..243ddb0494 100644
--- a/test/codegen/math.go
+++ b/test/codegen/math.go
@@ -55,6 +55,17 @@ func sqrt(x float64) float64 {
 	return math.Sqrt(x)
 }
 
+func sqrt32(x float32) float32 {
+	// amd64:"SQRTSS"
+	// 386/sse2:"SQRTSS" 386/softfloat:-"SQRTS"
+	// arm64:"FSQRTS"
+	// arm/7:"SQRTF"
+	// mips/hardfloat:"SQRTF" mips/softfloat:-"SQRTF"
+	// mips64/hardfloat:"SQRTF" mips64/softfloat:-"SQRTF"
+	// wasm:"F32Sqrt"
+	return float32(math.Sqrt(float64(x)))
+}
+
 // Check that it's using integer registers
 func abs(x, y float64) {
 	// amd64:"BTRQ\t[$]63"