Windows/ARM64: Support Windows calling conventions.

Dear Microsoft: your butchering of the (perfectly fine) ARM64 ABI is a disgrace.
Thanks to Peter Cawley. #593

4 files changed, 36 insertions, 9 deletions

diff --git a/src/lj_asm_arm64.h b/src/lj_asm_arm64.h
index 4dd6b711..c2b17737 100644
--- a/src/lj_asm_arm64.h
+++ b/src/lj_asm_arm64.h
@@ -432,6 +432,11 @@ static void asm_gencall(ASMState *as, const CCallInfo *ci, IRRef *args)
   for (gpr = REGARG_FIRSTGPR; gpr <= REGARG_LASTGPR; gpr++)
     as->cost[gpr] = REGCOST(~0u, ASMREF_L);
   gpr = REGARG_FIRSTGPR;
+#if LJ_HASFFI && LJ_ABI_WIN
+  if ((ci->flags & CCI_VARARG)) {
+    fpr = REGARG_LASTFPR+1;
+  }
+#endif
   for (n = 0; n < nargs; n++) { /* Setup args. */
     IRRef ref = args[n];
     IRIns *ir = IR(ref);
@@ -442,6 +447,11 @@ static void asm_gencall(ASMState *as, const CCallInfo *ci, IRRef *args)
                      "reg %d not free", fpr);  /* Must have been evicted. */
           ra_leftov(as, fpr, ref);
           fpr++;
+#if LJ_HASFFI && LJ_ABI_WIN
+        } else if ((ci->flags & CCI_VARARG) && (gpr <= REGARG_LASTGPR)) {
+          Reg rf = ra_alloc1(as, ref, RSET_FPR);
+          emit_dn(as, A64I_FMOV_R_D, gpr++, rf & 31);
+#endif
         } else {
           Reg r = ra_alloc1(as, ref, RSET_FPR);
           int32_t al = spalign;
@@ -1943,6 +1953,9 @@ static Reg asm_setup_call_slots(ASMState *as, IRIns *ir, const CCallInfo *ci)
     int ngpr = REGARG_NUMGPR, nfpr = REGARG_NUMFPR;
     int spofs = 0, spalign = LJ_TARGET_OSX ? 0 : 7, nslots;
     asm_collectargs(as, ir, ci, args);
+#if LJ_ABI_WIN
+    if ((ci->flags & CCI_VARARG)) nfpr = 0;
+#endif
     for (i = 0; i < nargs; i++) {
       int al = spalign;
       if (!args[i]) {
@@ -1954,7 +1967,9 @@ static Reg asm_setup_call_slots(ASMState *as, IRIns *ir, const CCallInfo *ci)
 #endif
       } else if (irt_isfp(IR(args[i])->t)) {
         if (nfpr > 0) { nfpr--; continue; }
-#if LJ_TARGET_OSX
+#if LJ_ABI_WIN
+        if ((ci->flags & CCI_VARARG) && ngpr > 0) { ngpr--; continue; }
+#elif LJ_TARGET_OSX
         al |= irt_isnum(IR(args[i])->t) ? 7 : 3;
 #endif
       } else {
diff --git a/src/lj_ccall.c b/src/lj_ccall.c
index 00e753b9..5f95f5d8 100644
--- a/src/lj_ccall.c
+++ b/src/lj_ccall.c
@@ -985,6 +985,14 @@ static int ccall_set_args(lua_State *L, CTState *cts, CType *ct,
     fid = ctf->sib;
   }
 
+#if LJ_TARGET_ARM64 && LJ_ABI_WIN
+  if ((ct->info & CTF_VARARG)) {
+    nsp -= maxgpr * CTSIZE_PTR;  /* May end up with negative nsp. */
+    ngpr = maxgpr;
+    nfpr = CCALL_NARG_FPR;
+  }
+#endif
+
   /* Walk through all passed arguments. */
   for (o = L->base+1, narg = 1; o < top; o++, narg++) {
     CTypeID did;
@@ -1035,9 +1043,14 @@ static int ccall_set_args(lua_State *L, CTState *cts, CType *ct,
         align = CTSIZE_PTR-1;
       nsp = (nsp + align) & ~align;
     }
+#if LJ_TARGET_ARM64 && LJ_ABI_WIN
+    /* A negative nsp points into cc->gpr. Blame MS for their messy ABI. */
+    dp = ((uint8_t *)cc->stack) + (int32_t)nsp;
+#else
     dp = ((uint8_t *)cc->stack) + nsp;
+#endif
     nsp += CCALL_PACK_STACKARG ? sz : n * CTSIZE_PTR;
-    if (nsp > CCALL_SIZE_STACK) {  /* Too many arguments. */
+    if ((int32_t)nsp > CCALL_SIZE_STACK) {  /* Too many arguments. */
     err_nyi:
       lj_err_caller(L, LJ_ERR_FFI_NYICALL);
     }
@@ -1099,6 +1112,9 @@ static int ccall_set_args(lua_State *L, CTState *cts, CType *ct,
 #endif
   }
   if (fid) lj_err_caller(L, LJ_ERR_FFI_NUMARG);  /* Too few arguments. */
+#if LJ_TARGET_ARM64 && LJ_ABI_WIN
+  if ((int32_t)nsp < 0) nsp = 0;
+#endif
 
 #if LJ_TARGET_X64 || (LJ_TARGET_PPC && !LJ_ABI_SOFTFP)
   cc->nfpr = nfpr;  /* Required for vararg functions. */
diff --git a/src/lj_crecord.c b/src/lj_crecord.c
index d7a522fb..55d0b3ef 100644
--- a/src/lj_crecord.c
+++ b/src/lj_crecord.c
@@ -1118,13 +1118,9 @@ static TRef crec_call_args(jit_State *J, RecordFFData *rd,
     ngpr = 1;
   else if (ctype_cconv(ct->info) == CTCC_FASTCALL)
     ngpr = 2;
-#elif LJ_TARGET_ARM64
-#if LJ_ABI_WIN
-#error "NYI: ARM64 Windows ABI calling conventions"
-#elif LJ_TARGET_OSX
+#elif LJ_TARGET_ARM64 && LJ_TARGET_OSX
   int ngpr = CCALL_NARG_GPR;
 #endif
-#endif
 
   /* Skip initial attributes. */
   fid = ct->sib;
diff --git a/src/lj_emit_arm64.h b/src/lj_emit_arm64.h
index fef5d973..3c510492 100644
--- a/src/lj_emit_arm64.h
+++ b/src/lj_emit_arm64.h
@@ -124,9 +124,9 @@ static LJ_AINLINE uint32_t emit_lso_pair_candidate(A64Ins ai, int ofs, int sc)
   }
 }
 
-static void emit_lso(ASMState *as, A64Ins ai, Reg rd, Reg rn, int64_t ofs)
+static void emit_lso(ASMState *as, A64Ins ai, Reg rd, Reg rn, int64_t ofs64)
 {
-  int ot = emit_checkofs(ai, ofs), sc = (ai >> 30) & 3;
+  int ot = emit_checkofs(ai, ofs64), sc = (ai >> 30) & 3, ofs = (int)ofs64;
   lj_assertA(ot, "load/store offset %d out of range", ofs);
   /* Combine LDR/STR pairs to LDP/STP. */
   if ((sc == 2 || sc == 3) &&