fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-05 00:58:05 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

aco: add GFX12 wait counters

Browse source 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29225>
This commit is contained in:
Rhys Perry 2024-05-03 12:02:21 +01:00 • committed by Marge Bot
parent c5e13af73c
commit cadce0f3b7
3 changed files with 66 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

39
src/amd/compiler/aco_ir.cpp
View file

@ -1196,10 +1196,13 @@ can_swap_operands(aco_ptr<Instruction>& instr, aco_opcode* new_op, unsigned idx0
}
}
wait_imm::wait_imm() : exp(unset_counter), lgkm(unset_counter), vm(unset_counter), vs(unset_counter)
wait_imm::wait_imm()
: exp(unset_counter), lgkm(unset_counter), vm(unset_counter), vs(unset_counter),
sample(unset_counter), bvh(unset_counter), km(unset_counter)
{}
wait_imm::wait_imm(uint16_t vm_, uint16_t exp_, uint16_t lgkm_, uint16_t vs_)
: exp(exp_), lgkm(lgkm_), vm(vm_), vs(vs_)
: exp(exp_), lgkm(lgkm_), vm(vm_), vs(vs_), sample(unset_counter), bvh(unset_counter),
km(unset_counter)
{}
uint16_t
@ -1241,6 +1244,9 @@ wait_imm::max(enum amd_gfx_level gfx_level)
imm.exp = 7;
imm.lgkm = gfx_level >= GFX10 ? 63 : 15;
imm.vs = gfx_level >= GFX10 ? 63 : 0;
imm.sample = gfx_level >= GFX12 ? 63 : 0;
imm.bvh = gfx_level >= GFX12 ? 7 : 0;
imm.km = gfx_level >= GFX12 ? 31 : 0;
return imm;
}
@ -1253,7 +1259,31 @@ wait_imm::unpack(enum amd_gfx_level gfx_level, const Instruction* instr)
aco_opcode op = instr->opcode;
uint16_t packed = instr->salu().imm;
if (op == aco_opcode::s_waitcnt_expcnt) {
if (op == aco_opcode::s_wait_loadcnt) {
vm = std::min<uint8_t>(vm, packed);
} else if (op == aco_opcode::s_wait_storecnt) {
vs = std::min<uint8_t>(vs, packed);
} else if (op == aco_opcode::s_wait_samplecnt) {
sample = std::min<uint8_t>(sample, packed);
} else if (op == aco_opcode::s_wait_bvhcnt) {
bvh = std::min<uint8_t>(bvh, packed);
} else if (op == aco_opcode::s_wait_expcnt) {
exp = std::min<uint8_t>(exp, packed);
} else if (op == aco_opcode::s_wait_dscnt) {
lgkm = std::min<uint8_t>(lgkm, packed);
} else if (op == aco_opcode::s_wait_kmcnt) {
km = std::min<uint8_t>(km, packed);
} else if (op == aco_opcode::s_wait_loadcnt_dscnt) {
uint32_t vm2 = (packed >> 8) & 0x3f;
uint32_t ds = packed & 0x3f;
vm = std::min<uint8_t>(vm, vm2 == 0x3f ? wait_imm::unset_counter : vm2);
lgkm = std::min<uint8_t>(lgkm, ds == 0x3f ? wait_imm::unset_counter : ds);
} else if (op == aco_opcode::s_wait_storecnt_dscnt) {
uint32_t vs2 = (packed >> 8) & 0x3f;
uint32_t ds = packed & 0x3f;
vs = std::min<uint8_t>(vs, vs2 == 0x3f ? wait_imm::unset_counter : vs2);
lgkm = std::min<uint8_t>(lgkm, ds == 0x3f ? wait_imm::unset_counter : ds);
} else if (op == aco_opcode::s_waitcnt_expcnt) {
exp = std::min<uint8_t>(exp, packed);
} else if (op == aco_opcode::s_waitcnt_lgkmcnt) {
lgkm = std::min<uint8_t>(lgkm, packed);
@ -1325,6 +1355,9 @@ wait_imm::print(FILE* output) const
names[wait_type_vm] = "vm";
names[wait_type_lgkm] = "lgkm";
names[wait_type_vs] = "vs";
names[wait_type_sample] = "sample";
names[wait_type_bvh] = "bvh";
names[wait_type_km] = "km";
for (unsigned i = 0; i < wait_type_num; i++) {
if ((*this)[i] != unset_counter)
fprintf(output, "%s: %u\n", names[i], (*this)[i]);

12
src/amd/compiler/aco_ir.h
View file

@ -179,7 +179,11 @@ enum wait_type {
wait_type_vm = 2,
/* GFX10+ */
wait_type_vs = 3,
wait_type_num = 4,
/* GFX12+ */
wait_type_sample = 4,
wait_type_bvh = 5,
wait_type_km = 6,
wait_type_num = 7,
};
struct Instruction;
@ -191,6 +195,9 @@ struct wait_imm {
uint8_t lgkm;
uint8_t vm;
uint8_t vs;
uint8_t sample;
uint8_t bvh;
uint8_t km;
wait_imm();
wait_imm(uint16_t vm_, uint16_t exp_, uint16_t lgkm_, uint16_t vs_);
@ -223,6 +230,9 @@ static_assert(offsetof(wait_imm, exp) == wait_type_exp);
static_assert(offsetof(wait_imm, lgkm) == wait_type_lgkm);
static_assert(offsetof(wait_imm, vm) == wait_type_vm);
static_assert(offsetof(wait_imm, vs) == wait_type_vs);
static_assert(offsetof(wait_imm, sample) == wait_type_sample);
static_assert(offsetof(wait_imm, bvh) == wait_type_bvh);
static_assert(offsetof(wait_imm, km) == wait_type_km);
/* s_wait_event immediate bits. */
enum wait_event_imm : uint16_t {

23
src/amd/compiler/aco_print_ir.cpp
View file

@ -274,20 +274,35 @@ print_instr_format_specific(enum amd_gfx_level gfx_level, const Instruction* ins
case Format::SOPP: {
uint16_t imm = instr->salu().imm;
switch (instr->opcode) {
case aco_opcode::s_waitcnt: {
case aco_opcode::s_waitcnt:
case aco_opcode::s_wait_loadcnt_dscnt:
case aco_opcode::s_wait_storecnt_dscnt: {
wait_imm unpacked;
unpacked.unpack(gfx_level, instr);
const char* names[wait_type_num];
names[wait_type_exp] = "expcnt";
names[wait_type_vm] = "vmcnt";
names[wait_type_lgkm] = "lgkmcnt";
names[wait_type_vs] = "vscnt";
names[wait_type_vm] = gfx_level >= GFX12 ? "loadcnt" : "vmcnt";
names[wait_type_lgkm] = gfx_level >= GFX12 ? "dscnt" : "lgkmcnt";
names[wait_type_vs] = gfx_level >= GFX12 ? "storecnt" : "vscnt";
names[wait_type_sample] = "samplecnt";
names[wait_type_bvh] = "bvhcnt";
names[wait_type_km] = "kmcnt";
for (unsigned i = 0; i < wait_type_num; i++) {
if (unpacked[i] != wait_imm::unset_counter)
fprintf(output, " %s(%d)", names[i], unpacked[i]);
}
break;
}
case aco_opcode::s_wait_expcnt:
case aco_opcode::s_wait_dscnt:
case aco_opcode::s_wait_loadcnt:
case aco_opcode::s_wait_storecnt:
case aco_opcode::s_wait_samplecnt:
case aco_opcode::s_wait_bvhcnt:
case aco_opcode::s_wait_kmcnt: {
fprintf(output, " imm:%u", imm);
break;
}
case aco_opcode::s_waitcnt_depctr: {
unsigned va_vdst = (imm >> 12) & 0xf;
unsigned va_sdst = (imm >> 9) & 0x7;