Add loongarch compilation support to makefile

Author: junchao-loongson

Makefile

@@ -351,6 +351,11 @@ ifneq ($(filter ppc64le%,$(UNAME_M)),)
 	CUDA_POWER_ARCH = 1
 endif
 
+ifneq ($(filter loongarch64%,$(UNAME_M)),)
+	MK_CFLAGS   += -mlasx
+	MK_CXXFLAGS += -mlasx
+endif
+
 else
 	MK_CFLAGS   += -march=rv64gcv -mabi=lp64d
 	MK_CXXFLAGS += -march=rv64gcv -mabi=lp64d

ggml-quants.c

@@ -584,7 +584,7 @@ static __m256 __lasx_xvreplfr2vr_s(float val)
 static inline __m256i ____m256i(__m128i in)
 {
     __m256i out = __lasx_xvldi(0);
-    asm volatile (
+    __asm__ volatile (
         ".irp i," __ALL_REGS                "\n\t"
         " .ifc %[out], " XREGS_PREFIX"\\i    \n\t"
         "  .irp j," __ALL_REGS              "\n\t"
@@ -602,7 +602,7 @@ static inline __m256i ____m256i(__m128i in)
 static inline __m256i lasx_set_q(__m128i inhi, __m128i inlo)
 {
     __m256i out;
-    asm volatile (
+    __asm__ volatile (
         ".irp i," __ALL_REGS                "\n\t"
         " .ifc %[hi], " VREGS_PREFIX "\\i    \n\t"
         "  .irp j," __ALL_REGS              "\n\t"
@@ -632,7 +632,7 @@ static inline __m256i lasx_set_q(__m128i inhi, __m128i inlo)
 static inline __m128i lasx_extracti128_lo(__m256i in)
 {
     __m128i out;
-    asm volatile (
+    __asm__ volatile (
         ".ifnc %[out], %[in]                 \n\t"
         ".irp i," __ALL_REGS                "\n\t"
         " .ifc %[out], " VREGS_PREFIX "\\i   \n\t"
@@ -652,7 +652,7 @@ static inline __m128i lasx_extracti128_lo(__m256i in)
 static inline __m128i lasx_extracti128_hi(__m256i in)
 {
     __m128i out;
-    asm volatile (
+    __asm__ volatile (
         ".irp i," __ALL_REGS                "\n\t"
         " .ifc %[out], " VREGS_PREFIX "\\i   \n\t"
         "  .irp j," __ALL_REGS              "\n\t"