MONO_ADD_INS (cfg->cbb, arg);
}
+#define NEW_VARSTORE(cfg,dest,var,vartype,inst) do { \
+ MONO_INST_NEW ((cfg), (dest), OP_MOVE); \
+ (dest)->opcode = mono_type_to_regmove ((cfg), (vartype)); \
+ (dest)->klass = (var)->klass; \
+ (dest)->sreg1 = (inst)->dreg; \
+ (dest)->dreg = (var)->dreg; \
+ if ((dest)->opcode == OP_VMOVE) (dest)->klass = mono_class_from_mono_type ((vartype)); \
+ } while (0)
+
+#define NEW_ARGSTORE(cfg,dest,num,inst) NEW_VARSTORE ((cfg), (dest), cfg->args [(num)], cfg->arg_types [(num)], (inst))
+
+#define EMIT_NEW_ARGSTORE(cfg,dest,num,inst) do { NEW_ARGSTORE ((cfg), (dest), (num), (inst)); MONO_ADD_INS ((cfg)->cbb, (dest)); } while (0)
+
void
mono_arch_emit_call (MonoCompile *cfg, MonoCallInst *call)
{
case ArgOnStack:
case ArgValuetypeInReg:
case ArgValuetypeAddrInIReg:
- if (ainfo->storage == ArgOnStack && call->tail_call)
- NOT_IMPLEMENTED;
- if ((i >= sig->hasthis) && (MONO_TYPE_ISSTRUCT(sig->params [i - sig->hasthis]))) {
+ if (ainfo->storage == ArgOnStack && call->tail_call) {
+ MonoInst *call_inst = (MonoInst*)call;
+ cfg->args [i]->flags |= MONO_INST_VOLATILE;
+ EMIT_NEW_ARGSTORE (cfg, call_inst, i, in);
+ } else if ((i >= sig->hasthis) && (MONO_TYPE_ISSTRUCT(sig->params [i - sig->hasthis]))) {
guint32 align;
guint32 size;
if (ainfo->pair_storage [0] == ArgInIReg) {
MONO_INST_NEW (cfg, arg, OP_X86_LEA_MEMBASE);
- arg->dreg = ainfo->pair_regs [0];
+ arg->dreg = mono_alloc_ireg (cfg);
arg->sreg1 = load->dreg;
arg->inst_imm = 0;
MONO_ADD_INS (cfg->cbb, arg);
+ mono_call_inst_add_outarg_reg (cfg, call, arg->dreg, ainfo->pair_regs [0], FALSE);
} else {
MONO_INST_NEW (cfg, arg, OP_X86_PUSH);
arg->sreg1 = load->dreg;
case OP_DIV_IMM:
case OP_REM_IMM:
case OP_IDIV_IMM:
- case OP_IREM_IMM:
case OP_IDIV_UN_IMM:
case OP_IREM_UN_IMM:
mono_decompose_op_imm (cfg, bb, ins);
break;
+ case OP_IREM_IMM:
+ /* Keep the opcode if we can implement it efficiently */
+ if (!((ins->inst_imm > 0) && (mono_is_power_of_two (ins->inst_imm) != -1)))
+ mono_decompose_op_imm (cfg, bb, ins);
+ break;
case OP_COMPARE_IMM:
case OP_LCOMPARE_IMM:
if (!amd64_is_imm32 (ins->inst_imm)) {
}
/*
- * emit_tls_get:
+ * mono_amd64_emit_tls_get:
* @code: buffer to store code to
* @dreg: hard register where to place the result
* @tls_offset: offset info
*
- * emit_tls_get emits in @code the native code that puts in the dreg register
- * the item in the thread local storage identified by tls_offset.
+ * mono_amd64_emit_tls_get emits in @code the native code that puts in
+ * the dreg register the item in the thread local storage identified
+ * by tls_offset.
*
* Returns: a pointer to the end of the stored code
*/
-static guint8*
-emit_tls_get (guint8* code, int dreg, int tls_offset)
+guint8*
+mono_amd64_emit_tls_get (guint8* code, int dreg, int tls_offset)
{
#ifdef PLATFORM_WIN32
g_assert (tls_offset < 64);
mono_debug_open_block (cfg, bb, offset);
+ if (mono_break_at_bb_method && mono_method_desc_full_match (mono_break_at_bb_method, cfg->method) && bb->block_num == mono_break_at_bb_bb_num)
+ x86_breakpoint (code);
+
MONO_BB_FOR_EACH_INS (bb, ins) {
offset = code - cfg->native_code;
x86_prefix (code, X86_REP_PREFIX);
x86_nop (code);
break;
+ case OP_HARD_NOP:
+ x86_nop (code);
+ break;
case OP_NOP:
case OP_DUMMY_USE:
case OP_DUMMY_STORE:
amd64_div_reg_size (code, ins->sreg2, FALSE, 4);
}
break;
+ case OP_IREM_IMM: {
+ int power = mono_is_power_of_two (ins->inst_imm);
+
+ g_assert (ins->sreg1 == X86_EAX);
+ g_assert (ins->dreg == X86_EAX);
+ g_assert (power >= 0);
+
+ if (power == 0) {
+ amd64_mov_reg_imm (code, ins->dreg, 0);
+ break;
+ }
+
+ /* Based on gcc code */
+
+ /* Add compensation for negative dividents */
+ amd64_mov_reg_reg_size (code, AMD64_RDX, AMD64_RAX, 4);
+ if (power > 1)
+ amd64_shift_reg_imm_size (code, X86_SAR, AMD64_RDX, 31, 4);
+ amd64_shift_reg_imm_size (code, X86_SHR, AMD64_RDX, 32 - power, 4);
+ amd64_alu_reg_reg_size (code, X86_ADD, AMD64_RAX, AMD64_RDX, 4);
+ /* Compute remainder */
+ amd64_alu_reg_imm_size (code, X86_AND, AMD64_RAX, (1 << power) - 1, 4);
+ /* Remove compensation */
+ amd64_alu_reg_reg_size (code, X86_SUB, AMD64_RAX, AMD64_RDX, 4);
+ break;
+ }
case OP_LMUL_OVF:
amd64_imul_reg_reg (code, ins->sreg1, ins->sreg2);
EMIT_COND_SYSTEM_EXCEPTION (X86_CC_O, FALSE, "OverflowException");
amd64_lea_membase (code, AMD64_RSI, ins->inst_basereg, ins->inst_offset);
else
amd64_mov_reg_reg (code, AMD64_RSI, ins->inst_basereg, 8);
- amd64_lea_membase (code, AMD64_RDI, AMD64_RSP, (3 * 8) + (size - ins->inst_imm));
- amd64_mov_reg_imm (code, AMD64_RCX, (ins->inst_imm >> 3));
+ amd64_lea_membase (code, AMD64_RDI, AMD64_RSP, (3 * 8));
+ amd64_mov_reg_imm (code, AMD64_RCX, (size >> 3));
amd64_cld (code);
amd64_prefix (code, X86_REP_PREFIX);
amd64_movsd (code);
amd64_alu_reg_imm (code, X86_ADD, AMD64_RSP, 16);
break;
case OP_TLS_GET: {
- code = emit_tls_get (code, ins->dreg, ins->inst_offset);
+ code = mono_amd64_emit_tls_get (code, ins->dreg, ins->inst_offset);
break;
}
case OP_MEMORY_BARRIER: {
if (appdomain_tls_offset != -1 && lmf_tls_offset != -1) {
guint8 *buf, *no_domain_branch;
- code = emit_tls_get (code, AMD64_RAX, appdomain_tls_offset);
+ code = mono_amd64_emit_tls_get (code, AMD64_RAX, appdomain_tls_offset);
if ((domain >> 32) == 0)
amd64_mov_reg_imm_size (code, AMD64_ARG_REG1, domain, 4);
else
amd64_alu_reg_reg (code, X86_CMP, AMD64_RAX, AMD64_ARG_REG1);
no_domain_branch = code;
x86_branch8 (code, X86_CC_NE, 0, 0);
- code = emit_tls_get ( code, AMD64_RAX, lmf_addr_tls_offset);
+ code = mono_amd64_emit_tls_get ( code, AMD64_RAX, lmf_addr_tls_offset);
amd64_test_reg_reg (code, AMD64_RAX, AMD64_RAX);
buf = code;
x86_branch8 (code, X86_CC_NE, 0, 0);
} else {
if (lmf_addr_tls_offset != -1) {
/* Load lmf quicky using the FS register */
- code = emit_tls_get (code, AMD64_RAX, lmf_addr_tls_offset);
+ code = mono_amd64_emit_tls_get (code, AMD64_RAX, lmf_addr_tls_offset);
#ifdef PLATFORM_WIN32
/* The TLS key actually contains a pointer to the MonoJitTlsData structure */
/* FIXME: Add a separate key for LMF to avoid this */
pos = 0;
if (method->save_lmf) {
+ /* check if we need to restore protection of the stack after a stack overflow */
+ if (mono_get_jit_tls_offset () != -1) {
+ guint8 *patch;
+ code = mono_amd64_emit_tls_get (code, X86_ECX, mono_get_jit_tls_offset ());
+ /* we load the value in a separate instruction: this mechanism may be
+ * used later as a safer way to do thread interruption
+ */
+ amd64_mov_reg_membase (code, X86_ECX, X86_ECX, G_STRUCT_OFFSET (MonoJitTlsData, restore_stack_prot), 8);
+ x86_alu_reg_imm (code, X86_CMP, X86_ECX, 0);
+ patch = code;
+ x86_branch8 (code, X86_CC_Z, 0, FALSE);
+ /* note that the call trampoline will preserve eax/edx */
+ x86_call_reg (code, X86_ECX);
+ x86_patch (patch, code);
+ } else {
+ /* FIXME: maybe save the jit tls in the prolog */
+ }
if ((lmf_tls_offset != -1) && !optimize_for_xen) {
/*
* Optimized version which uses the mono_lmf TLS variable instead of indirection
* LOCKING: called with the domain lock held
*/
gpointer
-mono_arch_build_imt_thunk (MonoVTable *vtable, MonoDomain *domain, MonoIMTCheckItem **imt_entries, int count)
+mono_arch_build_imt_thunk (MonoVTable *vtable, MonoDomain *domain, MonoIMTCheckItem **imt_entries, int count,
+ gpointer fail_tramp)
{
int i;
int size = 0;
if (item->is_equals) {
if (item->check_target_idx) {
if (!item->compare_done) {
- if (amd64_is_imm32 (item->method))
+ if (amd64_is_imm32 (item->key))
item->chunk_size += CMP_SIZE;
else
item->chunk_size += MOV_REG_IMM_SIZE + CMP_REG_REG_SIZE;
item->chunk_size += MOV_REG_IMM_SIZE;
item->chunk_size += BR_SMALL_SIZE + JUMP_REG_SIZE;
} else {
- if (vtable_is_32bit)
- item->chunk_size += MOV_REG_IMM_32BIT_SIZE;
- else
- item->chunk_size += MOV_REG_IMM_SIZE;
- item->chunk_size += JUMP_REG_SIZE;
- /* with assert below:
- * item->chunk_size += CMP_SIZE + BR_SMALL_SIZE + 1;
- */
+ if (fail_tramp) {
+ item->chunk_size += MOV_REG_IMM_SIZE * 3 + CMP_REG_REG_SIZE +
+ BR_SMALL_SIZE + JUMP_REG_SIZE * 2;
+ } else {
+ if (vtable_is_32bit)
+ item->chunk_size += MOV_REG_IMM_32BIT_SIZE;
+ else
+ item->chunk_size += MOV_REG_IMM_SIZE;
+ item->chunk_size += JUMP_REG_SIZE;
+ /* with assert below:
+ * item->chunk_size += CMP_SIZE + BR_SMALL_SIZE + 1;
+ */
+ }
}
} else {
- if (amd64_is_imm32 (item->method))
+ if (amd64_is_imm32 (item->key))
item->chunk_size += CMP_SIZE;
else
item->chunk_size += MOV_REG_IMM_SIZE + CMP_REG_REG_SIZE;
}
size += item->chunk_size;
}
- code = mono_code_manager_reserve (domain->code_mp, size);
+ if (fail_tramp)
+ code = mono_method_alloc_generic_virtual_thunk (domain, size);
+ else
+ code = mono_code_manager_reserve (domain->code_mp, size);
start = code;
for (i = 0; i < count; ++i) {
MonoIMTCheckItem *item = imt_entries [i];
if (item->is_equals) {
if (item->check_target_idx) {
if (!item->compare_done) {
- if (amd64_is_imm32 (item->method))
- amd64_alu_reg_imm (code, X86_CMP, MONO_ARCH_IMT_REG, (guint32)(gssize)item->method);
+ if (amd64_is_imm32 (item->key))
+ amd64_alu_reg_imm (code, X86_CMP, MONO_ARCH_IMT_REG, (guint32)(gssize)item->key);
else {
- amd64_mov_reg_imm (code, AMD64_R10, item->method);
+ amd64_mov_reg_imm (code, AMD64_R10, item->key);
amd64_alu_reg_reg (code, X86_CMP, MONO_ARCH_IMT_REG, AMD64_R10);
}
}
item->jmp_code = code;
amd64_branch8 (code, X86_CC_NE, 0, FALSE);
/* See the comment below about R10 */
- amd64_mov_reg_imm (code, AMD64_R10, & (vtable->vtable [item->vtable_slot]));
- amd64_jump_membase (code, AMD64_R10, 0);
+ if (fail_tramp) {
+ amd64_mov_reg_imm (code, AMD64_R10, item->value.target_code);
+ amd64_jump_reg (code, AMD64_R10);
+ } else {
+ amd64_mov_reg_imm (code, AMD64_R10, & (vtable->vtable [item->value.vtable_slot]));
+ amd64_jump_membase (code, AMD64_R10, 0);
+ }
} else {
- /* enable the commented code to assert on wrong method */
+ if (fail_tramp) {
+ if (amd64_is_imm32 (item->key))
+ amd64_alu_reg_imm (code, X86_CMP, MONO_ARCH_IMT_REG, (guint32)(gssize)item->key);
+ else {
+ amd64_mov_reg_imm (code, AMD64_R10, item->key);
+ amd64_alu_reg_reg (code, X86_CMP, MONO_ARCH_IMT_REG, AMD64_R10);
+ }
+ item->jmp_code = code;
+ amd64_branch8 (code, X86_CC_NE, 0, FALSE);
+ amd64_mov_reg_imm (code, AMD64_R10, item->value.target_code);
+ amd64_jump_reg (code, AMD64_R10);
+ amd64_patch (item->jmp_code, code);
+ amd64_mov_reg_imm (code, AMD64_R10, fail_tramp);
+ amd64_jump_reg (code, AMD64_R10);
+ item->jmp_code = NULL;
+
+ } else {
+ /* enable the commented code to assert on wrong method */
#if 0
- if (amd64_is_imm32 (item->method))
- amd64_alu_reg_imm (code, X86_CMP, MONO_ARCH_IMT_REG, (guint32)(gssize)item->method);
- else {
- amd64_mov_reg_imm (code, AMD64_R10, item->method);
- amd64_alu_reg_reg (code, X86_CMP, MONO_ARCH_IMT_REG, AMD64_R10);
- }
- item->jmp_code = code;
- amd64_branch8 (code, X86_CC_NE, 0, FALSE);
- /* See the comment below about R10 */
- amd64_mov_reg_imm (code, AMD64_R10, & (vtable->vtable [item->vtable_slot]));
- amd64_jump_membase (code, AMD64_R10, 0);
- amd64_patch (item->jmp_code, code);
- amd64_breakpoint (code);
- item->jmp_code = NULL;
+ if (amd64_is_imm32 (item->key))
+ amd64_alu_reg_imm (code, X86_CMP, MONO_ARCH_IMT_REG, (guint32)(gssize)item->key);
+ else {
+ amd64_mov_reg_imm (code, AMD64_R10, item->key);
+ amd64_alu_reg_reg (code, X86_CMP, MONO_ARCH_IMT_REG, AMD64_R10);
+ }
+ item->jmp_code = code;
+ amd64_branch8 (code, X86_CC_NE, 0, FALSE);
+ /* See the comment below about R10 */
+ amd64_mov_reg_imm (code, AMD64_R10, & (vtable->vtable [item->value.vtable_slot]));
+ amd64_jump_membase (code, AMD64_R10, 0);
+ amd64_patch (item->jmp_code, code);
+ amd64_breakpoint (code);
+ item->jmp_code = NULL;
#else
- /* We're using R10 here because R11
- needs to be preserved. R10 needs
- to be preserved for calls which
- require a runtime generic context,
- but interface calls don't. */
- amd64_mov_reg_imm (code, AMD64_R10, & (vtable->vtable [item->vtable_slot]));
- amd64_jump_membase (code, AMD64_R10, 0);
+ /* We're using R10 here because R11
+ needs to be preserved. R10 needs
+ to be preserved for calls which
+ require a runtime generic context,
+ but interface calls don't. */
+ amd64_mov_reg_imm (code, AMD64_R10, & (vtable->vtable [item->value.vtable_slot]));
+ amd64_jump_membase (code, AMD64_R10, 0);
#endif
+ }
}
} else {
- if (amd64_is_imm32 (item->method))
- amd64_alu_reg_imm (code, X86_CMP, MONO_ARCH_IMT_REG, (guint32)(gssize)item->method);
+ if (amd64_is_imm32 (item->key))
+ amd64_alu_reg_imm (code, X86_CMP, MONO_ARCH_IMT_REG, (guint32)(gssize)item->key);
else {
- amd64_mov_reg_imm (code, AMD64_R10, item->method);
+ amd64_mov_reg_imm (code, AMD64_R10, item->key);
amd64_alu_reg_reg (code, X86_CMP, MONO_ARCH_IMT_REG, AMD64_R10);
}
item->jmp_code = code;
}
}
}
-
- mono_stats.imt_thunks_size += code - start;
+
+ if (!fail_tramp)
+ mono_stats.imt_thunks_size += code - start;
g_assert (code - start <= size);
return start;
projects / mono.git / blobdiff