diff --git a/src/mono/mono/arch/arm64/arm64-codegen.h b/src/mono/mono/arch/arm64/arm64-codegen.h
index 6c2ce0d5fee55f..adb16f148bf21b 100644
--- a/src/mono/mono/arch/arm64/arm64-codegen.h
+++ b/src/mono/mono/arch/arm64/arm64-codegen.h
@@ -486,18 +486,19 @@ MONO_RESTORE_WARNING
 #define arm_neon_ldrq_lit(p, rd, target) arm_emit ((p), 0b00011100000000000000000000000000 | (0b10 << 30) | (arm_get_disp19 ((p), (target)) << 5) | (rd))
 #define arm_neon_ldrq_lit_fixup(p, target) *((guint32*)p) = (*((guint32*)p) & 0xff00001f) | (arm_get_disp19 ((p), (target)) << 5)
 
+#define ARM_MAX_ARITH_IMM (0xfff)
+
 /* Arithmetic (immediate) */
 static G_GNUC_UNUSED inline guint32
 arm_encode_arith_imm (int imm, guint32 *shift)
 {
 	// FIXME:
-	g_assert ((imm >= 0) && (imm < 0xfff));
+	g_assert ((imm >= 0) && (imm < ARM_MAX_ARITH_IMM));
 	*shift = 0;
 	return (guint32)imm;
 }
-
 // FIXME:
-#define arm_is_arith_imm(imm)  (((imm) >= 0) && ((imm) < 0xfff))
+#define arm_is_arith_imm(imm)  (((imm) >= 0) && ((imm) < ARM_MAX_ARITH_IMM))
 
 #define arm_format_alu_imm(p, sf, op, S, rd, rn, imm) do { \
 	guint32 _imm12, _shift; \
@@ -1054,7 +1055,12 @@ arm_encode_arith_imm (int imm, guint32 *shift)
 #define TYPE_F32 0
 #define TYPE_F64 1
 
-/* NEON :: move SIMD register*/
+/* NEON :: paired loads/stores */
+#define arm_neon_ldp_stp(p, opc, l, rt1, rt2, rn, imm7) arm_emit ((p), 0b00101101000000000000000000000000 | (opc) << 30 | (l) << 22 | (imm7) << 15 | (rt2) << 10 | (rn) << 5 | (rt1))
+#define arm_neon_stp_16b(p, rt1, rt2, rn, imm) arm_neon_ldp_stp ((p), 0b10, 0b0, (rt1), (rt2), (rn), arm_encode_imm7 (imm, 16))
+#define arm_neon_ldp_16b(p, rt1, rt2, rn, imm) arm_neon_ldp_stp ((p), 0b10, 0b1, (rt1), (rt2), (rn), arm_encode_imm7 (imm, 16))
+
+/* NEON :: move SIMD register */
 #define arm_neon_mov(p, rd, rn) arm_neon_orr ((p), VREG_FULL, (rd), (rn), (rn))
 #define arm_neon_mov_8b(p, rd, rn) arm_neon_orr ((p), VREG_LOW, (rd), (rn), (rn))
 
diff --git a/src/mono/mono/mini/tramp-arm64.c b/src/mono/mono/mini/tramp-arm64.c
index af394802ecd98a..d26e16b2bbd322 100644
--- a/src/mono/mono/mini/tramp-arm64.c
+++ b/src/mono/mono/mini/tramp-arm64.c
@@ -139,7 +139,7 @@ mono_arch_create_generic_trampoline (MonoTrampolineType tramp_type, MonoTrampInf
 	/* Setup stack frame */
 	imm = frame_size;
 	mono_add_unwind_op_def_cfa (unwind_ops, code, buf, ARMREG_SP, 0);
-	while (imm > 256) {
+	while (imm > 256) { // TODO: can this be changed to ARM_MAX_ARITH_IMM?
 		arm_subx_imm (code, ARMREG_SP, ARMREG_SP, 256);
 		imm -= 256;
 		mono_add_unwind_op_def_cfa_offset (unwind_ops, code, buf, frame_size - imm);
@@ -159,8 +159,15 @@ mono_arch_create_generic_trampoline (MonoTrampolineType tramp_type, MonoTrampInf
 	gregs_regset = ~((1 << ARMREG_FP) | (1 << ARMREG_SP));
 	code = mono_arm_emit_store_regarray (code, gregs_regset, ARMREG_FP, gregs_offset);
 	/* Save fregs */
-	for (i = 0; i < num_fregs; ++i)
-		arm_strfpq (code, i, ARMREG_FP, fregs_offset + (i * 16));
+	for (i = 0; i < num_fregs; ++i) {
+		int offs = fregs_offset + (i * 16);
+		if (i+1 < num_fregs && arm_is_imm7_scaled (offs, 16)) {
+			arm_neon_stp_16b (code, i, i+1, ARMREG_FP, offs);
+			i++;
+		} else {
+			arm_strfpq (code, i, ARMREG_FP, offs);
+		}
+	}
 	/* Save trampoline arg */
 	arm_strx (code, ARMREG_IP1, ARMREG_FP, arg_offset);
 
@@ -173,9 +180,9 @@ mono_arch_create_generic_trampoline (MonoTrampolineType tramp_type, MonoTrampInf
 	/* Save caller sp */
 	arm_movx (code, ARMREG_IP1, ARMREG_FP);
 	imm = frame_size;
-	while (imm > 256) {
-		arm_addx_imm (code, ARMREG_IP1, ARMREG_IP1, 256);
-		imm -= 256;
+	while (imm > ARM_MAX_ARITH_IMM) {
+		arm_addx_imm (code, ARMREG_IP1, ARMREG_IP1, ARM_MAX_ARITH_IMM);
+		imm -= ARM_MAX_ARITH_IMM;
 	}
 	arm_addx_imm (code, ARMREG_IP1, ARMREG_IP1, imm);
 	arm_strx (code, ARMREG_IP1, ARMREG_IP0, MONO_STRUCT_OFFSET (MonoLMF, gregs) + (MONO_ARCH_LMF_REG_SP * 8));
@@ -264,8 +271,15 @@ mono_arch_create_generic_trampoline (MonoTrampolineType tramp_type, MonoTrampInf
 	/* Only have to load the argument regs (r0..r8) and the rgctx reg */
 	code = mono_arm_emit_load_regarray (code, 0x1ff | (1 << ARMREG_LR) | (1 << MONO_ARCH_RGCTX_REG), ARMREG_FP, gregs_offset);
 	/* Restore fregs */
-	for (i = 0; i < num_fregs; ++i)
-		arm_ldrfpq (code, i, ARMREG_FP, fregs_offset + (i * 16));
+	for (i = 0; i < num_fregs; ++i) {
+		int offs = fregs_offset + (i * 16);
+		if (i+1 < num_fregs && arm_is_imm7_scaled (offs, 16)) {
+			arm_neon_ldp_16b (code, i, i+1, ARMREG_FP, offs);
+			i++;
+		} else {
+			arm_ldrfpq (code, i, ARMREG_FP, offs);
+		}
+	}
 
 	/* Load the result */
 	arm_ldrx (code, ARMREG_IP1, ARMREG_FP, res_offset);
@@ -567,9 +581,9 @@ mono_arch_create_sdb_trampoline (gboolean single_step, MonoTrampInfo **info, gbo
 
 	/* Setup stack frame */
 	imm = frame_size;
-	while (imm > 256) {
-		arm_subx_imm (code, ARMREG_SP, ARMREG_SP, 256);
-		imm -= 256;
+	while (imm > ARM_MAX_ARITH_IMM) {
+		arm_subx_imm (code, ARMREG_SP, ARMREG_SP, ARM_MAX_ARITH_IMM);
+		imm -= ARM_MAX_ARITH_IMM;
 	}
 	arm_subx_imm (code, ARMREG_SP, ARMREG_SP, imm);
 	arm_stpx (code, ARMREG_FP, ARMREG_LR, ARMREG_SP, 0);
@@ -585,9 +599,9 @@ mono_arch_create_sdb_trampoline (gboolean single_step, MonoTrampInfo **info, gbo
 	/* Save caller sp */
 	arm_movx (code, ARMREG_IP1, ARMREG_FP);
 	imm = frame_size;
-	while (imm > 256) {
-		arm_addx_imm (code, ARMREG_IP1, ARMREG_IP1, 256);
-		imm -= 256;
+	while (imm > ARM_MAX_ARITH_IMM) {
+		arm_addx_imm (code, ARMREG_IP1, ARMREG_IP1, ARM_MAX_ARITH_IMM);
+		imm -= ARM_MAX_ARITH_IMM;
 	}
 	arm_addx_imm (code, ARMREG_IP1, ARMREG_IP1, imm);
 	arm_strx (code, ARMREG_IP1, ARMREG_FP, ctx_offset + G_STRUCT_OFFSET (MonoContext, regs) + (ARMREG_SP * 8));