feat(riscv): add RISC-V RV64A, FP and CSR instruction support

Assisted-by: Kimi K3

testdata/amo_fp_riscv64.s

@@ -0,0 +1,20 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+#include "textflag.h"
+
+// func atomicAdd(ptr *int64, val int64) int64
+TEXT ·atomicAdd(SB), NOSPLIT, $0-24
+	LD a+0(FP), A0
+	LD b+8(FP), A1
+	AMOADDD A1, (A0), A2
+	SD A2, ret+16(FP)
+	RET
+
+// func fpAdd(a, b float64) float64
+TEXT ·fpAdd(SB), NOSPLIT, $0-24
+	FLD a+0(FP), FA0
+	FLD b+8(FP), FA1
+	FADDD FA0, FA1, FA2
+	FSD FA2, ret+16(FP)
+	RET

testdata/csr_riscv64.s

@@ -0,0 +1,27 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+#include "textflag.h"
+
+// func readCSR(csr int64) int64
+// Reads a CSR into the return value.
+TEXT ·readCSR(SB), NOSPLIT, $0-16
+	LD a+0(FP), A0
+	CSRRS $0x300, X0, A1
+	SD A1, ret+8(FP)
+	RET
+
+// func setCSRBit(csr, bit int64) int64
+TEXT ·setCSRBit(SB), NOSPLIT, $0-24
+	LD a+0(FP), A0
+	LD b+8(FP), A1
+	CSRRS $0x304, A1, A2
+	SD A2, ret+16(FP)
+	RET
+
+// func writeCSR(val int64) int64
+TEXT ·writeCSR(SB), NOSPLIT, $0-16
+	LD a+0(FP), A0
+	CSRRW $0x305, A0, A1
+	SD A1, ret+8(FP)
+	RET

testdata/fma_riscv64.s

@@ -0,0 +1,24 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+#include "textflag.h"
+
+// func fma(a, b, c float64) float64
+// Computes a*b + c using fused multiply-add.
+TEXT ·fma(SB), NOSPLIT, $0-32
+	FLD a+0(FP), FA0
+	FLD b+8(FP), FA1
+	FLD c+16(FP), FA2
+	FMADDD FA0, FA1, FA2, FA3
+	FSD FA3, ret+24(FP)
+	RET
+
+// func fms(a, b, c float64) float64
+// Computes a*b - c using fused multiply-subtract.
+TEXT ·fms(SB), NOSPLIT, $0-32
+	FLD a+0(FP), FA0
+	FLD b+8(FP), FA1
+	FLD c+16(FP), FA2
+	FMSUBD FA0, FA1, FA2, FA3
+	FSD FA3, ret+24(FP)
+	RET

testdata/lrsc_cvt_riscv64.s

@@ -0,0 +1,37 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+#include "textflag.h"
+
+// func casLoop(ptr *int64, old, new int64) bool
+// Compare-and-swap using LR/SC.
+TEXT ·casLoop(SB), NOSPLIT, $0-32
+cas_retry:
+	LD a+0(FP), A0
+	LRD (A0), A1
+	LD b+8(FP), A2
+	BNE A1, A2, cas_fail
+	LD c+16(FP), A3
+	SCD A3, (A0), A4
+	BNE A4, X0, cas_retry
+	ADDI X0, $1, A5
+	SD A5, ret+24(FP)
+	RET
+cas_fail:
+	SD X0, ret+24(FP)
+	RET
+
+// func intToFloat(x int64) float64
+TEXT ·intToFloat(SB), NOSPLIT, $0-16
+	LD a+0(FP), A0
+	FCVTDL A0, FA0
+	FSD FA0, ret+8(FP)
+	RET
+
+// func compare(a, b float64) bool
+TEXT ·compare(SB), NOSPLIT, $0-24
+	FLD a+0(FP), FA0
+	FLD b+8(FP), FA1
+	FLTD FA0, FA1, A0
+	SD A0, ret+16(FP)
+	RET