OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
e34add229ff06203fe852b15cb7ff32cf9e4c344
third_party_mesa3d/src/panfrost/bifrost/bi_pack.c
Alyssa Rosenzweig e34add229f pan/bi: Fix backwards registers ports 
Will matter when packing multiple instructions per bundle.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4505>
2020-04-10 16:53:56 +02:00

1224 lines
41 KiB
C
Raw Blame History

/*
* Copyright (C) 2020 Collabora, Ltd.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include "compiler.h"
#define RETURN_PACKED(str) { \
uint64_t temp = 0; \
memcpy(&temp, &str, sizeof(str)); \
return temp; \
}
/* This file contains the final passes of the compiler. Running after
* scheduling and RA, the IR is now finalized, so we need to emit it to actual
* bits on the wire (as well as fixup branches) */
static uint64_t
bi_pack_header(bi_clause *clause, bi_clause *next, bool is_fragment)
{
struct bifrost_header header = {
.back_to_back = clause->back_to_back,
.no_end_of_shader = (next != NULL),
.elide_writes = is_fragment,
.branch_cond = clause->branch_conditional,
.datareg_writebarrier = clause->data_register_write_barrier,
.datareg = clause->data_register,
.scoreboard_deps = next ? next->dependencies : 0,
.scoreboard_index = clause->scoreboard_id,
.clause_type = clause->clause_type,
.next_clause_type = next ? next->clause_type : 0,
};
uint64_t u = 0;
memcpy(&u, &header, sizeof(header));
return u;
}
/* Represents the assignment of ports for a given bundle */
struct bi_registers {
/* Register to assign to each port */
unsigned port[4];
/* Read ports can be disabled */
bool enabled[2];
/* Should we write FMA? what about ADD? If only a single port is
* enabled it is in port 2, else ADD/FMA is 2/3 respectively */
bool write_fma, write_add;
/* Should we read with port 3? */
bool read_port3;
/* Packed uniform/constant */
uint8_t uniform_constant;
/* Whether writes are actually for the last instruction */
bool first_instruction;
};
static inline void
bi_print_ports(struct bi_registers *regs)
{
for (unsigned i = 0; i < 2; ++i) {
if (regs->enabled[i])
printf("port %u: %u\n", i, regs->port[i]);
}
if (regs->write_fma || regs->write_add) {
printf("port 2 (%s): %u\n",
regs->write_add ? "ADD" : "FMA",
regs->port[2]);
}
if ((regs->write_fma && regs->write_add) || regs->read_port3) {
printf("port 3 (%s): %u\n",
regs->read_port3 ? "read" : "FMA",
regs->port[3]);
}
}
/* The uniform/constant slot allows loading a contiguous 64-bit immediate or
* pushed uniform per bundle. Figure out which one we need in the bundle (the
* scheduler needs to ensure we only have one type per bundle), validate
* everything, and rewrite away the register/uniform indices to use 3-bit
* sources directly. */
static unsigned
bi_lookup_constant(bi_clause *clause, uint64_t cons, bool *hi, bool b64)
{
uint64_t want = (cons >> 4);
for (unsigned i = 0; i < clause->constant_count; ++i) {
/* Only check top 60-bits since that's what's actually embedded
* in the clause, the bottom 4-bits are bundle-inline */
unsigned candidates[2] = {
clause->constants[i] >> 4,
clause->constants[i] >> 36
};
if (!b64)
candidates[0] &= 0xFFFFFFFF;
if (candidates[0] == want)
return i;
if (candidates[1] == want && !b64) {
*hi = true;
return i;
}
}
unreachable("Invalid constant accessed");
}
static unsigned
bi_constant_field(unsigned idx)
{
assert(idx <= 5);
const unsigned values[] = {
4, 5, 6, 7, 2, 3
};
return values[idx] << 4;
}
static bool
bi_assign_uniform_constant_single(
struct bi_registers *regs,
bi_clause *clause,
bi_instruction *ins, bool assigned, bool fast_zero)
{
if (!ins)
return assigned;
bi_foreach_src(ins, s) {
if (s == 0 && (ins->type == BI_LOAD_VAR_ADDRESS || ins->type == BI_LOAD_ATTR)) continue;
if (ins->src[s] & BIR_INDEX_CONSTANT) {
bool hi = false;
bool b64 = nir_alu_type_get_type_size(ins->src_types[s]) > 32;
uint64_t cons = bi_get_immediate(ins, ins->src[s]);
unsigned idx = bi_lookup_constant(clause, cons, &hi, b64);
unsigned f = bi_constant_field(idx) | (cons & 0xF);
if (assigned && regs->uniform_constant != f)
unreachable("Mismatched uniform/const field: imm");
regs->uniform_constant = f;
ins->src[s] = BIR_INDEX_PASS | (hi ? BIFROST_SRC_CONST_HI : BIFROST_SRC_CONST_LO);
assigned = true;
} else if (ins->src[s] & BIR_INDEX_ZERO && (ins->type == BI_LOAD_UNIFORM || ins->type == BI_LOAD_VAR)) {
/* XXX: HACK UNTIL WE HAVE HI MATCHING DUE TO OVERFLOW XXX */
ins->src[s] = BIR_INDEX_PASS | BIFROST_SRC_CONST_HI;
} else if (ins->src[s] & BIR_INDEX_ZERO && !fast_zero) {
/* FMAs have a fast zero port, ADD needs to use the
* uniform/const port's special 0 mode handled here */
unsigned f = 0;
if (assigned && regs->uniform_constant != f)
unreachable("Mismatched uniform/const field: 0");
regs->uniform_constant = f;
ins->src[s] = BIR_INDEX_PASS | BIFROST_SRC_CONST_LO;
assigned = true;
} else if (s & BIR_INDEX_UNIFORM) {
unreachable("Push uniforms not implemented yet");
}
}
return assigned;
}
static void
bi_assign_uniform_constant(
bi_clause *clause,
struct bi_registers *regs,
bi_bundle bundle)
{
bool assigned =
bi_assign_uniform_constant_single(regs, clause, bundle.fma, false, true);
bi_assign_uniform_constant_single(regs, clause, bundle.add, assigned, false);
}
/* Assigns a port for reading, before anything is written */
static void
bi_assign_port_read(struct bi_registers *regs, unsigned src)
{
/* We only assign for registers */
if (!(src & BIR_INDEX_REGISTER))
return;
unsigned reg = src & ~BIR_INDEX_REGISTER;
/* Check if we already assigned the port */
for (unsigned i = 0; i <= 1; ++i) {
if (regs->port[i] == reg && regs->enabled[i])
return;
}
if (regs->port[3] == reg && regs->read_port3)
return;
/* Assign it now */
for (unsigned i = 0; i <= 1; ++i) {
if (!regs->enabled[i]) {
regs->port[i] = reg;
regs->enabled[i] = true;
return;
}
}
if (!regs->read_port3) {
regs->port[3] = reg;
regs->read_port3 = true;
return;
}
bi_print_ports(regs);
unreachable("Failed to find a free port for src");
}
static struct bi_registers
bi_assign_ports(bi_bundle now, bi_bundle prev)
{
struct bi_registers regs = { 0 };
/* We assign ports for the main register mechanism. Special ops
* use the data registers, which has its own mechanism entirely
* and thus gets skipped over here. */
unsigned read_dreg = now.add &&
bi_class_props[now.add->type] & BI_DATA_REG_SRC;
unsigned write_dreg = prev.add &&
bi_class_props[prev.add->type] & BI_DATA_REG_DEST;
/* First, assign reads */
if (now.fma)
bi_foreach_src(now.fma, src)
bi_assign_port_read(&regs, now.fma->src[src]);
if (now.add) {
bi_foreach_src(now.add, src) {
if (!(src == 0 && read_dreg))
bi_assign_port_read(&regs, now.add->src[src]);
}
}
/* Next, assign writes */
if (prev.add && prev.add->dest & BIR_INDEX_REGISTER && !write_dreg) {
regs.port[2] = prev.add->dest & ~BIR_INDEX_REGISTER;
regs.write_add = true;
}
if (prev.fma && prev.fma->dest & BIR_INDEX_REGISTER) {
unsigned r = prev.fma->dest & ~BIR_INDEX_REGISTER;
if (regs.write_add) {
/* Scheduler constraint: cannot read 3 and write 2 */
assert(!regs.read_port3);
regs.port[3] = r;
} else {
regs.port[2] = r;
}
regs.write_fma = true;
}
/* Finally, ensure port 1 > port 0 for the 63-x trick to function */
if (regs.enabled[0] && regs.enabled[1] && regs.port[1] < regs.port[0]) {
unsigned temp = regs.port[0];
regs.port[0] = regs.port[1];
regs.port[1] = temp;
}
return regs;
}
/* Determines the register control field, ignoring the first? flag */
static enum bifrost_reg_control
bi_pack_register_ctrl_lo(struct bi_registers r)
{
if (r.write_fma) {
if (r.write_add) {
assert(!r.read_port3);
return BIFROST_WRITE_ADD_P2_FMA_P3;
} else {
if (r.read_port3)
return BIFROST_WRITE_FMA_P2_READ_P3;
else
return BIFROST_WRITE_FMA_P2;
}
} else if (r.write_add) {
if (r.read_port3)
return BIFROST_WRITE_ADD_P2_READ_P3;
else
return BIFROST_WRITE_ADD_P2;
} else if (r.read_port3)
return BIFROST_READ_P3;
else
return BIFROST_REG_NONE;
}
/* Ditto but account for the first? flag this time */
static enum bifrost_reg_control
bi_pack_register_ctrl(struct bi_registers r)
{
enum bifrost_reg_control ctrl = bi_pack_register_ctrl_lo(r);
if (r.first_instruction) {
if (ctrl == BIFROST_REG_NONE)
ctrl = BIFROST_FIRST_NONE;
else if (ctrl == BIFROST_WRITE_FMA_P2_READ_P3)
ctrl = BIFROST_FIRST_WRITE_FMA_P2_READ_P3;
else
ctrl |= BIFROST_FIRST_NONE;
}
return ctrl;
}
static uint64_t
bi_pack_registers(struct bi_registers regs)
{
enum bifrost_reg_control ctrl = bi_pack_register_ctrl(regs);
struct bifrost_regs s = { 0 };
uint64_t packed = 0;
if (regs.enabled[1]) {
/* Gotta save that bit!~ Required by the 63-x trick */
assert(regs.port[1] > regs.port[0]);
assert(regs.enabled[0]);
/* Do the 63-x trick, see docs/disasm */
if (regs.port[0] > 31) {
regs.port[0] = 63 - regs.port[0];
regs.port[1] = 63 - regs.port[1];
}
assert(regs.port[0] <= 31);
assert(regs.port[1] <= 63);
s.ctrl = ctrl;
s.reg1 = regs.port[1];
s.reg0 = regs.port[0];
} else {
/* Port 1 disabled, so set to zero and use port 1 for ctrl */
s.ctrl = 0;
s.reg1 = ctrl << 2;
if (regs.enabled[0]) {
/* Bit 0 upper bit of port 0 */
s.reg1 |= (regs.port[0] >> 5);
/* Rest of port 0 in usual spot */
s.reg0 = (regs.port[0] & 0b11111);
} else {
/* Bit 1 set if port 0 also disabled */
s.reg1 |= (1 << 1);
}
}
/* When port 3 isn't used, we have to set it to port 2, and vice versa,
* or INSTR_INVALID_ENC is raised. The reason is unknown. */
bool has_port2 = regs.write_fma || regs.write_add;
bool has_port3 = regs.read_port3 || (regs.write_fma && regs.write_add);
if (!has_port3)
regs.port[3] = regs.port[2];
if (!has_port2)
regs.port[2] = regs.port[3];
s.reg3 = regs.port[3];
s.reg2 = regs.port[2];
s.uniform_const = regs.uniform_constant;
memcpy(&packed, &s, sizeof(s));
return packed;
}
static void
bi_set_data_register(bi_clause *clause, unsigned idx)
{
assert(idx & BIR_INDEX_REGISTER);
unsigned reg = idx & ~BIR_INDEX_REGISTER;
assert(reg <= 63);
clause->data_register = reg;
}
static void
bi_read_data_register(bi_clause *clause, bi_instruction *ins)
{
bi_set_data_register(clause, ins->src[0]);
}
static void
bi_write_data_register(bi_clause *clause, bi_instruction *ins)
{
bi_set_data_register(clause, ins->dest);
}
static enum bifrost_packed_src
bi_get_src_reg_port(struct bi_registers *regs, unsigned src)
{
unsigned reg = src & ~BIR_INDEX_REGISTER;
if (regs->port[0] == reg && regs->enabled[0])
return BIFROST_SRC_PORT0;
else if (regs->port[1] == reg && regs->enabled[1])
return BIFROST_SRC_PORT1;
else if (regs->port[3] == reg && regs->read_port3)
return BIFROST_SRC_PORT3;
else
unreachable("Tried to access register with no port");
}
static enum bifrost_packed_src
bi_get_src(bi_instruction *ins, struct bi_registers *regs, unsigned s, bool is_fma)
{
unsigned src = ins->src[s];
if (src & BIR_INDEX_REGISTER)
return bi_get_src_reg_port(regs, src);
else if (src & BIR_INDEX_ZERO && is_fma)
return BIFROST_SRC_STAGE;
else if (src & BIR_INDEX_PASS)
return src & ~BIR_INDEX_PASS;
else
unreachable("Unknown src");
}
/* Constructs a packed 2-bit swizzle for a 16-bit vec2 source. Source must be
* 16-bit and written components must correspond to valid swizzles (component x
* or y). */
static unsigned
bi_swiz16(bi_instruction *ins, unsigned src)
{
assert(nir_alu_type_get_type_size(ins->src_types[src]) == 16);
unsigned swizzle = 0;
for (unsigned c = 0; c < 2; ++c) {
if (!bi_writes_component(ins, src)) continue;
unsigned k = ins->swizzle[src][c];
assert(k < 1);
swizzle |= (k << c);
}
return swizzle;
}
static unsigned
bi_pack_fma_fma(bi_instruction *ins, struct bi_registers *regs)
{
/* (-a)(-b) = ab, so we only need one negate bit */
bool negate_mul = ins->src_neg[0] ^ ins->src_neg[1];
if (ins->dest_type == nir_type_float32) {
struct bifrost_fma_fma pack = {
.src0 = bi_get_src(ins, regs, 0, true),
.src1 = bi_get_src(ins, regs, 1, true),
.src2 = bi_get_src(ins, regs, 2, true),
.src0_abs = ins->src_abs[0],
.src1_abs = ins->src_abs[1],
.src2_abs = ins->src_abs[2],
.src0_neg = negate_mul,
.src2_neg = ins->src_neg[2],
.outmod = ins->outmod,
.roundmode = ins->roundmode,
.op = BIFROST_FMA_OP_FMA
};
RETURN_PACKED(pack);
} else if (ins->dest_type == nir_type_float16) {
struct bifrost_fma_fma16 pack = {
.src0 = bi_get_src(ins, regs, 0, true),
.src1 = bi_get_src(ins, regs, 1, true),
.src2 = bi_get_src(ins, regs, 2, true),
.swizzle_0 = bi_swiz16(ins, 0),
.swizzle_1 = bi_swiz16(ins, 1),
.swizzle_2 = bi_swiz16(ins, 2),
.src0_neg = negate_mul,
.src2_neg = ins->src_neg[2],
.outmod = ins->outmod,
.roundmode = ins->roundmode,
.op = BIFROST_FMA_OP_FMA16
};
RETURN_PACKED(pack);
} else {
unreachable("Invalid fma dest type");
}
}
static unsigned
bi_pack_fma_addmin_f32(bi_instruction *ins, struct bi_registers *regs)
{
unsigned op =
(ins->type == BI_ADD) ? BIFROST_FMA_OP_FADD32 :
(ins->op.minmax == BI_MINMAX_MIN) ? BIFROST_FMA_OP_FMIN32 :
BIFROST_FMA_OP_FMAX32;
struct bifrost_fma_add pack = {
.src0 = bi_get_src(ins, regs, 0, true),
.src1 = bi_get_src(ins, regs, 1, true),
.src0_abs = ins->src_abs[0],
.src1_abs = ins->src_abs[1],
.src0_neg = ins->src_neg[0],
.src1_neg = ins->src_neg[1],
.unk = 0x0,
.outmod = ins->outmod,
.roundmode = (ins->type == BI_ADD) ? ins->roundmode : ins->minmax,
.op = op
};
RETURN_PACKED(pack);
}
static unsigned
bi_pack_fma_addmin_f16(bi_instruction *ins, struct bi_registers *regs)
{
unsigned op =
(ins->type == BI_ADD) ? BIFROST_FMA_OP_FADD16 :
(ins->op.minmax == BI_MINMAX_MIN) ? BIFROST_FMA_OP_FMIN16 :
BIFROST_FMA_OP_FMAX16;
/* Absolute values are packed in a quirky way. Let k = src1 < src0. Let
* l be an auxiliary bit we encode. Then the hardware determines:
*
* abs0 = l || k
* abs1 = l && k
*
* Since add/min/max are commutative, this saves a bit by using the
* order of the operands as a bit (k). To pack this, first note:
*
* (l && k) implies (l || k).
*
* That is, if the second argument is abs'd, then the first argument
* also has abs. So there are three cases:
*
* Case 0: Neither src has absolute value. Then we have l = k = 0.
*
* Case 1: Exactly one src has absolute value. Assign that source to
* src0 and the other source to src1. Compute k = src1 < src0 based on
* that assignment. Then l = ~k.
*
* Case 2: Both sources have absolute value. Then we have l = k = 1.
* Note to force k = 1 requires that (src1 < src0) OR (src0 < src1).
* That is, this encoding is only valid if src1 and src0 are distinct.
* This is a scheduling restriction (XXX); if an op of this type
* requires both identical sources to have abs value, then we must
* schedule to ADD (which does not use this ordering trick).
*/
unsigned abs_0 = ins->src_abs[0], abs_1 = ins->src_abs[1];
unsigned src_0 = bi_get_src(ins, regs, 0, true);
unsigned src_1 = bi_get_src(ins, regs, 1, true);
bool l = false;
bool flip = false;
if (!abs_0 && !abs_1) {
/* Force k = 0 <===> NOT(src1 < src0) */
flip = (src_1 < src_0);
} else if (abs_0 && !abs_1) {
l = src_1 >= src_0;
} else if (abs_1 && !abs_0) {
flip = true;
l = src_0 >= src_1;
} else {
flip = (src_0 >= src_1);
l = true;
}
struct bifrost_fma_add_minmax16 pack = {
.src0 = flip ? src_1 : src_0,
.src1 = flip ? src_0 : src_1,
.src0_neg = ins->src_neg[flip ? 1 : 0],
.src1_neg = ins->src_neg[flip ? 0 : 1],
.abs1 = l,
.outmod = ins->outmod,
.mode = (ins->type == BI_ADD) ? ins->roundmode : ins->minmax,
.op = op
};
RETURN_PACKED(pack);
}
static unsigned
bi_pack_fma_addmin(bi_instruction *ins, struct bi_registers *regs)
{
if (ins->dest_type == nir_type_float32)
return bi_pack_fma_addmin_f32(ins, regs);
else if(ins->dest_type == nir_type_float16)
return bi_pack_fma_addmin_f16(ins, regs);
else
unreachable("Unknown FMA/ADD type");
}
static unsigned
bi_pack_fma_1src(bi_instruction *ins, struct bi_registers *regs, unsigned op)
{
struct bifrost_fma_inst pack = {
.src0 = bi_get_src(ins, regs, 0, true),
.op = op
};
RETURN_PACKED(pack);
}
static unsigned
bi_pack_add_1src(bi_instruction *ins, struct bi_registers *regs, unsigned op)
{
struct bifrost_add_inst pack = {
.src0 = bi_get_src(ins, regs, 0, true),
.op = op
};
RETURN_PACKED(pack);
}
static enum bifrost_csel_cond
bi_cond_to_csel(enum bi_cond cond, bool *flip, bool *invert, nir_alu_type T)
{
nir_alu_type B = nir_alu_type_get_base_type(T);
unsigned idx = (B == nir_type_float) ? 0 :
((B == nir_type_int) ? 1 : 2);
switch (cond){
case BI_COND_LT:
*flip = true;
case BI_COND_GT: {
const enum bifrost_csel_cond ops[] = {
BIFROST_FGT_F,
BIFROST_IGT_I,
BIFROST_UGT_I
};
return ops[idx];
}
case BI_COND_LE:
*flip = true;
case BI_COND_GE: {
const enum bifrost_csel_cond ops[] = {
BIFROST_FGE_F,
BIFROST_IGE_I,
BIFROST_UGE_I
};
return ops[idx];
}
case BI_COND_NE:
*invert = true;
case BI_COND_EQ: {
const enum bifrost_csel_cond ops[] = {
BIFROST_FEQ_F,
BIFROST_IEQ_F,
BIFROST_IEQ_F /* sign is irrelevant */
};
return ops[idx];
}
default:
unreachable("Invalid op for csel");
}
}
static unsigned
bi_pack_fma_csel(bi_instruction *ins, struct bi_registers *regs)
{
/* TODO: Use csel3 as well */
bool flip = false, invert = false;
enum bifrost_csel_cond cond =
bi_cond_to_csel(ins->csel_cond, &flip, &invert, ins->src_types[0]);
unsigned size = nir_alu_type_get_type_size(ins->dest_type);
unsigned cmp_0 = (flip ? 1 : 0);
unsigned cmp_1 = (flip ? 0 : 1);
unsigned res_0 = (invert ? 3 : 2);
unsigned res_1 = (invert ? 2 : 3);
struct bifrost_csel4 pack = {
.src0 = bi_get_src(ins, regs, cmp_0, true),
.src1 = bi_get_src(ins, regs, cmp_1, true),
.src2 = bi_get_src(ins, regs, res_0, true),
.src3 = bi_get_src(ins, regs, res_1, true),
.cond = cond,
.op = (size == 16) ? BIFROST_FMA_OP_CSEL4_V16 :
BIFROST_FMA_OP_CSEL4
};
RETURN_PACKED(pack);
}
/* We have a single convert opcode in the IR but a number of opcodes that could
* come out. In particular we have native opcodes for:
*
* [ui]16 --> [fui]32 -- int16_to_32
* f16 --> f32 -- float16_to_32
* f32 --> f16 -- float32_to_16
* f32 --> [ui]32 -- float32_to_int
* [ui]32 --> f32 -- int_to_float32
* [fui]16 --> [fui]16 -- f2i_i2f16
*/
static unsigned
bi_pack_fma_convert(bi_instruction *ins, struct bi_registers *regs)
{
nir_alu_type from_base = nir_alu_type_get_base_type(ins->src_types[0]);
unsigned from_size = nir_alu_type_get_type_size(ins->src_types[0]);
bool from_unsigned = from_base == nir_type_uint;
nir_alu_type to_base = nir_alu_type_get_base_type(ins->dest_type);
unsigned to_size = nir_alu_type_get_type_size(ins->dest_type);
bool to_unsigned = to_base == nir_type_uint;
/* Sanity check */
assert((from_base != to_base) || (from_size != to_size));
assert((MAX2(from_size, to_size) / MIN2(from_size, to_size)) <= 2);
if (from_size == 16 && to_size == 16) {
/* f2i_i2f16 */
unreachable("i16 not yet implemented");
} else if (from_size == 32 && to_size == 32) {
unsigned op = 0;
if (from_base == nir_type_float) {
op = BIFROST_FMA_FLOAT32_TO_INT(to_unsigned);
} else {
op = BIFROST_FMA_INT_TO_FLOAT32(from_unsigned);
}
return bi_pack_fma_1src(ins, regs, op);
} else if (from_size == 16 && to_size == 32) {
bool from_y = ins->swizzle[0][0];
if (from_base == nir_type_float) {
return bi_pack_fma_1src(ins, regs,
BIFROST_FMA_FLOAT16_TO_32(from_y));
} else {
unreachable("i16 not yet implemented");
}
} else if (from_size == 32 && to_size == 16) {
if (from_base == nir_type_float) {
/* TODO: second vectorized source? */
struct bifrost_fma_2src pack = {
.src0 = bi_get_src(ins, regs, 0, true),
.src1 = BIFROST_SRC_STAGE, /* 0 */
.op = BIFROST_FMA_FLOAT32_TO_16
};
RETURN_PACKED(pack);
} else {
unreachable("i16 not yet implemented");
}
}
unreachable("Unknown convert");
}
static unsigned
bi_pack_fma(bi_clause *clause, bi_bundle bundle, struct bi_registers *regs)
{
if (!bundle.fma)
return BIFROST_FMA_NOP;
switch (bundle.fma->type) {
case BI_ADD:
return bi_pack_fma_addmin(bundle.fma, regs);
case BI_CMP:
case BI_BITWISE:
return BIFROST_FMA_NOP;
case BI_CONVERT:
return bi_pack_fma_convert(bundle.fma, regs);
case BI_CSEL:
return bi_pack_fma_csel(bundle.fma, regs);
case BI_FMA:
return bi_pack_fma_fma(bundle.fma, regs);
case BI_FREXP:
case BI_ISUB:
return BIFROST_FMA_NOP;
case BI_MINMAX:
return bi_pack_fma_addmin(bundle.fma, regs);
case BI_MOV:
return bi_pack_fma_1src(bundle.fma, regs, BIFROST_FMA_OP_MOV);
case BI_SHIFT:
case BI_SWIZZLE:
case BI_ROUND:
return BIFROST_FMA_NOP;
default:
unreachable("Cannot encode class as FMA");
}
}
static unsigned
bi_pack_add_ld_vary(bi_clause *clause, bi_instruction *ins, struct bi_registers *regs)
{
unsigned size = nir_alu_type_get_type_size(ins->dest_type);
assert(size == 32 || size == 16);
unsigned op = (size == 32) ?
BIFROST_ADD_OP_LD_VAR_32 :
BIFROST_ADD_OP_LD_VAR_16;
unsigned cmask = bi_from_bytemask(ins->writemask, size / 8);
unsigned channels = util_bitcount(cmask);
assert(cmask == ((1 << channels) - 1));
unsigned packed_addr = 0;
if (ins->src[0] & BIR_INDEX_CONSTANT) {
/* Direct uses address field directly */
packed_addr = bi_get_immediate(ins, ins->src[0]);
assert(packed_addr < 0b1000);
} else {
/* Indirect gets an extra source */
packed_addr = bi_get_src(ins, regs, 0, false) | 0b11000;
}
/* The destination is thrown in the data register */
assert(ins->dest & BIR_INDEX_REGISTER);
clause->data_register = ins->dest & ~BIR_INDEX_REGISTER;
assert(channels >= 1 && channels <= 4);
struct bifrost_ld_var pack = {
.src0 = bi_get_src(ins, regs, 1, false),
.addr = packed_addr,
.channels = MALI_POSITIVE(channels),
.interp_mode = ins->load_vary.interp_mode,
.reuse = ins->load_vary.reuse,
.flat = ins->load_vary.flat,
.op = op
};
RETURN_PACKED(pack);
}
static unsigned
bi_pack_add_2src(bi_instruction *ins, struct bi_registers *regs, unsigned op)
{
struct bifrost_add_2src pack = {
.src0 = bi_get_src(ins, regs, 0, true),
.src1 = bi_get_src(ins, regs, 1, true),
.op = op
};
RETURN_PACKED(pack);
}
static unsigned
bi_pack_add_addmin_f32(bi_instruction *ins, struct bi_registers *regs)
{
unsigned op =
(ins->type == BI_ADD) ? BIFROST_ADD_OP_FADD32 :
(ins->op.minmax == BI_MINMAX_MIN) ? BIFROST_ADD_OP_FMIN32 :
BIFROST_ADD_OP_FMAX32;
struct bifrost_add_faddmin pack = {
.src0 = bi_get_src(ins, regs, 0, true),
.src1 = bi_get_src(ins, regs, 1, true),
.src0_abs = ins->src_abs[0],
.src1_abs = ins->src_abs[1],
.src0_neg = ins->src_neg[0],
.src1_neg = ins->src_neg[1],
.outmod = ins->outmod,
.mode = (ins->type == BI_ADD) ? ins->roundmode : ins->minmax,
.op = op
};
RETURN_PACKED(pack);
}
static unsigned
bi_pack_add_addmin(bi_instruction *ins, struct bi_registers *regs)
{
if (ins->dest_type == nir_type_float32)
return bi_pack_add_addmin_f32(ins, regs);
else if(ins->dest_type == nir_type_float16)
unreachable("todo");
//return bi_pack_add_addmin_f16(ins, regs);
else
unreachable("Unknown FMA/ADD type");
}
static unsigned
bi_pack_add_ld_ubo(bi_clause *clause, bi_instruction *ins, struct bi_registers *regs)
{
unsigned components = bi_load32_components(ins);
const unsigned ops[4] = {
BIFROST_ADD_OP_LD_UBO_1,
BIFROST_ADD_OP_LD_UBO_2,
BIFROST_ADD_OP_LD_UBO_3,
BIFROST_ADD_OP_LD_UBO_4
};
bi_write_data_register(clause, ins);
return bi_pack_add_2src(ins, regs, ops[components - 1]);
}
static enum bifrost_ldst_type
bi_pack_ldst_type(nir_alu_type T)
{
switch (T) {
case nir_type_float16: return BIFROST_LDST_F16;
case nir_type_float32: return BIFROST_LDST_F32;
case nir_type_int32: return BIFROST_LDST_I32;
case nir_type_uint32: return BIFROST_LDST_U32;
default: unreachable("Invalid type loaded");
}
}
static unsigned
bi_pack_add_ld_var_addr(bi_clause *clause, bi_instruction *ins, struct bi_registers *regs)
{
struct bifrost_ld_var_addr pack = {
.src0 = bi_get_src(ins, regs, 1, false),
.src1 = bi_get_src(ins, regs, 2, false),
.location = bi_get_immediate(ins, ins->src[0]),
.type = bi_pack_ldst_type(ins->src_types[3]),
.op = BIFROST_ADD_OP_LD_VAR_ADDR
};
bi_write_data_register(clause, ins);
RETURN_PACKED(pack);
}
static unsigned
bi_pack_add_ld_attr(bi_clause *clause, bi_instruction *ins, struct bi_registers *regs)
{
struct bifrost_ld_attr pack = {
.src0 = bi_get_src(ins, regs, 1, false),
.src1 = bi_get_src(ins, regs, 2, false),
.location = bi_get_immediate(ins, ins->src[0]),
.channels = MALI_POSITIVE(bi_load32_components(ins)),
.type = bi_pack_ldst_type(ins->dest_type),
.op = BIFROST_ADD_OP_LD_ATTR
};
bi_write_data_register(clause, ins);
RETURN_PACKED(pack);
}
static unsigned
bi_pack_add_st_vary(bi_clause *clause, bi_instruction *ins, struct bi_registers *regs)
{
assert(ins->store_channels >= 1 && ins->store_channels <= 4);
struct bifrost_st_vary pack = {
.src0 = bi_get_src(ins, regs, 1, false),
.src1 = bi_get_src(ins, regs, 2, false),
.src2 = bi_get_src(ins, regs, 3, false),
.channels = MALI_POSITIVE(ins->store_channels),
.op = BIFROST_ADD_OP_ST_VAR
};
bi_read_data_register(clause, ins);
RETURN_PACKED(pack);
}
static unsigned
bi_pack_add_atest(bi_clause *clause, bi_instruction *ins, struct bi_registers *regs)
{
/* TODO: fp16 */
assert(ins->src_types[1] == nir_type_float32);
struct bifrost_add_atest pack = {
.src0 = bi_get_src(ins, regs, 0, false),
.src1 = bi_get_src(ins, regs, 1, false),
.component = 1, /* Set for fp32 */
.op = BIFROST_ADD_OP_ATEST,
};
/* Despite *also* writing with the usual mechanism... quirky and
* perhaps unnecessary, but let's match the blob */
clause->data_register = ins->dest & ~BIR_INDEX_REGISTER;
RETURN_PACKED(pack);
}
static unsigned
bi_pack_add_blend(bi_instruction *ins, struct bi_registers *regs)
{
struct bifrost_add_inst pack = {
.src0 = bi_get_src(ins, regs, 0, false),
.op = BIFROST_ADD_OP_BLEND
};
/* TODO: Pack location in uniform_const */
assert(ins->blend_location == 0);
RETURN_PACKED(pack);
}
static unsigned
bi_pack_add_special(bi_instruction *ins, struct bi_registers *regs)
{
unsigned op = 0;
bool fp16 = ins->dest_type == nir_type_float16;
bool Y = ins->swizzle[0][0];
if (ins->op.special == BI_SPECIAL_FRCP) {
op = fp16 ?
(Y ? BIFROST_ADD_OP_FRCP_FAST_F16_Y :
BIFROST_ADD_OP_FRCP_FAST_F16_X) :
BIFROST_ADD_OP_FRCP_FAST_F32;
} else {
op = fp16 ?
(Y ? BIFROST_ADD_OP_FRSQ_FAST_F16_Y :
BIFROST_ADD_OP_FRSQ_FAST_F16_X) :
BIFROST_ADD_OP_FRSQ_FAST_F32;
}
return bi_pack_add_1src(ins, regs, op);
}
static unsigned
bi_pack_add(bi_clause *clause, bi_bundle bundle, struct bi_registers *regs)
{
if (!bundle.add)
return BIFROST_ADD_NOP;
switch (bundle.add->type) {
case BI_ADD:
return bi_pack_add_addmin(bundle.add, regs);
case BI_ATEST:
return bi_pack_add_atest(clause, bundle.add, regs);
case BI_BRANCH:
case BI_CMP:
return BIFROST_ADD_NOP;
case BI_BLEND:
return bi_pack_add_blend(bundle.add, regs);
case BI_BITWISE:
case BI_CONVERT:
case BI_DISCARD:
case BI_FREXP:
case BI_ISUB:
case BI_LOAD:
return BIFROST_ADD_NOP;
case BI_LOAD_ATTR:
return bi_pack_add_ld_attr(clause, bundle.add, regs);
case BI_LOAD_UNIFORM:
return bi_pack_add_ld_ubo(clause, bundle.add, regs);
case BI_LOAD_VAR:
return bi_pack_add_ld_vary(clause, bundle.add, regs);
case BI_LOAD_VAR_ADDRESS:
return bi_pack_add_ld_var_addr(clause, bundle.add, regs);
case BI_MINMAX:
return bi_pack_add_addmin(bundle.add, regs);
case BI_MOV:
case BI_SHIFT:
case BI_STORE:
return BIFROST_ADD_NOP;
case BI_STORE_VAR:
return bi_pack_add_st_vary(clause, bundle.add, regs);
case BI_SPECIAL:
return bi_pack_add_special(bundle.add, regs);
case BI_SWIZZLE:
case BI_TEX:
case BI_ROUND:
return BIFROST_ADD_NOP;
default:
unreachable("Cannot encode class as ADD");
}
}
struct bi_packed_bundle {
uint64_t lo;
uint64_t hi;
};
static struct bi_packed_bundle
bi_pack_bundle(bi_clause *clause, bi_bundle bundle, bi_bundle prev, bool first_bundle)
{
struct bi_registers regs = bi_assign_ports(bundle, prev);
bi_assign_uniform_constant(clause, &regs, bundle);
regs.first_instruction = first_bundle;
uint64_t reg = bi_pack_registers(regs);
uint64_t fma = bi_pack_fma(clause, bundle, &regs);
uint64_t add = bi_pack_add(clause, bundle, &regs);
struct bi_packed_bundle packed = {
.lo = reg | (fma << 35) | ((add & 0b111111) << 58),
.hi = add >> 6
};
return packed;
}
/* Packs the next two constants as a dedicated constant quadword at the end of
* the clause, returning the number packed. */
static unsigned
bi_pack_constants(bi_context *ctx, bi_clause *clause,
unsigned index,
struct util_dynarray *emission)
{
/* After these two, are we done? Determines tag */
bool done = clause->constant_count <= (index + 2);
bool only = clause->constant_count <= (index + 1);
/* TODO: Pos */
assert(index == 0 && clause->bundle_count == 1);
struct bifrost_fmt_constant quad = {
.pos = 0, /* TODO */
.tag = done ? BIFROST_FMTC_FINAL : BIFROST_FMTC_CONSTANTS,
.imm_1 = clause->constants[index + 0] >> 4,
.imm_2 = only ? 0 : clause->constants[index + 1] >> 4
};
/* XXX: On G71, Connor observed that the difference of the top 4 bits
* of the second constant with the first must be less than 8, otherwise
* we have to swap them. I am not able to reproduce this on G52,
* further investigation needed. Possibly an errata. XXX */
util_dynarray_append(emission, struct bifrost_fmt_constant, quad);
return 2;
}
static void
bi_pack_clause(bi_context *ctx, bi_clause *clause, bi_clause *next,
struct util_dynarray *emission)
{
struct bi_packed_bundle ins_1 = bi_pack_bundle(clause, clause->bundles[0], clause->bundles[0], true);
assert(clause->bundle_count == 1);
/* Used to decide if we elide writes */
bool is_fragment = ctx->stage == MESA_SHADER_FRAGMENT;
/* State for packing constants throughout */
unsigned constant_index = 0;
struct bifrost_fmt1 quad_1 = {
.tag = clause->constant_count ? BIFROST_FMT1_CONSTANTS : BIFROST_FMT1_FINAL,
.header = bi_pack_header(clause, next, is_fragment),
.ins_1 = ins_1.lo,
.ins_2 = ins_1.hi & ((1 << 11) - 1),
.ins_0 = (ins_1.hi >> 11) & 0b111,
};
util_dynarray_append(emission, struct bifrost_fmt1, quad_1);
/* Pack the remaining constants */
while (constant_index < clause->constant_count) {
constant_index += bi_pack_constants(ctx, clause,
constant_index, emission);
}
}
static bi_clause *
bi_next_clause(bi_context *ctx, pan_block *block, bi_clause *clause)
{
/* Try the next clause in this block */
if (clause->link.next != &((bi_block *) block)->clauses)
return list_first_entry(&(clause->link), bi_clause, link);
/* Try the next block, or the one after that if it's empty, etc .*/
pan_block *next_block = pan_next_block(block);
bi_foreach_block_from(ctx, next_block, block) {
bi_block *blk = (bi_block *) block;
if (!list_is_empty(&blk->clauses))
return list_first_entry(&(blk->clauses), bi_clause, link);
}
return NULL;
}
void
bi_pack(bi_context *ctx, struct util_dynarray *emission)
{
util_dynarray_init(emission, NULL);
bi_foreach_block(ctx, _block) {
bi_block *block = (bi_block *) _block;
bi_foreach_clause_in_block(block, clause) {
bi_clause *next = bi_next_clause(ctx, _block, clause);
bi_pack_clause(ctx, clause, next, emission);
}
}
}
Reference in New Issue View Git Blame Copy Permalink