void channel_loop(body_t body, tail_t tail) {
size_t num_loops = num_c16_blocks_ / unroll_regs_;
size_t loop_tail = num_c16_blocks_ - num_loops * unroll_regs_;
mov(reg_coff_s8, reg_soff);
xor_(reg_coff_f32, reg_coff_f32);
if (num_loops) {
xor_(reg_tmp, reg_tmp);
add(reg_tmp, c_in_xmm_ * unroll_regs_);
Label c_loop;
L(c_loop); {
body(unroll_regs_);
add(reg_coff_s8, c_in_xmm_ * unroll_regs_);
add(reg_coff_f32, sizeof(float) * c_in_xmm_ * unroll_regs_);
add(reg_tmp, c_in_xmm_ * unroll_regs_);
cmp(reg_tmp, reg_coff_max);
jle(c_loop);
}
}
if (loop_tail)
body(loop_tail);
if (c_tail_) {
add(reg_coff_s8, c_in_xmm_ * loop_tail);
add(reg_coff_f32, sizeof(float) * c_in_xmm_ * loop_tail);
tail();
}
}
void GSDrawScanlineCodeGenerator::Generate()
{
// TODO: on linux/mac rsi, rdi, xmm6-xmm15 are all caller saved
push(rbx);
push(rsi);
push(rdi);
push(rbp);
push(r12);
push(r13);
sub(rsp, 8 + 10 * 16);
for(int i = 6; i < 16; i++)
{
vmovdqa(ptr[rsp + (i - 6) * 16], Xmm(i));
}
mov(r10, (size_t)&m_test[0]);
mov(r11, (size_t)&m_local);
mov(r12, (size_t)m_local.gd);
mov(r13, (size_t)m_local.gd->vm);
Init();
// rcx = steps
// rsi = fza_base
// rdi = fza_offset
// r10 = &m_test[0]
// r11 = &m_local
// r12 = m_local->gd
// r13 = m_local->gd.vm
// xmm7 = vf (sprite && ltf)
// xmm8 = z
// xmm9 = f
// xmm10 = s
// xmm11 = t
// xmm12 = q
// xmm13 = rb
// xmm14 = ga
// xmm15 = test
if(!m_sel.edge)
{
align(16);
}
L("loop");
TestZ(xmm5, xmm6);
// ebp = za
if(m_sel.mmin)
{
SampleTextureLOD();
}
else
{
SampleTexture();
}
// ebp = za
// xmm2 = rb
// xmm3 = ga
AlphaTFX();
// ebp = za
// xmm2 = rb
// xmm3 = ga
ReadMask();
// ebp = za
// xmm2 = rb
// xmm3 = ga
// xmm4 = fm
// xmm5 = zm
TestAlpha();
// ebp = za
// xmm2 = rb
// xmm3 = ga
// xmm4 = fm
// xmm5 = zm
ColorTFX();
// ebp = za
// xmm2 = rb
// xmm3 = ga
// xmm4 = fm
// xmm5 = zm
Fog();
// ebp = za
// xmm2 = rb
// xmm3 = ga
// xmm4 = fm
// xmm5 = zm
ReadFrame();
// ebx = fa
// ebp = za
// xmm2 = rb
// xmm3 = ga
// xmm4 = fm
// xmm5 = zm
// xmm6 = fd
TestDestAlpha();
// ebx = fa
// ebp = za
// xmm2 = rb
// xmm3 = ga
// xmm4 = fm
// xmm5 = zm
// xmm6 = fd
WriteMask();
// ebx = fa
// edx = fzm
// ebp = za
// xmm2 = rb
// xmm3 = ga
// xmm4 = fm
// xmm5 = zm
// xmm6 = fd
WriteZBuf();
// ebx = fa
// edx = fzm
// xmm2 = rb
// xmm3 = ga
// xmm4 = fm
// xmm6 = fd
AlphaBlend();
// ebx = fa
// edx = fzm
// xmm2 = rb
// xmm3 = ga
// xmm4 = fm
// xmm6 = fd
WriteFrame();
L("step");
// if(steps <= 0) break;
if(!m_sel.edge)
{
test(rcx, rcx);
jle("exit", T_NEAR);
Step();
jmp("loop", T_NEAR);
}
L("exit");
for(int i = 6; i < 16; i++)
{
vmovdqa(Xmm(i), ptr[rsp + (i - 6) * 16]);
}
add(rsp, 8 + 10 * 16);
pop(r13);
pop(r12);
pop(rbp);
pop(rdi);
pop(rsi);
pop(rbx);
ret();
}
void GPUDrawScanlineCodeGenerator::Generate()
{
push(esi);
push(edi);
Init();
align(16);
L("loop");
// GSVector4i test = m_test[7 + (steps & (steps >> 31))];
mov(edx, ecx);
sar(edx, 31);
and(edx, ecx);
shl(edx, 4);
movdqa(xmm7, ptr[edx + (size_t)&m_test[7]]);
// movdqu(xmm1, ptr[edi]);
movq(xmm1, qword[edi]);
movhps(xmm1, qword[edi + 8]);
// ecx = steps
// esi = tex (tme)
// edi = fb
// xmm1 = fd
// xmm2 = s
// xmm3 = t
// xmm4 = r
// xmm5 = g
// xmm6 = b
// xmm7 = test
TestMask();
SampleTexture();
// xmm1 = fd
// xmm3 = a
// xmm4 = r
// xmm5 = g
// xmm6 = b
// xmm7 = test
// xmm0, xmm2 = free
ColorTFX();
AlphaBlend();
Dither();
WriteFrame();
L("step");
// if(steps <= 0) break;
test(ecx, ecx);
jle("exit", T_NEAR);
Step();
jmp("loop", T_NEAR);
L("exit");
pop(edi);
pop(esi);
ret(8);
}
void GSDrawScanlineCodeGenerator::Generate()
{
push(ebx);
push(esi);
push(edi);
push(ebp);
const int params = 16;
Init(params);
if(!m_sel.edge)
{
align(16);
}
L("loop");
// ecx = steps
// esi = fzbr
// edi = fzbc
// xmm0 = z/zi
// xmm2 = u (tme)
// xmm3 = v (tme)
// xmm5 = rb (!tme)
// xmm6 = ga (!tme)
// xmm7 = test
bool tme = m_sel.tfx != TFX_NONE;
TestZ(tme ? xmm5 : xmm2, tme ? xmm6 : xmm3);
// ecx = steps
// esi = fzbr
// edi = fzbc
// - xmm0
// xmm2 = u (tme)
// xmm3 = v (tme)
// xmm5 = rb (!tme)
// xmm6 = ga (!tme)
// xmm7 = test
SampleTexture();
// ecx = steps
// esi = fzbr
// edi = fzbc
// ebp = za
// - xmm2
// - xmm3
// - xmm4
// xmm5 = rb
// xmm6 = ga
// xmm7 = test
AlphaTFX();
// ecx = steps
// esi = fzbr
// edi = fzbc
// ebp = za
// xmm2 = gaf (TFX_HIGHLIGHT || TFX_HIGHLIGHT2 && !tcc)
// xmm5 = rb
// xmm6 = ga
// xmm7 = test
if(m_sel.fwrite)
{
movdqa(xmm3, xmmword[&m_env.fm]);
}
if(m_sel.zwrite)
{
movdqa(xmm4, xmmword[&m_env.zm]);
}
// ecx = steps
// esi = fzbr
// edi = fzbc
// ebp = za
// xmm2 = gaf (TFX_HIGHLIGHT || TFX_HIGHLIGHT2 && !tcc)
// xmm3 = fm
// xmm4 = zm
// xmm5 = rb
// xmm6 = ga
// xmm7 = test
TestAlpha();
// ecx = steps
// esi = fzbr
// edi = fzbc
// ebp = za
// xmm2 = gaf (TFX_HIGHLIGHT || TFX_HIGHLIGHT2 && !tcc)
// xmm3 = fm
// xmm4 = zm
// xmm5 = rb
// xmm6 = ga
// xmm7 = test
ColorTFX();
// ecx = steps
// esi = fzbr
// edi = fzbc
// ebp = za
// xmm3 = fm
// xmm4 = zm
// xmm5 = rb
// xmm6 = ga
// xmm7 = test
Fog();
// ecx = steps
// esi = fzbr
// edi = fzbc
// ebp = za
// xmm3 = fm
// xmm4 = zm
// xmm5 = rb
// xmm6 = ga
// xmm7 = test
ReadFrame();
// ecx = steps
// esi = fzbr
// edi = fzbc
// ebp = za
// xmm2 = fd
// xmm3 = fm
// xmm4 = zm
// xmm5 = rb
// xmm6 = ga
// xmm7 = test
TestDestAlpha();
// fm |= test;
// zm |= test;
if(m_sel.fwrite)
{
por(xmm3, xmm7);
}
if(m_sel.zwrite)
{
por(xmm4, xmm7);
}
// int fzm = ~(fm == GSVector4i::xffffffff()).ps32(zm == GSVector4i::xffffffff()).mask();
pcmpeqd(xmm1, xmm1);
if(m_sel.fwrite && m_sel.zwrite)
{
movdqa(xmm0, xmm1);
pcmpeqd(xmm1, xmm3);
pcmpeqd(xmm0, xmm4);
packssdw(xmm1, xmm0);
}
else if(m_sel.fwrite)
{
pcmpeqd(xmm1, xmm3);
packssdw(xmm1, xmm1);
}
else if(m_sel.zwrite)
{
pcmpeqd(xmm1, xmm4);
packssdw(xmm1, xmm1);
}
pmovmskb(edx, xmm1);
not(edx);
// ebx = fa
// ecx = steps
// edx = fzm
// esi = fzbr
// edi = fzbc
// ebp = za
// xmm2 = fd
// xmm3 = fm
// xmm4 = zm
// xmm5 = rb
// xmm6 = ga
WriteZBuf();
// ebx = fa
// ecx = steps
// edx = fzm
// esi = fzbr
// edi = fzbc
// - ebp
// xmm2 = fd
// xmm3 = fm
// - xmm4
// xmm5 = rb
// xmm6 = ga
AlphaBlend();
// ebx = fa
// ecx = steps
// edx = fzm
// esi = fzbr
// edi = fzbc
// xmm2 = fd
// xmm3 = fm
// xmm5 = rb
// xmm6 = ga
WriteFrame(params);
L("step");
// if(steps <= 0) break;
if(!m_sel.edge)
{
test(ecx, ecx);
jle("exit", T_NEAR);
Step();
jmp("loop", T_NEAR);
}
L("exit");
pop(ebp);
pop(edi);
pop(esi);
pop(ebx);
ret(8);
}
int encode_op(char *opcode, char *op_data)
{
int rd,rs,rt,imm,funct,shaft,target;
char tmp[256];
const char *fi = "%s %d";
const char *fg = "%s %%g%d";
const char *ff = "%s %%f%d";
const char *fl = "%s %s";
const char *fgi = "%s %%g%d, %d";
const char *fgl = "%s %%g%d, %s";
const char *fgg = "%s %%g%d, %%g%d";
const char *fggl = "%s %%g%d, %%g%d, %s";
const char *fggi = "%s %%g%d, %%g%d, %d";
const char *fggg = "%s %%g%d, %%g%d, %%g%d";
const char *fff = "%s %%f%d, %%f%d";
const char *fgf = "%s %%g%d, %%f%d";
const char *ffg = "%s %%f%d, %%g%d";
const char *fffl = "%s %%f%d, %%f%d, %s";
const char *ffff = "%s %%f%d, %%f%d, %%f%d";
const char *ffgi = "%s %%f%d, %%g%d, %d";
const char *ffgg = "%s %%f%d, %%g%d, %%g%d";
char lname[256];
shaft = funct = target = 0;
if(strcmp(opcode, "mvhi") == 0){
if(sscanf(op_data, fgi, tmp, &rs, &imm) == 3)
return mvhi(rs,0,imm);
}
if(strcmp(opcode, "mvlo") == 0){
if(sscanf(op_data, fgi, tmp, &rs, &imm) == 3)
return mvlo(rs,0,imm);
}
if(strcmp(opcode, "add") == 0){
if(sscanf(op_data, fggg, tmp, &rd, &rs,&rt) == 4)
return add(rs,rt,rd,0);
}
if(strcmp(opcode, "nor") == 0){
if(sscanf(op_data, fggg, tmp, &rd, &rs,&rt) == 4)
return nor(rs,rt,rd,0);
}
if(strcmp(opcode, "sub") == 0){
if(sscanf(op_data, fggg, tmp, &rd, &rs,&rt) == 4)
return sub(rs,rt,rd,0);
}
if(strcmp(opcode, "mul") == 0){
if(sscanf(op_data, fggg, tmp, &rd, &rs,&rt) == 4)
return mul(rs,rt,rd,0);
}
if(strcmp(opcode, "addi") == 0){
if(sscanf(op_data, fggi, tmp, &rt, &rs, &imm) == 4)
return addi(rs,rt,imm);
}
if(strcmp(opcode, "subi") == 0){
if(sscanf(op_data, fggi, tmp, &rt, &rs, &imm) == 4)
return subi(rs,rt,imm);
}
if(strcmp(opcode, "muli") == 0){
if(sscanf(op_data, fggi, tmp, &rt, &rs, &imm) == 4)
return muli(rs,rt,imm);
}
if(strcmp(opcode, "input") == 0){
if(sscanf(op_data, fg, tmp, &rd) == 2)
return input(0,0,rd,0);
}
if(strcmp(opcode, "inputw") == 0){
if(sscanf(op_data, fg, tmp, &rd) == 2)
return inputw(0,0,rd,0);
}
if(strcmp(opcode, "inputf") == 0){
if(sscanf(op_data, ff, tmp, &rd) == 2)
return inputf(0,0,rd,0);
}
if(strcmp(opcode, "output") == 0){
if(sscanf(op_data, fg, tmp, &rs) == 2)
return output(rs,0,0,0);
}
if(strcmp(opcode, "outputw") == 0){
if(sscanf(op_data, fg, tmp, &rs) == 2)
return outputw(rs,0,0,0);
}
if(strcmp(opcode, "outputf") == 0){
if(sscanf(op_data, ff, tmp, &rs) == 2)
return outputf(rs,0,0,0);
}
if(strcmp(opcode, "and") == 0){
if(sscanf(op_data, fggg, tmp, &rd, &rs,&rt) == 4)
return _and(rs,rt,rd,0);
}
if(strcmp(opcode, "or") == 0){
if(sscanf(op_data, fggg, tmp, &rd, &rs,&rt) == 4)
return _or(rs,rt,rd,0);
}
if(strcmp(opcode, "sll") == 0){
if(sscanf(op_data, fggg, tmp, &rd, &rs,&rt) == 4)
return sll(rs,rt,rd,0);
}
if(strcmp(opcode, "srl") == 0){
if(sscanf(op_data, fggg, tmp, &rd, &rs,&rt) == 4)
return srl(rs,rt,rd,0);
}
if(strcmp(opcode, "slli") == 0){
if(sscanf(op_data, fggi, tmp, &rt, &rs, &imm) == 4)
return slli(rs,rt,imm);
}
if(strcmp(opcode, "srli") == 0){
if(sscanf(op_data, fggi, tmp, &rt, &rs, &imm) == 4)
return srli(rs,rt,imm);
}
if(strcmp(opcode, "b") == 0){
if(sscanf(op_data, fg, tmp, &rs) == 2)
return b(rs,0,0,0);
}
if(strcmp(opcode, "jmp") == 0){
if(sscanf(op_data, fl, tmp, lname) == 2) {
strcpy(label_name[label_cnt],lname);
return jmp(label_cnt++);
}
}
if(strcmp(opcode, "jeq") == 0){
if(sscanf(op_data, fggl, tmp, &rs, &rt, lname) == 4) {
strcpy(label_name[label_cnt],lname);
return jeq(rs,rt,label_cnt++);
}
}
if(strcmp(opcode, "jne") == 0){
if(sscanf(op_data, fggl, tmp, &rs, &rt, lname) == 4) {
strcpy(label_name[label_cnt],lname);
return jne(rs,rt,label_cnt++);
}
}
if(strcmp(opcode, "jlt") == 0){
if(sscanf(op_data, fggl, tmp, &rs, &rt, lname) == 4) {
strcpy(label_name[label_cnt],lname);
return jlt(rs,rt,label_cnt++);
}
}
if(strcmp(opcode, "jle") == 0){
if(sscanf(op_data, fggl, tmp, &rs, &rt, lname) == 4) {
strcpy(label_name[label_cnt],lname);
return jle(rs,rt,label_cnt++);
}
}
if(strcmp(opcode, "call") == 0){
if(sscanf(op_data, fl, tmp, lname) == 2) {
strcpy(label_name[label_cnt],lname);
return call(label_cnt++);
}
}
if(strcmp(opcode, "callR") == 0){
if(sscanf(op_data, fg, tmp, &rs) == 2)
return callr(rs,0,0,0);
}
if(strcmp(opcode, "return") == 0){
return _return(0);
}
if(strcmp(opcode, "ld") == 0){
if(sscanf(op_data, fggg, tmp, &rd, &rs,&rt) == 4)
return ld(rs,rt,rd,0);
}
if(strcmp(opcode, "ldi") == 0){
if(sscanf(op_data, fggi, tmp, &rt, &rs, &imm) == 4)
return ldi(rs,rt,imm);
}
if(strcmp(opcode, "ldlr") == 0){
if(sscanf(op_data, fgi, tmp, &rs, &imm) == 3)
return ldlr(rs,0,imm);
}
if(strcmp(opcode, "fld") == 0){
if(sscanf(op_data, ffgg, tmp, &rd, &rs,&rt) == 4)
return fld(rs,rt,rd,0);
}
if(strcmp(opcode, "st") == 0){
if(sscanf(op_data, fggg, tmp, &rd, &rs,&rt) == 4)
return st(rs,rt,rd,0);
}
if(strcmp(opcode, "sti") == 0){
if(sscanf(op_data, fggi, tmp, &rt, &rs, &imm) == 4)
return sti(rs,rt,imm);
}
if(strcmp(opcode, "stlr") == 0){
if(sscanf(op_data, fgi, tmp, &rs, &imm) == 3)
return stlr(rs,0,imm);
}
if(strcmp(opcode, "fst") == 0){
if(sscanf(op_data, ffgg, tmp, &rd, &rs,&rt) == 4)
return fst(rs,rt,rd,0);
}
if(strcmp(opcode, "fadd") == 0){
if(sscanf(op_data, ffff, tmp, &rd, &rs, &rt) == 4)
return fadd(rs,rt,rd,0);
}
if(strcmp(opcode, "fsub") == 0){
if(sscanf(op_data, ffff, tmp, &rd, &rs, &rt) == 4)
return fsub(rs,rt,rd,0);
}
if(strcmp(opcode, "fmul") == 0){
if(sscanf(op_data, ffff, tmp, &rd, &rs, &rt) == 4)
return fmul(rs,rt,rd,0);
}
if(strcmp(opcode, "fdiv") == 0){
if(sscanf(op_data, ffff, tmp, &rd, &rs, &rt) == 4)
return fdiv(rs,rt,rd,0);
}
if(strcmp(opcode, "fsqrt") == 0){
if(sscanf(op_data, fff, tmp, &rd, &rs) == 3)
return fsqrt(rs,0,rd,0);
}
if(strcmp(opcode, "fabs") == 0){
if(sscanf(op_data, fff, tmp, &rd, &rs) == 3)
return _fabs(rs,0,rd,0);
}
if(strcmp(opcode, "fmov") == 0){
if(sscanf(op_data, fff, tmp, &rd, &rs) == 3)
return fmov(rs,0,rd,0);
}
if(strcmp(opcode, "fneg") == 0){
if(sscanf(op_data, fff, tmp, &rd, &rs) == 3)
return fneg(rs,0,rd,0);
}
if(strcmp(opcode, "fldi") == 0){
if(sscanf(op_data, ffgi, tmp, &rt, &rs, &imm) == 4)
return fldi(rs,rt,imm);
}
if(strcmp(opcode, "fsti") == 0){
if(sscanf(op_data, ffgi, tmp, &rt, &rs, &imm) == 4)
return fsti(rs,rt,imm);
}
if(strcmp(opcode, "fjeq") == 0){
if(sscanf(op_data, fffl, tmp, &rs, &rt, lname) == 4) {
strcpy(label_name[label_cnt],lname);
return fjeq(rs,rt,label_cnt++);
}
}
if(strcmp(opcode, "fjlt") == 0){
if(sscanf(op_data, fffl, tmp, &rs, &rt, lname) == 4) {
strcpy(label_name[label_cnt],lname);
return fjlt(rs,rt,label_cnt++);
}
}
if(strcmp(opcode, "halt") == 0){
return halt(0,0,0,0);
}
if(strcmp(opcode, "setL") == 0){
if(sscanf(op_data, fgl, tmp, &rd, lname) == 3) {
strcpy(label_name[label_cnt],lname);
return setl(0,rd,label_cnt++);
}
}
if(strcmp(opcode, "padd") == 0){
if(sscanf(op_data, fgi, tmp, &rt, &imm) == 3) {
return padd(0,rt,imm);
}
}
if(strcmp(opcode, "link") == 0){
if(sscanf(op_data, fi, tmp, &imm) == 2) {
return link(0,0,imm);
}
}
if(strcmp(opcode, "movlr") == 0){
return movlr(0,0,0,0);
}
if(strcmp(opcode, "btmplr") == 0){
return btmplr(0,0,0,0);
}
/*
if(strcmp(opcode, "padd") == 0){
if(sscanf(op_data, fgg, tmp, &rd, &rt) == 3) {
return padd(0,rt,d,0);
}
}
*/
return -1;
}