Implement objc_msgSend function for the LoongArch CPU architecture

Author: basilisk-dev

CMakeLists.txt

@@ -213,7 +213,7 @@ if (WIN32 AND NOT MINGW)
 		COMMAND echo ${ASSEMBLER} ${ASM_TARGET} -c "${CMAKE_SOURCE_DIR}/objc_msgSend.S" -o "${CMAKE_BINARY_DIR}/objc_msgSend.obj"
 		COMMAND ${ASSEMBLER} ${ASM_TARGET} -c "${CMAKE_SOURCE_DIR}/objc_msgSend.S" -o "${CMAKE_BINARY_DIR}/objc_msgSend.obj"
 		MAIN_DEPENDENCY objc_msgSend.S
-		DEPENDS objc_msgSend.aarch64.S objc_msgSend.arm.S objc_msgSend.mips.S objc_msgSend.x86-32.S objc_msgSend.x86-64.S
+		DEPENDS objc_msgSend.aarch64.S objc_msgSend.arm.S objc_msgSend.loongarch64.S objc_msgSend.mips.S objc_msgSend.x86-32.S objc_msgSend.x86-64.S
 	)
 	set(libobjc_ASM_OBJS block_trampolines.obj objc_msgSend.obj)
 endif()

objc/message.h

@@ -9,6 +9,8 @@
 #if defined(__x86_64) || defined(__i386) || defined(__arm__) ||                \
     defined(__mips_n64) || defined(__mips_n32) ||                              \
     defined(__ARM_ARCH_ISA_A64) ||                                             \
+    (defined(__loongarch__) && defined(__loongarch_lp64) &&                    \
+     defined(__loongarch_double_float)) ||                                     \
     (defined(__riscv) && __riscv_xlen == 64 &&                                 \
      defined(__riscv_float_abi_double))
 

objc_msgSend.S

@@ -10,6 +10,8 @@
 #include "objc_msgSend.aarch64.S"
 #elif defined(__riscv) && (__riscv_xlen == 64) && defined(__riscv_float_abi_double)
 #include "objc_msgSend.riscv64.S"
+#elif defined(__loongarch__) && defined(__loongarch_lp64) && defined(__loongarch_double_float)
+#include "objc_msgSend.loongarch64.S"
 #elif defined(__mips_n64) || defined(__mips_n32)
 #include "objc_msgSend.mips.S"
 #else

objc_msgSend.loongarch64.S

@@ -0,0 +1,134 @@
+#define ARGUMENT_SPILL_SIZE (10*8 + 8*8)
+
+.macro MSGSEND receiver, sel
+	.cfi_startproc
+	beqz     \receiver, 3f                // Skip everything if receiver is nil
+
+	andi     $t0, \receiver, SMALLOBJ_MASK
+	bnez     $t0, 5f
+
+	ld.d     $t0, \receiver, 0            // Load class into t0
+0:
+	ld.d     $t0, $t0, DTABLE_OFFSET      // dtable -> t0
+	ld.d     $t1, \sel, 0                 // selector->index -> t1
+	ld.w     $t2, $t0, SHIFT_OFFSET       // dtable->shift -> t2
+
+	addi.d   $t3, $zero, 8
+	beq      $t2, $t3, 1f
+	beqz     $t2, 2f
+
+	srli.d   $t2, $t1, 13                 // Extract byte 3 of sel index and multiply by 2^3
+	andi     $t2, $t2, 0x7f8
+	add.d    $t2, $t0, $t2
+	ld.d     $t0, $t2, DATA_OFFSET
+1:
+	srli.d   $t2, $t1, 5                  // Extract byte 2 of sel index and multiply by 2^3
+	andi     $t2, $t2, 0x7f8
+	add.d    $t2, $t0, $t2
+	ld.d     $t0, $t2, DATA_OFFSET
+2:
+	slli.d   $t2, $t1, 3                  // Multiply by 2^3
+	andi     $t2, $t2, 0x7f8
+	add.d    $t2, $t0, $t2
+	ld.d     $t0, $t2, DATA_OFFSET        // Slot pointer is now in t0
+
+	beqz     $t0, 4f                      // If the slot is nil, go to the C path
+
+	ld.d     $t0, $t0, SLOT_OFFSET        // Load the method from the slot
+	jr       $t0                          // Tail-call the method
+
+3:
+	move     \receiver, $zero
+	move     \sel, $zero
+	movgr2fr.d $fa0, $zero
+	movgr2fr.d $fa1, $zero
+	ret
+
+4:
+	addi.d   $sp, $sp, -ARGUMENT_SPILL_SIZE
+
+	// Spill function arguments.
+	st.d     $a0, $sp, 0
+	st.d     $a1, $sp, 8
+	st.d     $a2, $sp, 16
+	st.d     $a3, $sp, 24
+	st.d     $a4, $sp, 32
+	st.d     $a5, $sp, 40
+	st.d     $a6, $sp, 48
+	st.d     $a7, $sp, 56
+
+	// Spill FP arguments.
+	fst.d    $fa0, $sp, 64
+	fst.d    $fa1, $sp, 72
+	fst.d    $fa2, $sp, 80
+	fst.d    $fa3, $sp, 88
+	fst.d    $fa4, $sp, 96
+	fst.d    $fa5, $sp, 104
+	fst.d    $fa6, $sp, 112
+	fst.d    $fa7, $sp, 120
+
+	st.d     $fp, $sp, 128
+	st.d     $ra, $sp, 136
+
+	addi.d   $fp, $sp, 128
+	addi.d   $sp, $sp, -16
+	st.d     \receiver, $sp, 0            // Keep &self at sp for slowMsgLookup
+
+	.cfi_def_cfa fp, 16
+	.cfi_offset fp, -16
+	.cfi_offset ra, -8
+
+	move     $a0, $sp                     // &self in first argument
+	move     $a1, \sel
+	pcaddu18i $ra, %call36(CDECL(slowMsgLookup))
+	jirl     $ra, $ra, 0
+
+	move     $t0, $a0                     // IMP -> t0
+
+	ld.d     $a0, $sp, 16
+	ld.d     $a1, $sp, 24
+	ld.d     $a2, $sp, 32
+	ld.d     $a3, $sp, 40
+	ld.d     $a4, $sp, 48
+	ld.d     $a5, $sp, 56
+	ld.d     $a6, $sp, 64
+	ld.d     $a7, $sp, 72
+
+	fld.d    $fa0, $sp, 80
+	fld.d    $fa1, $sp, 88
+	fld.d    $fa2, $sp, 96
+	fld.d    $fa3, $sp, 104
+	fld.d    $fa4, $sp, 112
+	fld.d    $fa5, $sp, 120
+	fld.d    $fa6, $sp, 128
+	fld.d    $fa7, $sp, 136
+
+	ld.d     $fp, $sp, 144
+	ld.d     $ra, $sp, 152
+	ld.d     \receiver, $sp, 0
+
+	addi.d   $sp, $sp, ARGUMENT_SPILL_SIZE
+	addi.d   $sp, $sp, 16
+
+	jr       $t0                          // Tail-call the method
+
+5:
+	pcalau12i $t1, %got_pc_hi20(CDECL(SmallObjectClasses))
+	ld.d     $t1, $t1, %got_pc_lo12(CDECL(SmallObjectClasses))
+	slli.d   $t0, $t0, 3
+	ldx.d    $t0, $t1, $t0
+	b        0b
+	.cfi_endproc
+.endm
+
+.globl CDECL(objc_msgSend_fpret)
+TYPE_DIRECTIVE(CDECL(objc_msgSend_fpret), %function)
+.globl CDECL(objc_msgSend)
+TYPE_DIRECTIVE(CDECL(objc_msgSend), %function)
+.globl CDECL(objc_msgSend_stret)
+TYPE_DIRECTIVE(CDECL(objc_msgSend_stret), %function)
+CDECL(objc_msgSend):
+CDECL(objc_msgSend_fpret):
+	MSGSEND $a0, $a1
+CDECL(objc_msgSend_stret):
+	MSGSEND $a1, $a2 // Pointer to stack frame in a0