mirror of
https://github.com/PabloMK7/citra.git
synced 2024-11-25 00:44:05 +00:00
Merge pull request #379 from lioncash/sh
dyncom: Implement SHADD8/SHADD16/SHSUB8/SHSUB16/SHASX/SHSAX
This commit is contained in:
commit
fa4cc502e7
1 changed files with 110 additions and 8 deletions
|
@ -2252,13 +2252,48 @@ ARM_INST_PTR INTERPRETER_TRANSLATE(sel)(unsigned int inst, int index)
|
||||||
|
|
||||||
return inst_base;
|
return inst_base;
|
||||||
}
|
}
|
||||||
|
|
||||||
ARM_INST_PTR INTERPRETER_TRANSLATE(setend)(unsigned int inst, int index) { UNIMPLEMENTED_INSTRUCTION("SETEND"); }
|
ARM_INST_PTR INTERPRETER_TRANSLATE(setend)(unsigned int inst, int index) { UNIMPLEMENTED_INSTRUCTION("SETEND"); }
|
||||||
ARM_INST_PTR INTERPRETER_TRANSLATE(shadd16)(unsigned int inst, int index) { UNIMPLEMENTED_INSTRUCTION("SHADD16"); }
|
|
||||||
ARM_INST_PTR INTERPRETER_TRANSLATE(shadd8)(unsigned int inst, int index) { UNIMPLEMENTED_INSTRUCTION("SHADD8"); }
|
ARM_INST_PTR INTERPRETER_TRANSLATE(shadd8)(unsigned int inst, int index)
|
||||||
ARM_INST_PTR INTERPRETER_TRANSLATE(shaddsubx)(unsigned int inst, int index) { UNIMPLEMENTED_INSTRUCTION("SHADDSUBX"); }
|
{
|
||||||
ARM_INST_PTR INTERPRETER_TRANSLATE(shsub16)(unsigned int inst, int index) { UNIMPLEMENTED_INSTRUCTION("SHSUB16"); }
|
arm_inst* const inst_base = (arm_inst*)AllocBuffer(sizeof(arm_inst) + sizeof(generic_arm_inst));
|
||||||
ARM_INST_PTR INTERPRETER_TRANSLATE(shsub8)(unsigned int inst, int index) { UNIMPLEMENTED_INSTRUCTION("SHSUB8"); }
|
generic_arm_inst* const inst_cream = (generic_arm_inst*)inst_base->component;
|
||||||
ARM_INST_PTR INTERPRETER_TRANSLATE(shsubaddx)(unsigned int inst, int index) { UNIMPLEMENTED_INSTRUCTION("SHSUBADDX"); }
|
|
||||||
|
inst_base->cond = BITS(inst, 28, 31);
|
||||||
|
inst_base->idx = index;
|
||||||
|
inst_base->br = NON_BRANCH;
|
||||||
|
inst_base->load_r15 = 0;
|
||||||
|
|
||||||
|
inst_cream->op1 = BITS(inst, 20, 21);
|
||||||
|
inst_cream->op2 = BITS(inst, 5, 7);
|
||||||
|
inst_cream->Rm = BITS(inst, 0, 3);
|
||||||
|
inst_cream->Rn = BITS(inst, 16, 19);
|
||||||
|
inst_cream->Rd = BITS(inst, 12, 15);
|
||||||
|
|
||||||
|
return inst_base;
|
||||||
|
}
|
||||||
|
ARM_INST_PTR INTERPRETER_TRANSLATE(shadd16)(unsigned int inst, int index)
|
||||||
|
{
|
||||||
|
return INTERPRETER_TRANSLATE(shadd8)(inst, index);
|
||||||
|
}
|
||||||
|
ARM_INST_PTR INTERPRETER_TRANSLATE(shaddsubx)(unsigned int inst, int index)
|
||||||
|
{
|
||||||
|
return INTERPRETER_TRANSLATE(shadd8)(inst, index);
|
||||||
|
}
|
||||||
|
ARM_INST_PTR INTERPRETER_TRANSLATE(shsub8)(unsigned int inst, int index)
|
||||||
|
{
|
||||||
|
return INTERPRETER_TRANSLATE(shadd8)(inst, index);
|
||||||
|
}
|
||||||
|
ARM_INST_PTR INTERPRETER_TRANSLATE(shsub16)(unsigned int inst, int index)
|
||||||
|
{
|
||||||
|
return INTERPRETER_TRANSLATE(shadd8)(inst, index);
|
||||||
|
}
|
||||||
|
ARM_INST_PTR INTERPRETER_TRANSLATE(shsubaddx)(unsigned int inst, int index)
|
||||||
|
{
|
||||||
|
return INTERPRETER_TRANSLATE(shadd8)(inst, index);
|
||||||
|
}
|
||||||
|
|
||||||
ARM_INST_PTR INTERPRETER_TRANSLATE(smla)(unsigned int inst, int index)
|
ARM_INST_PTR INTERPRETER_TRANSLATE(smla)(unsigned int inst, int index)
|
||||||
{
|
{
|
||||||
arm_inst *inst_base = (arm_inst *)AllocBuffer(sizeof(arm_inst) + sizeof(smla_inst));
|
arm_inst *inst_base = (arm_inst *)AllocBuffer(sizeof(arm_inst) + sizeof(smla_inst));
|
||||||
|
@ -5229,12 +5264,79 @@ unsigned InterpreterMainLoop(ARMul_State* state) {
|
||||||
}
|
}
|
||||||
|
|
||||||
SETEND_INST:
|
SETEND_INST:
|
||||||
SHADD16_INST:
|
|
||||||
SHADD8_INST:
|
SHADD8_INST:
|
||||||
|
SHADD16_INST:
|
||||||
SHADDSUBX_INST:
|
SHADDSUBX_INST:
|
||||||
SHSUB16_INST:
|
|
||||||
SHSUB8_INST:
|
SHSUB8_INST:
|
||||||
|
SHSUB16_INST:
|
||||||
SHSUBADDX_INST:
|
SHSUBADDX_INST:
|
||||||
|
{
|
||||||
|
if (inst_base->cond == 0xE || CondPassed(cpu, inst_base->cond)) {
|
||||||
|
generic_arm_inst* const inst_cream = (generic_arm_inst*)inst_base->component;
|
||||||
|
|
||||||
|
const u8 op2 = inst_cream->op2;
|
||||||
|
const u32 rm_val = RM;
|
||||||
|
const u32 rn_val = RN;
|
||||||
|
|
||||||
|
if (op2 == 0x00 || op2 == 0x01 || op2 == 0x02 || op2 == 0x03) {
|
||||||
|
s32 lo_result = 0;
|
||||||
|
s32 hi_result = 0;
|
||||||
|
|
||||||
|
// SHADD16
|
||||||
|
if (op2 == 0x00) {
|
||||||
|
lo_result = ((s16)(rn_val & 0xFFFF) + (s16)(rm_val & 0xFFFF)) >> 1;
|
||||||
|
hi_result = ((s16)((rn_val >> 16) & 0xFFFF) + (s16)((rm_val >> 16) & 0xFFFF)) >> 1;
|
||||||
|
}
|
||||||
|
// SHASX
|
||||||
|
else if (op2 == 0x01) {
|
||||||
|
lo_result = ((s16)(rn_val & 0xFFFF) - (s16)((rm_val >> 16) & 0xFFFF)) >> 1;
|
||||||
|
hi_result = ((s16)((rn_val >> 16) & 0xFFFF) + (s16)(rm_val & 0xFFFF)) >> 1;
|
||||||
|
}
|
||||||
|
// SHSAX
|
||||||
|
else if (op2 == 0x02) {
|
||||||
|
lo_result = ((s16)(rn_val & 0xFFFF) + (s16)((rm_val >> 16) & 0xFFFF)) >> 1;
|
||||||
|
hi_result = ((s16)((rn_val >> 16) & 0xFFFF) - (s16)(rm_val & 0xFFFF)) >> 1;
|
||||||
|
}
|
||||||
|
// SHSUB16
|
||||||
|
else if (op2 == 0x03) {
|
||||||
|
lo_result = ((s16)(rn_val & 0xFFFF) - (s16)(rm_val & 0xFFFF)) >> 1;
|
||||||
|
hi_result = ((s16)((rn_val >> 16) & 0xFFFF) - (s16)((rm_val >> 16) & 0xFFFF)) >> 1;
|
||||||
|
}
|
||||||
|
|
||||||
|
RD = ((lo_result & 0xFFFF) | ((hi_result & 0xFFFF) << 16));
|
||||||
|
}
|
||||||
|
else if (op2 == 0x04 || op2 == 0x07) {
|
||||||
|
s16 lo_val1, lo_val2;
|
||||||
|
s16 hi_val1, hi_val2;
|
||||||
|
|
||||||
|
// SHADD8
|
||||||
|
if (op2 == 0x04) {
|
||||||
|
lo_val1 = ((s8)(rn_val & 0xFF) + (s8)(rm_val & 0xFF)) >> 1;
|
||||||
|
lo_val2 = ((s8)((rn_val >> 8) & 0xFF) + (s8)((rm_val >> 8) & 0xFF)) >> 1;
|
||||||
|
|
||||||
|
hi_val1 = ((s8)((rn_val >> 16) & 0xFF) + (s8)((rm_val >> 16) & 0xFF)) >> 1;
|
||||||
|
hi_val2 = ((s8)((rn_val >> 24) & 0xFF) + (s8)((rm_val >> 24) & 0xFF)) >> 1;
|
||||||
|
}
|
||||||
|
// SHSUB8
|
||||||
|
else {
|
||||||
|
lo_val1 = ((s8)(rn_val & 0xFF) - (s8)(rm_val & 0xFF)) >> 1;
|
||||||
|
lo_val2 = ((s8)((rn_val >> 8) & 0xFF) - (s8)((rm_val >> 8) & 0xFF)) >> 1;
|
||||||
|
|
||||||
|
hi_val1 = ((s8)((rn_val >> 16) & 0xFF) - (s8)((rm_val >> 16) & 0xFF)) >> 1;
|
||||||
|
hi_val2 = ((s8)((rn_val >> 24) & 0xFF) - (s8)((rm_val >> 24) & 0xFF)) >> 1;
|
||||||
|
}
|
||||||
|
|
||||||
|
RD = (lo_val1 & 0xFF) | ((lo_val2 & 0xFF) << 8) | ((hi_val1 & 0xFF) << 16) | ((hi_val2 & 0xFF) << 24);
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
cpu->Reg[15] += GET_INST_SIZE(cpu);
|
||||||
|
INC_PC(sizeof(generic_arm_inst));
|
||||||
|
FETCH_INST;
|
||||||
|
GOTO_NEXT_INST;
|
||||||
|
}
|
||||||
|
|
||||||
SMLA_INST:
|
SMLA_INST:
|
||||||
{
|
{
|
||||||
if ((inst_base->cond == 0xe) || CondPassed(cpu, inst_base->cond)) {
|
if ((inst_base->cond == 0xe) || CondPassed(cpu, inst_base->cond)) {
|
||||||
|
|
Loading…
Reference in a new issue