mirrors
/
mesa
mirror of https://gitlab.freedesktop.org/mesa/mesa
1
0
Fork 0
Code Issues 9 Packages Projects Releases Wiki Activity
mesa/src/intel/compiler/brw_fs_generator.cpp

1527 lines
54 KiB
C++
Raw Permalink Blame History

/*
* Copyright © 2010 Intel Corporation
*
* 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.
*/
/** @file brw_fs_generator.cpp
*
* This file supports generating code from the FS LIR to the actual
* native instructions.
*/
#include "brw_eu.h"
#include "brw_disasm_info.h"
#include "brw_fs.h"
#include "brw_cfg.h"
#include "dev/intel_debug.h"
#include "util/mesa-sha1.h"
#include "util/half_float.h"
static uint32_t
brw_math_function(enum opcode op)
{
switch (op) {
case SHADER_OPCODE_RCP:
return BRW_MATH_FUNCTION_INV;
case SHADER_OPCODE_RSQ:
return BRW_MATH_FUNCTION_RSQ;
case SHADER_OPCODE_SQRT:
return BRW_MATH_FUNCTION_SQRT;
case SHADER_OPCODE_EXP2:
return BRW_MATH_FUNCTION_EXP;
case SHADER_OPCODE_LOG2:
return BRW_MATH_FUNCTION_LOG;
case SHADER_OPCODE_POW:
return BRW_MATH_FUNCTION_POW;
case SHADER_OPCODE_SIN:
return BRW_MATH_FUNCTION_SIN;
case SHADER_OPCODE_COS:
return BRW_MATH_FUNCTION_COS;
case SHADER_OPCODE_INT_QUOTIENT:
return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT;
case SHADER_OPCODE_INT_REMAINDER:
return BRW_MATH_FUNCTION_INT_DIV_REMAINDER;
default:
unreachable("not reached: unknown math function");
}
}
static struct brw_reg
brw_reg_from_fs_reg(const struct intel_device_info *devinfo, fs_inst *inst,
fs_reg *reg)
{
struct brw_reg brw_reg;
switch (reg->file) {
case ARF:
case FIXED_GRF:
case IMM:
assert(reg->offset == 0);
brw_reg = reg->as_brw_reg();
break;
case BAD_FILE:
/* Probably unused. */
brw_reg = brw_null_reg();
break;
case VGRF:
case ATTR:
case UNIFORM:
unreachable("not reached");
}
return brw_reg;
}
fs_generator::fs_generator(const struct brw_compiler *compiler,
const struct brw_compile_params *params,
struct brw_stage_prog_data *prog_data,
gl_shader_stage stage)
: compiler(compiler), params(params),
devinfo(compiler->devinfo),
prog_data(prog_data), dispatch_width(0),
debug_flag(false),
shader_name(NULL), stage(stage), mem_ctx(params->mem_ctx)
{
p = rzalloc(mem_ctx, struct brw_codegen);
brw_init_codegen(&compiler->isa, p, mem_ctx);
}
fs_generator::~fs_generator()
{
}
class ip_record : public exec_node {
public:
DECLARE_RALLOC_CXX_OPERATORS(ip_record)
ip_record(int ip)
{
this->ip = ip;
}
int ip;
};
bool
fs_generator::patch_halt_jumps()
{
if (this->discard_halt_patches.is_empty())
return false;
int scale = brw_jump_scale(p->devinfo);
/* There is a somewhat strange undocumented requirement of using
* HALT, according to the simulator. If some channel has HALTed to
* a particular UIP, then by the end of the program, every channel
* must have HALTed to that UIP. Furthermore, the tracking is a
* stack, so you can't do the final halt of a UIP after starting
* halting to a new UIP.
*
* Symptoms of not emitting this instruction on actual hardware
* included GPU hangs and sparkly rendering on the piglit discard
* tests.
*/
brw_inst *last_halt = brw_HALT(p);
brw_inst_set_uip(p->devinfo, last_halt, 1 * scale);
brw_inst_set_jip(p->devinfo, last_halt, 1 * scale);
int ip = p->nr_insn;
foreach_in_list(ip_record, patch_ip, &discard_halt_patches) {
brw_inst *patch = &p->store[patch_ip->ip];
assert(brw_inst_opcode(p->isa, patch) == BRW_OPCODE_HALT);
/* HALT takes a half-instruction distance from the pre-incremented IP. */
brw_inst_set_uip(p->devinfo, patch, (ip - patch_ip->ip) * scale);
}
this->discard_halt_patches.make_empty();
return true;
}
void
fs_generator::generate_send(fs_inst *inst,
struct brw_reg dst,
struct brw_reg desc,
struct brw_reg ex_desc,
struct brw_reg payload,
struct brw_reg payload2)
{
const unsigned rlen = inst->dst.is_null() ? 0 : inst->size_written / REG_SIZE;
uint32_t desc_imm = inst->desc |
brw_message_desc(devinfo, inst->mlen, rlen, inst->header_size);
uint32_t ex_desc_imm = inst->ex_desc |
brw_message_ex_desc(devinfo, inst->ex_mlen);
if (ex_desc.file != BRW_IMMEDIATE_VALUE || ex_desc.ud || ex_desc_imm ||
inst->send_ex_desc_scratch) {
/* If we have any sort of extended descriptor, then we need SENDS. This
* also covers the dual-payload case because ex_mlen goes in ex_desc.
*/
brw_send_indirect_split_message(p, inst->sfid, dst, payload, payload2,
desc, desc_imm, ex_desc, ex_desc_imm,
inst->send_ex_desc_scratch,
inst->send_ex_bso, inst->eot);
if (inst->check_tdr)
brw_inst_set_opcode(p->isa, brw_last_inst,
devinfo->ver >= 12 ? BRW_OPCODE_SENDC : BRW_OPCODE_SENDSC);
} else {
brw_send_indirect_message(p, inst->sfid, dst, payload, desc, desc_imm,
inst->eot);
if (inst->check_tdr)
brw_inst_set_opcode(p->isa, brw_last_inst, BRW_OPCODE_SENDC);
}
}
void
fs_generator::generate_mov_indirect(fs_inst *inst,
struct brw_reg dst,
struct brw_reg reg,
struct brw_reg indirect_byte_offset)
{
assert(indirect_byte_offset.type == BRW_TYPE_UD);
assert(indirect_byte_offset.file == BRW_GENERAL_REGISTER_FILE);
assert(!reg.abs && !reg.negate);
/* Gen12.5 adds the following region restriction:
*
* "Vx1 and VxH indirect addressing for Float, Half-Float, Double-Float
* and Quad-Word data must not be used."
*
* We require the source and destination types to match so stomp to an
* unsigned integer type.
*/
assert(reg.type == dst.type);
reg.type = dst.type =
brw_type_with_size(BRW_TYPE_UD, brw_type_size_bits(reg.type));
unsigned imm_byte_offset = reg.nr * REG_SIZE + reg.subnr;
if (indirect_byte_offset.file == BRW_IMMEDIATE_VALUE) {
imm_byte_offset += indirect_byte_offset.ud;
reg.nr = imm_byte_offset / REG_SIZE;
reg.subnr = imm_byte_offset % REG_SIZE;
if (brw_type_size_bytes(reg.type) > 4 && !devinfo->has_64bit_int) {
brw_MOV(p, subscript(dst, BRW_TYPE_D, 0),
subscript(reg, BRW_TYPE_D, 0));
brw_set_default_swsb(p, tgl_swsb_null());
brw_MOV(p, subscript(dst, BRW_TYPE_D, 1),
subscript(reg, BRW_TYPE_D, 1));
} else {
brw_MOV(p, dst, reg);
}
} else {
/* We use VxH indirect addressing, clobbering a0.0 through a0.7. */
struct brw_reg addr = vec8(brw_address_reg(0));
/* Whether we can use destination dependency control without running the
* risk of a hang if an instruction gets shot down.
*/
const bool use_dep_ctrl = !inst->predicate &&
inst->exec_size == dispatch_width;
brw_inst *insn;
/* The destination stride of an instruction (in bytes) must be greater
* than or equal to the size of the rest of the instruction. Since the
* address register is of type UW, we can't use a D-type instruction.
* In order to get around this, re retype to UW and use a stride.
*/
indirect_byte_offset =
retype(spread(indirect_byte_offset, 2), BRW_TYPE_UW);
/* There are a number of reasons why we don't use the base offset here.
* One reason is that the field is only 9 bits which means we can only
* use it to access the first 16 GRFs. Also, from the Haswell PRM
* section "Register Region Restrictions":
*
* "The lower bits of the AddressImmediate must not overflow to
* change the register address. The lower 5 bits of Address
* Immediate when added to lower 5 bits of address register gives
* the sub-register offset. The upper bits of Address Immediate
* when added to upper bits of address register gives the register
* address. Any overflow from sub-register offset is dropped."
*
* Since the indirect may cause us to cross a register boundary, this
* makes the base offset almost useless. We could try and do something
* clever where we use a actual base offset if base_offset % 32 == 0 but
* that would mean we were generating different code depending on the
* base offset. Instead, for the sake of consistency, we'll just do the
* add ourselves. This restriction is only listed in the Haswell PRM
* but empirical testing indicates that it applies on all older
* generations and is lifted on Broadwell.
*
* In the end, while base_offset is nice to look at in the generated
* code, using it saves us 0 instructions and would require quite a bit
* of case-by-case work. It's just not worth it.
*
* Due to a hardware bug some platforms (particularly Gfx11+) seem to
* require the address components of all channels to be valid whether or
* not they're active, which causes issues if we use VxH addressing
* under non-uniform control-flow. We can easily work around that by
* initializing the whole address register with a pipelined NoMask MOV
* instruction.
*/
insn = brw_MOV(p, addr, brw_imm_uw(imm_byte_offset));
brw_inst_set_mask_control(devinfo, insn, BRW_MASK_DISABLE);
brw_inst_set_pred_control(devinfo, insn, BRW_PREDICATE_NONE);
if (devinfo->ver >= 12)
brw_set_default_swsb(p, tgl_swsb_null());
else
brw_inst_set_no_dd_clear(devinfo, insn, use_dep_ctrl);
insn = brw_ADD(p, addr, indirect_byte_offset, brw_imm_uw(imm_byte_offset));
if (devinfo->ver >= 12)
brw_set_default_swsb(p, tgl_swsb_regdist(1));
else
brw_inst_set_no_dd_check(devinfo, insn, use_dep_ctrl);
if (brw_type_size_bytes(reg.type) > 4 &&
(intel_device_info_is_9lp(devinfo) || !devinfo->has_64bit_int)) {
/* From the Cherryview PRM Vol 7. "Register Region Restrictions":
*
* "When source or destination datatype is 64b or operation is
* integer DWord multiply, indirect addressing must not be used."
*
* We may also not support Q/UQ types.
*
* To work around both of these, we do two integer MOVs instead
* of one 64-bit MOV. Because no double value should ever cross
* a register boundary, it's safe to use the immediate offset in
* the indirect here to handle adding 4 bytes to the offset and
* avoid the extra ADD to the register file.
*/
brw_MOV(p, subscript(dst, BRW_TYPE_D, 0),
retype(brw_VxH_indirect(0, 0), BRW_TYPE_D));
brw_set_default_swsb(p, tgl_swsb_null());
brw_MOV(p, subscript(dst, BRW_TYPE_D, 1),
retype(brw_VxH_indirect(0, 4), BRW_TYPE_D));
} else {
struct brw_reg ind_src = brw_VxH_indirect(0, 0);
brw_MOV(p, dst, retype(ind_src, reg.type));
}
}
}
void
fs_generator::generate_shuffle(fs_inst *inst,
struct brw_reg dst,
struct brw_reg src,
struct brw_reg idx)
{
assert(src.file == BRW_GENERAL_REGISTER_FILE);
assert(!src.abs && !src.negate);
/* Ivy bridge has some strange behavior that makes this a real pain to
* implement for 64-bit values so we just don't bother.
*/
assert(devinfo->has_64bit_float || brw_type_size_bytes(src.type) <= 4);
/* Gen12.5 adds the following region restriction:
*
* "Vx1 and VxH indirect addressing for Float, Half-Float, Double-Float
* and Quad-Word data must not be used."
*
* We require the source and destination types to match so stomp to an
* unsigned integer type.
*/
assert(src.type == dst.type);
src.type = dst.type =
brw_type_with_size(BRW_TYPE_UD, brw_type_size_bits(src.type));
/* Because we're using the address register, we're limited to 16-wide
* by the address register file and 8-wide for 64-bit types. We could try
* and make this instruction splittable higher up in the compiler but that
* gets weird because it reads all of the channels regardless of execution
* size. It's easier just to split it here.
*/
const unsigned lower_width =
element_sz(src) > 4 || element_sz(dst) > 4 ? 8 :
MIN2(16, inst->exec_size);
brw_set_default_exec_size(p, cvt(lower_width) - 1);
for (unsigned group = 0; group < inst->exec_size; group += lower_width) {
brw_set_default_group(p, group);
if ((src.vstride == 0 && src.hstride == 0) ||
idx.file == BRW_IMMEDIATE_VALUE) {
/* Trivial, the source is already uniform or the index is a constant.
* We will typically not get here if the optimizer is doing its job,
* but asserting would be mean.
*/
const unsigned i = idx.file == BRW_IMMEDIATE_VALUE ? idx.ud : 0;
struct brw_reg group_src = stride(suboffset(src, i), 0, 1, 0);
struct brw_reg group_dst = suboffset(dst, group << (dst.hstride - 1));
brw_MOV(p, group_dst, group_src);
} else {
/* We use VxH indirect addressing, clobbering a0.0 through a0.7. */
struct brw_reg addr = vec8(brw_address_reg(0));
struct brw_reg group_idx = suboffset(idx, group);
if (lower_width == 8 && group_idx.width == BRW_WIDTH_16) {
/* Things get grumpy if the register is too wide. */
group_idx.width--;
group_idx.vstride--;
}
assert(brw_type_size_bytes(group_idx.type) <= 4);
if (brw_type_size_bytes(group_idx.type) == 4) {
/* The destination stride of an instruction (in bytes) must be
* greater than or equal to the size of the rest of the
* instruction. Since the address register is of type UW, we
* can't use a D-type instruction. In order to get around this,
* re retype to UW and use a stride.
*/
group_idx = retype(spread(group_idx, 2), BRW_TYPE_W);
}
uint32_t src_start_offset = src.nr * REG_SIZE + src.subnr;
/* From the Haswell PRM:
*
* "When a sequence of NoDDChk and NoDDClr are used, the last
* instruction that completes the scoreboard clear must have a
* non-zero execution mask. This means, if any kind of predication
* can change the execution mask or channel enable of the last
* instruction, the optimization must be avoided. This is to
* avoid instructions being shot down the pipeline when no writes
* are required."
*
* Whenever predication is enabled or the instructions being emitted
* aren't the full width, it's possible that it will be run with zero
* channels enabled so we can't use dependency control without
* running the risk of a hang if an instruction gets shot down.
*/
const bool use_dep_ctrl = !inst->predicate &&
lower_width == dispatch_width;
brw_inst *insn;
/* Due to a hardware bug some platforms (particularly Gfx11+) seem
* to require the address components of all channels to be valid
* whether or not they're active, which causes issues if we use VxH
* addressing under non-uniform control-flow. We can easily work
* around that by initializing the whole address register with a
* pipelined NoMask MOV instruction.
*/
insn = brw_MOV(p, addr, brw_imm_uw(src_start_offset));
brw_inst_set_mask_control(devinfo, insn, BRW_MASK_DISABLE);
brw_inst_set_pred_control(devinfo, insn, BRW_PREDICATE_NONE);
if (devinfo->ver >= 12)
brw_set_default_swsb(p, tgl_swsb_null());
else
brw_inst_set_no_dd_clear(devinfo, insn, use_dep_ctrl);
/* Take into account the component size and horizontal stride. */
assert(src.vstride == src.hstride + src.width);
insn = brw_SHL(p, addr, group_idx,
brw_imm_uw(util_logbase2(brw_type_size_bytes(src.type)) +
src.hstride - 1));
if (devinfo->ver >= 12)
brw_set_default_swsb(p, tgl_swsb_regdist(1));
else
brw_inst_set_no_dd_check(devinfo, insn, use_dep_ctrl);
/* Add on the register start offset */
brw_ADD(p, addr, addr, brw_imm_uw(src_start_offset));
brw_MOV(p, suboffset(dst, group << (dst.hstride - 1)),
retype(brw_VxH_indirect(0, 0), src.type));
}
brw_set_default_swsb(p, tgl_swsb_null());
}
}
void
fs_generator::generate_quad_swizzle(const fs_inst *inst,
struct brw_reg dst, struct brw_reg src,
unsigned swiz)
{
/* Requires a quad. */
assert(inst->exec_size >= 4);
if (src.file == BRW_IMMEDIATE_VALUE ||
has_scalar_region(src)) {
/* The value is uniform across all channels */
brw_MOV(p, dst, src);
} else if (devinfo->ver < 11 && brw_type_size_bytes(src.type) == 4) {
/* This only works on 8-wide 32-bit values */
assert(inst->exec_size == 8);
assert(src.hstride == BRW_HORIZONTAL_STRIDE_1);
assert(src.vstride == src.width + 1);
brw_set_default_access_mode(p, BRW_ALIGN_16);
struct brw_reg swiz_src = stride(src, 4, 4, 1);
swiz_src.swizzle = swiz;
brw_MOV(p, dst, swiz_src);
} else {
assert(src.hstride == BRW_HORIZONTAL_STRIDE_1);
assert(src.vstride == src.width + 1);
const struct brw_reg src_0 = suboffset(src, BRW_GET_SWZ(swiz, 0));
switch (swiz) {
case BRW_SWIZZLE_XXXX:
case BRW_SWIZZLE_YYYY:
case BRW_SWIZZLE_ZZZZ:
case BRW_SWIZZLE_WWWW:
brw_MOV(p, dst, stride(src_0, 4, 4, 0));
break;
case BRW_SWIZZLE_XXZZ:
case BRW_SWIZZLE_YYWW:
brw_MOV(p, dst, stride(src_0, 2, 2, 0));
break;
case BRW_SWIZZLE_XYXY:
case BRW_SWIZZLE_ZWZW:
assert(inst->exec_size == 4);
brw_MOV(p, dst, stride(src_0, 0, 2, 1));
break;
default:
assert(inst->force_writemask_all);
brw_set_default_exec_size(p, cvt(inst->exec_size / 4) - 1);
for (unsigned c = 0; c < 4; c++) {
brw_inst *insn = brw_MOV(
p, stride(suboffset(dst, c),
4 * inst->dst.stride, 1, 4 * inst->dst.stride),
stride(suboffset(src, BRW_GET_SWZ(swiz, c)), 4, 1, 0));
if (devinfo->ver < 12) {
brw_inst_set_no_dd_clear(devinfo, insn, c < 3);
brw_inst_set_no_dd_check(devinfo, insn, c > 0);
}
brw_set_default_swsb(p, tgl_swsb_null());
}
break;
}
}
}
void
fs_generator::generate_barrier(fs_inst *, struct brw_reg src)
{
brw_barrier(p, src);
if (devinfo->ver >= 12) {
brw_set_default_swsb(p, tgl_swsb_null());
brw_SYNC(p, TGL_SYNC_BAR);
} else {
brw_WAIT(p);
}
}
/* For OPCODE_DDX and OPCODE_DDY, per channel of output we've got input
* looking like:
*
* arg0: ss0.tl ss0.tr ss0.bl ss0.br ss1.tl ss1.tr ss1.bl ss1.br
*
* Ideally, we want to produce:
*
* DDX DDY
* dst: (ss0.tr - ss0.tl) (ss0.tl - ss0.bl)
* (ss0.tr - ss0.tl) (ss0.tr - ss0.br)
* (ss0.br - ss0.bl) (ss0.tl - ss0.bl)
* (ss0.br - ss0.bl) (ss0.tr - ss0.br)
* (ss1.tr - ss1.tl) (ss1.tl - ss1.bl)
* (ss1.tr - ss1.tl) (ss1.tr - ss1.br)
* (ss1.br - ss1.bl) (ss1.tl - ss1.bl)
* (ss1.br - ss1.bl) (ss1.tr - ss1.br)
*
* and add another set of two more subspans if in 16-pixel dispatch mode.
*
* For DDX, it ends up being easy: width = 2, horiz=0 gets us the same result
* for each pair, and vertstride = 2 jumps us 2 elements after processing a
* pair. But the ideal approximation may impose a huge performance cost on
* sample_d. On at least Haswell, sample_d instruction does some
* optimizations if the same LOD is used for all pixels in the subspan.
*
* For DDY, we need to use ALIGN16 mode since it's capable of doing the
* appropriate swizzling.
*/
void
fs_generator::generate_ddx(const fs_inst *inst,
struct brw_reg dst, struct brw_reg src)
{
unsigned vstride, width;
if (inst->opcode == FS_OPCODE_DDX_FINE) {
/* produce accurate derivatives */
vstride = BRW_VERTICAL_STRIDE_2;
width = BRW_WIDTH_2;
} else {
/* replicate the derivative at the top-left pixel to other pixels */
vstride = BRW_VERTICAL_STRIDE_4;
width = BRW_WIDTH_4;
}
struct brw_reg src0 = byte_offset(src, brw_type_size_bytes(src.type));;
struct brw_reg src1 = src;
src0.vstride = vstride;
src0.width = width;
src0.hstride = BRW_HORIZONTAL_STRIDE_0;
src1.vstride = vstride;
src1.width = width;
src1.hstride = BRW_HORIZONTAL_STRIDE_0;
brw_ADD(p, dst, src0, negate(src1));
}
/* The negate_value boolean is used to negate the derivative computation for
* FBOs, since they place the origin at the upper left instead of the lower
* left.
*/
void
fs_generator::generate_ddy(const fs_inst *inst,
struct brw_reg dst, struct brw_reg src)
{
const uint32_t type_size = brw_type_size_bytes(src.type);
if (inst->opcode == FS_OPCODE_DDY_FINE) {
/* produce accurate derivatives.
*
* From the Broadwell PRM, Volume 7 (3D-Media-GPGPU)
* "Register Region Restrictions", Section "1. Special Restrictions":
*
* "In Align16 mode, the channel selects and channel enables apply to
* a pair of half-floats, because these parameters are defined for
* DWord elements ONLY. This is applicable when both source and
* destination are half-floats."
*
* So for half-float operations we use the Gfx11+ Align1 path. CHV
* inherits its FP16 hardware from SKL, so it is not affected.
*/
if (devinfo->ver >= 11) {
src = stride(src, 0, 2, 1);
brw_push_insn_state(p);
brw_set_default_exec_size(p, BRW_EXECUTE_4);
for (uint32_t g = 0; g < inst->exec_size; g += 4) {
brw_set_default_group(p, inst->group + g);
brw_ADD(p, byte_offset(dst, g * type_size),
negate(byte_offset(src, g * type_size)),
byte_offset(src, (g + 2) * type_size));
brw_set_default_swsb(p, tgl_swsb_null());
}
brw_pop_insn_state(p);
} else {
struct brw_reg src0 = stride(src, 4, 4, 1);
struct brw_reg src1 = stride(src, 4, 4, 1);
src0.swizzle = BRW_SWIZZLE_XYXY;
src1.swizzle = BRW_SWIZZLE_ZWZW;
brw_push_insn_state(p);
brw_set_default_access_mode(p, BRW_ALIGN_16);
brw_ADD(p, dst, negate(src0), src1);
brw_pop_insn_state(p);
}
} else {
/* replicate the derivative at the top-left pixel to other pixels */
struct brw_reg src0 = byte_offset(stride(src, 4, 4, 0), 0 * type_size);
struct brw_reg src1 = byte_offset(stride(src, 4, 4, 0), 2 * type_size);
brw_ADD(p, dst, negate(src0), src1);
}
}
void
fs_generator::generate_halt(fs_inst *)
{
/* This HALT will be patched up at FB write time to point UIP at the end of
* the program, and at brw_uip_jip() JIP will be set to the end of the
* current block (or the program).
*/
this->discard_halt_patches.push_tail(new(mem_ctx) ip_record(p->nr_insn));
brw_HALT(p);
}
/* The A32 messages take a buffer base address in header.5:[31:0] (See
* MH1_A32_PSM for typed messages or MH_A32_GO for byte/dword scattered
* and OWord block messages in the SKL PRM Vol. 2d for more details.)
* Unfortunately, there are a number of subtle differences:
*
* For the block read/write messages:
*
* - We always stomp header.2 to fill in the actual scratch address (in
* units of OWORDs) so we don't care what's in there.
*
* - They rely on per-thread scratch space value in header.3[3:0] to do
* bounds checking so that needs to be valid. The upper bits of
* header.3 are ignored, though, so we can copy all of g0.3.
*
* - They ignore header.5[9:0] and assumes the address is 1KB aligned.
*
*
* For the byte/dword scattered read/write messages:
*
* - We want header.2 to be zero because that gets added to the per-channel
* offset in the non-header portion of the message.
*
* - Contrary to what the docs claim, they don't do any bounds checking so
* the value of header.3[3:0] doesn't matter.
*
* - They consider all of header.5 for the base address and header.5[9:0]
* are not ignored. This means that we can't copy g0.5 verbatim because
* g0.5[9:0] contains the FFTID on most platforms. Instead, we have to
* use an AND to mask off the bottom 10 bits.
*
*
* For block messages, just copying g0 gives a valid header because all the
* garbage gets ignored except for header.2 which we stomp as part of message
* setup. For byte/dword scattered messages, we can just zero out the header
* and copy over the bits we need from g0.5. This opcode, however, tries to
* satisfy the requirements of both by starting with 0 and filling out the
* information required by either set of opcodes.
*/
void
fs_generator::generate_scratch_header(fs_inst *inst, struct brw_reg dst)
{
assert(inst->exec_size == 8 && inst->force_writemask_all);
assert(dst.file == BRW_GENERAL_REGISTER_FILE);
dst.type = BRW_TYPE_UD;
brw_inst *insn = brw_MOV(p, dst, brw_imm_ud(0));
if (devinfo->ver >= 12)
brw_set_default_swsb(p, tgl_swsb_null());
else
brw_inst_set_no_dd_clear(p->devinfo, insn, true);
/* Copy the per-thread scratch space size from g0.3[3:0] */
brw_set_default_exec_size(p, BRW_EXECUTE_1);
insn = brw_AND(p, suboffset(dst, 3),
retype(brw_vec1_grf(0, 3), BRW_TYPE_UD),
brw_imm_ud(INTEL_MASK(3, 0)));
if (devinfo->ver < 12) {
brw_inst_set_no_dd_clear(p->devinfo, insn, true);
brw_inst_set_no_dd_check(p->devinfo, insn, true);
}
/* Copy the scratch base address from g0.5[31:10] */
insn = brw_AND(p, suboffset(dst, 5),
retype(brw_vec1_grf(0, 5), BRW_TYPE_UD),
brw_imm_ud(INTEL_MASK(31, 10)));
if (devinfo->ver < 12)
brw_inst_set_no_dd_check(p->devinfo, insn, true);
}
void
fs_generator::enable_debug(const char *shader_name)
{
debug_flag = true;
this->shader_name = shader_name;
}
static gfx12_systolic_depth
translate_systolic_depth(unsigned d)
{
/* Could also return (ffs(d) - 1) & 3. */
switch (d) {
case 2: return BRW_SYSTOLIC_DEPTH_2;
case 4: return BRW_SYSTOLIC_DEPTH_4;
case 8: return BRW_SYSTOLIC_DEPTH_8;
case 16: return BRW_SYSTOLIC_DEPTH_16;
default: unreachable("Invalid systolic depth.");
}
}
int
fs_generator::generate_code(const cfg_t *cfg, int dispatch_width,
struct shader_stats shader_stats,
const brw::performance &perf,
struct brw_compile_stats *stats,
unsigned max_polygons)
{
/* align to 64 byte boundary. */
brw_realign(p, 64);
this->dispatch_width = dispatch_width;
int start_offset = p->next_insn_offset;
int loop_count = 0, send_count = 0, nop_count = 0, sync_nop_count = 0;
bool is_accum_used = false;
struct disasm_info *disasm_info = disasm_initialize(p->isa, cfg);
foreach_block_and_inst (block, fs_inst, inst, cfg) {
if (inst->opcode == SHADER_OPCODE_UNDEF)
continue;
struct brw_reg src[4], dst;
unsigned int last_insn_offset = p->next_insn_offset;
bool multiple_instructions_emitted = false;
tgl_swsb swsb = inst->sched;
/* From the Broadwell PRM, Volume 7, "3D-Media-GPGPU", in the
* "Register Region Restrictions" section: for BDW, SKL:
*
* "A POW/FDIV operation must not be followed by an instruction
* that requires two destination registers."
*
* The documentation is often lacking annotations for Atom parts,
* and empirically this affects CHV as well.
*/
if (devinfo->ver <= 9 &&
p->nr_insn > 1 &&
brw_inst_opcode(p->isa, brw_last_inst) == BRW_OPCODE_MATH &&
brw_inst_math_function(devinfo, brw_last_inst) == BRW_MATH_FUNCTION_POW &&
inst->dst.component_size(inst->exec_size) > REG_SIZE) {
brw_NOP(p);
last_insn_offset = p->next_insn_offset;
/* In order to avoid spurious instruction count differences when the
* instruction schedule changes, keep track of the number of inserted
* NOPs.
*/
nop_count++;
}
/* Wa_14010017096:
*
* Clear accumulator register before end of thread.
*/
if (inst->eot && is_accum_used &&
intel_needs_workaround(devinfo, 14010017096)) {
brw_set_default_exec_size(p, BRW_EXECUTE_16);
brw_set_default_group(p, 0);
brw_set_default_mask_control(p, BRW_MASK_DISABLE);
brw_set_default_predicate_control(p, BRW_PREDICATE_NONE);
brw_set_default_flag_reg(p, 0, 0);
brw_set_default_swsb(p, tgl_swsb_src_dep(swsb));
brw_MOV(p, brw_acc_reg(8), brw_imm_f(0.0f));
last_insn_offset = p->next_insn_offset;
swsb = tgl_swsb_dst_dep(swsb, 1);
}
if (!is_accum_used && !inst->eot) {
is_accum_used = inst->writes_accumulator_implicitly(devinfo) ||
inst->dst.is_accumulator();
}
/* Wa_14013672992:
*
* Always use @1 SWSB for EOT.
*/
if (inst->eot && intel_needs_workaround(devinfo, 14013672992)) {
if (tgl_swsb_src_dep(swsb).mode) {
brw_set_default_exec_size(p, BRW_EXECUTE_1);
brw_set_default_mask_control(p, BRW_MASK_DISABLE);
brw_set_default_predicate_control(p, BRW_PREDICATE_NONE);
brw_set_default_flag_reg(p, 0, 0);
brw_set_default_swsb(p, tgl_swsb_src_dep(swsb));
brw_SYNC(p, TGL_SYNC_NOP);
last_insn_offset = p->next_insn_offset;
}
swsb = tgl_swsb_dst_dep(swsb, 1);
}
if (unlikely(debug_flag))
disasm_annotate(disasm_info, inst, p->next_insn_offset);
if (devinfo->ver >= 20 && inst->group % 8 != 0) {
assert(inst->force_writemask_all);
assert(!inst->predicate && !inst->conditional_mod);
assert(!inst->writes_accumulator_implicitly(devinfo) &&
!inst->reads_accumulator_implicitly());
assert(inst->opcode != SHADER_OPCODE_SEL_EXEC);
brw_set_default_group(p, 0);
} else {
brw_set_default_group(p, inst->group);
}
for (unsigned int i = 0; i < inst->sources; i++) {
src[i] = brw_reg_from_fs_reg(devinfo, inst, &inst->src[i]);
/* The accumulator result appears to get used for the
* conditional modifier generation. When negating a UD
* value, there is a 33rd bit generated for the sign in the
* accumulator value, so now you can't check, for example,
* equality with a 32-bit value. See piglit fs-op-neg-uvec4.
*/
assert(!inst->conditional_mod ||
inst->src[i].type != BRW_TYPE_UD ||
!inst->src[i].negate);
}
dst = brw_reg_from_fs_reg(devinfo, inst, &inst->dst);
brw_set_default_access_mode(p, BRW_ALIGN_1);
brw_set_default_predicate_control(p, inst->predicate);
brw_set_default_predicate_inverse(p, inst->predicate_inverse);
/* On gfx7 and above, hardware automatically adds the group onto the
* flag subregister number.
*/
const unsigned flag_subreg = inst->flag_subreg;
brw_set_default_flag_reg(p, flag_subreg / 2, flag_subreg % 2);
brw_set_default_saturate(p, inst->saturate);
brw_set_default_mask_control(p, inst->force_writemask_all);
if (devinfo->ver >= 20 && inst->writes_accumulator) {
assert(inst->dst.is_accumulator() ||
inst->opcode == BRW_OPCODE_ADDC ||
inst->opcode == BRW_OPCODE_MACH ||
inst->opcode == BRW_OPCODE_SUBB);
} else {
brw_set_default_acc_write_control(p, inst->writes_accumulator);
}
brw_set_default_swsb(p, swsb);
unsigned exec_size = inst->exec_size;
brw_set_default_exec_size(p, cvt(exec_size) - 1);
assert(inst->force_writemask_all || inst->exec_size >= 4);
assert(inst->force_writemask_all || inst->group % inst->exec_size == 0);
assert(inst->mlen <= BRW_MAX_MSG_LENGTH * reg_unit(devinfo));
switch (inst->opcode) {
case BRW_OPCODE_NOP:
brw_NOP(p);
break;
case BRW_OPCODE_SYNC:
assert(src[0].file == BRW_IMMEDIATE_VALUE);
brw_SYNC(p, tgl_sync_function(src[0].ud));
if (tgl_sync_function(src[0].ud) == TGL_SYNC_NOP)
++sync_nop_count;
break;
case BRW_OPCODE_MOV:
brw_MOV(p, dst, src[0]);
break;
case BRW_OPCODE_ADD:
brw_ADD(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_MUL:
brw_MUL(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_AVG:
brw_AVG(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_MACH:
brw_MACH(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_DP4A:
assert(devinfo->ver >= 12);
brw_DP4A(p, dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_LINE:
brw_LINE(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_DPAS:
assert(devinfo->verx10 >= 125);
brw_DPAS(p, translate_systolic_depth(inst->sdepth), inst->rcount,
dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_MAD:
if (devinfo->ver < 10)
brw_set_default_access_mode(p, BRW_ALIGN_16);
brw_MAD(p, dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_LRP:
assert(devinfo->ver <= 10);
if (devinfo->ver < 10)
brw_set_default_access_mode(p, BRW_ALIGN_16);
brw_LRP(p, dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_ADD3:
assert(devinfo->verx10 >= 125);
brw_ADD3(p, dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_FRC:
brw_FRC(p, dst, src[0]);
break;
case BRW_OPCODE_RNDD:
brw_RNDD(p, dst, src[0]);
break;
case BRW_OPCODE_RNDE:
brw_RNDE(p, dst, src[0]);
break;
case BRW_OPCODE_RNDZ:
brw_RNDZ(p, dst, src[0]);
break;
case BRW_OPCODE_AND:
brw_AND(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_OR:
brw_OR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_XOR:
brw_XOR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_NOT:
brw_NOT(p, dst, src[0]);
break;
case BRW_OPCODE_ASR:
brw_ASR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_SHR:
brw_SHR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_SHL:
brw_SHL(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_ROL:
assert(devinfo->ver >= 11);
assert(src[0].type == dst.type);
brw_ROL(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_ROR:
assert(devinfo->ver >= 11);
assert(src[0].type == dst.type);
brw_ROR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_CMP:
brw_CMP(p, dst, inst->conditional_mod, src[0], src[1]);
break;
case BRW_OPCODE_CMPN:
brw_CMPN(p, dst, inst->conditional_mod, src[0], src[1]);
break;
case BRW_OPCODE_SEL:
brw_SEL(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_CSEL:
if (devinfo->ver < 10)
brw_set_default_access_mode(p, BRW_ALIGN_16);
brw_CSEL(p, dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_BFREV:
brw_BFREV(p, retype(dst, BRW_TYPE_UD), retype(src[0], BRW_TYPE_UD));
break;
case BRW_OPCODE_FBH:
brw_FBH(p, retype(dst, src[0].type), src[0]);
break;
case BRW_OPCODE_FBL:
brw_FBL(p, retype(dst, BRW_TYPE_UD), retype(src[0], BRW_TYPE_UD));
break;
case BRW_OPCODE_LZD:
brw_LZD(p, dst, src[0]);
break;
case BRW_OPCODE_CBIT:
brw_CBIT(p, retype(dst, BRW_TYPE_UD), retype(src[0], BRW_TYPE_UD));
break;
case BRW_OPCODE_ADDC:
brw_ADDC(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_SUBB:
brw_SUBB(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_MAC:
brw_MAC(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_BFE:
if (devinfo->ver < 10)
brw_set_default_access_mode(p, BRW_ALIGN_16);
brw_BFE(p, dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_BFI1:
brw_BFI1(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_BFI2:
if (devinfo->ver < 10)
brw_set_default_access_mode(p, BRW_ALIGN_16);
brw_BFI2(p, dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_IF:
brw_IF(p, brw_get_default_exec_size(p));
break;
case BRW_OPCODE_ELSE:
brw_ELSE(p);
break;
case BRW_OPCODE_ENDIF:
brw_ENDIF(p);
break;
case BRW_OPCODE_DO:
brw_DO(p, brw_get_default_exec_size(p));
break;
case BRW_OPCODE_BREAK:
brw_BREAK(p);
break;
case BRW_OPCODE_CONTINUE:
brw_CONT(p);
break;
case BRW_OPCODE_WHILE:
brw_WHILE(p);
loop_count++;
break;
case SHADER_OPCODE_RCP:
case SHADER_OPCODE_RSQ:
case SHADER_OPCODE_SQRT:
case SHADER_OPCODE_EXP2:
case SHADER_OPCODE_LOG2:
case SHADER_OPCODE_SIN:
case SHADER_OPCODE_COS:
assert(inst->conditional_mod == BRW_CONDITIONAL_NONE);
assert(inst->mlen == 0);
gfx6_math(p, dst, brw_math_function(inst->opcode),
src[0], brw_null_reg());
break;
case SHADER_OPCODE_INT_QUOTIENT:
case SHADER_OPCODE_INT_REMAINDER:
case SHADER_OPCODE_POW:
assert(devinfo->verx10 < 125);
assert(inst->conditional_mod == BRW_CONDITIONAL_NONE);
assert(inst->mlen == 0);
assert(inst->opcode == SHADER_OPCODE_POW || inst->exec_size == 8);
gfx6_math(p, dst, brw_math_function(inst->opcode), src[0], src[1]);
break;
case BRW_OPCODE_PLN:
/* PLN reads:
* / in SIMD16 \
* -----------------------------------
* | src1+0 | src1+1 | src1+2 | src1+3 |
* |-----------------------------------|
* |(x0, x1)|(y0, y1)|(x2, x3)|(y2, y3)|
* -----------------------------------
*/
brw_PLN(p, dst, src[0], src[1]);
break;
case FS_OPCODE_PIXEL_X:
assert(src[0].type == BRW_TYPE_UW);
assert(src[1].type == BRW_TYPE_UW);
src[0].subnr = 0 * brw_type_size_bytes(src[0].type);
if (src[1].file == BRW_IMMEDIATE_VALUE) {
assert(src[1].ud == 0);
brw_MOV(p, dst, stride(src[0], 8, 4, 1));
} else {
/* Coarse pixel case */
brw_ADD(p, dst, stride(src[0], 8, 4, 1), src[1]);
}
break;
case FS_OPCODE_PIXEL_Y:
assert(src[0].type == BRW_TYPE_UW);
assert(src[1].type == BRW_TYPE_UW);
src[0].subnr = 4 * brw_type_size_bytes(src[0].type);
if (src[1].file == BRW_IMMEDIATE_VALUE) {
assert(src[1].ud == 0);
brw_MOV(p, dst, stride(src[0], 8, 4, 1));
} else {
/* Coarse pixel case */
brw_ADD(p, dst, stride(src[0], 8, 4, 1), src[1]);
}
break;
case SHADER_OPCODE_SEND:
generate_send(inst, dst, src[0], src[1], src[2],
inst->ex_mlen > 0 ? src[3] : brw_null_reg());
send_count++;
break;
case FS_OPCODE_DDX_COARSE:
case FS_OPCODE_DDX_FINE:
generate_ddx(inst, dst, src[0]);
break;
case FS_OPCODE_DDY_COARSE:
case FS_OPCODE_DDY_FINE:
generate_ddy(inst, dst, src[0]);
break;
case SHADER_OPCODE_SCRATCH_HEADER:
generate_scratch_header(inst, dst);
break;
case SHADER_OPCODE_MOV_INDIRECT:
generate_mov_indirect(inst, dst, src[0], src[1]);
break;
case SHADER_OPCODE_MOV_RELOC_IMM:
assert(src[0].file == BRW_IMMEDIATE_VALUE);
assert(src[1].file == BRW_IMMEDIATE_VALUE);
brw_MOV_reloc_imm(p, dst, dst.type, src[0].ud, src[1].ud);
break;
case BRW_OPCODE_HALT:
generate_halt(inst);
break;
case SHADER_OPCODE_INTERLOCK:
case SHADER_OPCODE_MEMORY_FENCE: {
assert(src[1].file == BRW_IMMEDIATE_VALUE);
assert(src[2].file == BRW_IMMEDIATE_VALUE);
const enum opcode send_op = inst->opcode == SHADER_OPCODE_INTERLOCK ?
BRW_OPCODE_SENDC : BRW_OPCODE_SEND;
brw_memory_fence(p, dst, src[0], send_op,
brw_message_target(inst->sfid),
inst->desc,
/* commit_enable */ src[1].ud,
/* bti */ src[2].ud);
send_count++;
break;
}
case FS_OPCODE_SCHEDULING_FENCE:
if (inst->sources == 0 && swsb.regdist == 0 &&
swsb.mode == TGL_SBID_NULL) {
if (unlikely(debug_flag))
disasm_info->use_tail = true;
break;
}
if (devinfo->ver >= 12) {
/* Use the available SWSB information to stall. A single SYNC is
* sufficient since if there were multiple dependencies, the
* scoreboard algorithm already injected other SYNCs before this
* instruction.
*/
brw_SYNC(p, TGL_SYNC_NOP);
} else {
for (unsigned i = 0; i < inst->sources; i++) {
/* Emit a MOV to force a stall until the instruction producing the
* registers finishes.
*/
brw_MOV(p, retype(brw_null_reg(), BRW_TYPE_UW),
retype(src[i], BRW_TYPE_UW));
}
if (inst->sources > 1)
multiple_instructions_emitted = true;
}
break;
case SHADER_OPCODE_FIND_LIVE_CHANNEL:
case SHADER_OPCODE_FIND_LAST_LIVE_CHANNEL:
case SHADER_OPCODE_LOAD_LIVE_CHANNELS:
unreachable("Should be lowered by lower_find_live_channel()");
break;
case FS_OPCODE_LOAD_LIVE_CHANNELS: {
assert(inst->force_writemask_all && inst->group == 0);
assert(inst->dst.file == BAD_FILE);
brw_set_default_exec_size(p, BRW_EXECUTE_1);
brw_MOV(p, retype(brw_flag_subreg(inst->flag_subreg), BRW_TYPE_UD),
retype(brw_mask_reg(0), BRW_TYPE_UD));
break;
}
case SHADER_OPCODE_BROADCAST:
assert(inst->force_writemask_all);
brw_broadcast(p, dst, src[0], src[1]);
break;
case SHADER_OPCODE_SHUFFLE:
generate_shuffle(inst, dst, src[0], src[1]);
break;
case SHADER_OPCODE_SEL_EXEC:
assert(inst->force_writemask_all);
assert(devinfo->has_64bit_float || brw_type_size_bytes(dst.type) <= 4);
brw_set_default_mask_control(p, BRW_MASK_DISABLE);
brw_MOV(p, dst, src[1]);
brw_set_default_mask_control(p, BRW_MASK_ENABLE);
brw_set_default_swsb(p, tgl_swsb_null());
brw_MOV(p, dst, src[0]);
break;
case SHADER_OPCODE_QUAD_SWIZZLE:
assert(src[1].file == BRW_IMMEDIATE_VALUE);
assert(src[1].type == BRW_TYPE_UD);
generate_quad_swizzle(inst, dst, src[0], src[1].ud);
break;
case SHADER_OPCODE_CLUSTER_BROADCAST: {
assert((!intel_device_info_is_9lp(devinfo) &&
devinfo->has_64bit_float) || brw_type_size_bytes(src[0].type) <= 4);
assert(!src[0].negate && !src[0].abs);
assert(src[1].file == BRW_IMMEDIATE_VALUE);
assert(src[1].type == BRW_TYPE_UD);
assert(src[2].file == BRW_IMMEDIATE_VALUE);
assert(src[2].type == BRW_TYPE_UD);
const unsigned component = src[1].ud;
const unsigned cluster_size = src[2].ud;
assert(inst->src[0].file != ARF);
unsigned s;
if (inst->src[0].file == FIXED_GRF) {
s = inst->src[0].hstride ? 1 << (inst->src[0].hstride - 1) : 0;
} else {
s = inst->src[0].stride;
}
unsigned vstride = cluster_size * s;
unsigned width = cluster_size;
/* The maximum exec_size is 32, but the maximum width is only 16. */
if (inst->exec_size == width) {
vstride = 0;
width = 1;
}
struct brw_reg strided = stride(suboffset(src[0], component * s),
vstride, width, 0);
brw_MOV(p, dst, strided);
break;
}
case SHADER_OPCODE_HALT_TARGET:
/* This is the place where the final HALT needs to be inserted if
* we've emitted any discards. If not, this will emit no code.
*/
if (!patch_halt_jumps()) {
if (unlikely(debug_flag)) {
disasm_info->use_tail = true;
}
}
break;
case SHADER_OPCODE_BARRIER:
generate_barrier(inst, src[0]);
send_count++;
break;
case SHADER_OPCODE_RND_MODE: {
assert(src[0].file == BRW_IMMEDIATE_VALUE);
/*
* Changes the floating point rounding mode updating the control
* register field defined at cr0.0[5-6] bits.
*/
enum brw_rnd_mode mode =
(enum brw_rnd_mode) (src[0].d << BRW_CR0_RND_MODE_SHIFT);
brw_float_controls_mode(p, mode, BRW_CR0_RND_MODE_MASK);
}
break;
case SHADER_OPCODE_FLOAT_CONTROL_MODE:
assert(src[0].file == BRW_IMMEDIATE_VALUE);
assert(src[1].file == BRW_IMMEDIATE_VALUE);
brw_float_controls_mode(p, src[0].d, src[1].d);
break;
case SHADER_OPCODE_READ_ARCH_REG:
if (devinfo->ver >= 12) {
/* There is a SWSB restriction that requires that any time sr0 is
* accessed both the instruction doing the access and the next one
* have SWSB set to RegDist(1).
*/
if (brw_get_default_swsb(p).mode != TGL_SBID_NULL)
brw_SYNC(p, TGL_SYNC_NOP);
brw_set_default_swsb(p, tgl_swsb_regdist(1));
brw_MOV(p, dst, src[0]);
brw_set_default_swsb(p, tgl_swsb_regdist(1));
brw_AND(p, dst, dst, brw_imm_ud(0xffffffff));
} else {
brw_MOV(p, dst, src[0]);
}
break;
default:
unreachable("Unsupported opcode");
case SHADER_OPCODE_LOAD_PAYLOAD:
unreachable("Should be lowered by lower_load_payload()");
}
if (multiple_instructions_emitted)
continue;
if (inst->no_dd_clear || inst->no_dd_check || inst->conditional_mod) {
assert(p->next_insn_offset == last_insn_offset + 16 ||
!"conditional_mod, no_dd_check, or no_dd_clear set for IR "
"emitting more than 1 instruction");
brw_inst *last = &p->store[last_insn_offset / 16];
if (inst->conditional_mod)
brw_inst_set_cond_modifier(p->devinfo, last, inst->conditional_mod);
if (devinfo->ver < 12) {
brw_inst_set_no_dd_clear(p->devinfo, last, inst->no_dd_clear);
brw_inst_set_no_dd_check(p->devinfo, last, inst->no_dd_check);
}
}
/* When enabled, insert sync NOP after every instruction and make sure
* that current instruction depends on the previous instruction.
*/
if (INTEL_DEBUG(DEBUG_SWSB_STALL) && devinfo->ver >= 12) {
brw_set_default_swsb(p, tgl_swsb_regdist(1));
brw_SYNC(p, TGL_SYNC_NOP);
}
}
brw_set_uip_jip(p, start_offset);
/* end of program sentinel */
disasm_new_inst_group(disasm_info, p->next_insn_offset);
/* `send_count` explicitly does not include spills or fills, as we'd
* like to use it as a metric for intentional memory access or other
* shared function use. Otherwise, subtle changes to scheduling or
* register allocation could cause it to fluctuate wildly - and that
* effect is already counted in spill/fill counts.
*/
send_count -= shader_stats.spill_count;
send_count -= shader_stats.fill_count;
#ifndef NDEBUG
bool validated =
#else
if (unlikely(debug_flag))
#endif
brw_validate_instructions(&compiler->isa, p->store,
start_offset,
p->next_insn_offset,
disasm_info);
int before_size = p->next_insn_offset - start_offset;
brw_compact_instructions(p, start_offset, disasm_info);
int after_size = p->next_insn_offset - start_offset;
bool dump_shader_bin = brw_should_dump_shader_bin();
unsigned char sha1[21];
char sha1buf[41];
if (unlikely(debug_flag || dump_shader_bin)) {
_mesa_sha1_compute(p->store + start_offset / sizeof(brw_inst),
after_size, sha1);
_mesa_sha1_format(sha1buf, sha1);
}
if (unlikely(dump_shader_bin))
brw_dump_shader_bin(p->store, start_offset, p->next_insn_offset,
sha1buf);
if (unlikely(debug_flag)) {
fprintf(stderr, "Native code for %s (src_hash 0x%08x) (sha1 %s)\n"
"SIMD%d shader: %d instructions. %d loops. %u cycles. "
"%d:%d spills:fills, %u sends, "
"scheduled with mode %s. "
"Promoted %u constants. "
"Compacted %d to %d bytes (%.0f%%)\n",
shader_name, params->source_hash, sha1buf,
dispatch_width, before_size / 16,
loop_count, perf.latency,
shader_stats.spill_count,
shader_stats.fill_count,
send_count,
shader_stats.scheduler_mode,
shader_stats.promoted_constants,
before_size, after_size,
100.0f * (before_size - after_size) / before_size);
/* overriding the shader makes disasm_info invalid */
if (!brw_try_override_assembly(p, start_offset, sha1buf)) {
dump_assembly(p->store, start_offset, p->next_insn_offset,
disasm_info, perf.block_latency);
} else {
fprintf(stderr, "Successfully overrode shader with sha1 %s\n\n", sha1buf);
}
}
ralloc_free(disasm_info);
#ifndef NDEBUG
if (!validated && !debug_flag) {
fprintf(stderr,
"Validation failed. Rerun with INTEL_DEBUG=shaders to get more information.\n");
}
#endif
assert(validated);
brw_shader_debug_log(compiler, params->log_data,
"%s SIMD%d shader: %d inst, %d loops, %u cycles, "
"%d:%d spills:fills, %u sends, "
"scheduled with mode %s, "
"Promoted %u constants, "
"compacted %d to %d bytes.\n",
_mesa_shader_stage_to_abbrev(stage),
dispatch_width,
before_size / 16 - nop_count - sync_nop_count,
loop_count, perf.latency,
shader_stats.spill_count,
shader_stats.fill_count,
send_count,
shader_stats.scheduler_mode,
shader_stats.promoted_constants,
before_size, after_size);
if (stats) {
stats->dispatch_width = dispatch_width;
stats->max_polygons = max_polygons;
stats->max_dispatch_width = dispatch_width;
stats->instructions = before_size / 16 - nop_count - sync_nop_count;
stats->sends = send_count;
stats->loops = loop_count;
stats->cycles = perf.latency;
stats->spills = shader_stats.spill_count;
stats->fills = shader_stats.fill_count;
stats->max_live_registers = shader_stats.max_register_pressure;
}
return start_offset;
}
void
fs_generator::add_const_data(void *data, unsigned size)
{
assert(prog_data->const_data_size == 0);
if (size > 0) {
prog_data->const_data_size = size;
prog_data->const_data_offset = brw_append_data(p, data, size, 32);
}
}
void
fs_generator::add_resume_sbt(unsigned num_resume_shaders, uint64_t *sbt)
{
assert(brw_shader_stage_is_bindless(stage));
struct brw_bs_prog_data *bs_prog_data = brw_bs_prog_data(prog_data);
if (num_resume_shaders > 0) {
bs_prog_data->resume_sbt_offset =
brw_append_data(p, sbt, num_resume_shaders * sizeof(uint64_t), 32);
for (unsigned i = 0; i < num_resume_shaders; i++) {
size_t offset = bs_prog_data->resume_sbt_offset + i * sizeof(*sbt);
assert(offset <= UINT32_MAX);
brw_add_reloc(p, BRW_SHADER_RELOC_SHADER_START_OFFSET,
BRW_SHADER_RELOC_TYPE_U32,
(uint32_t)offset, (uint32_t)sbt[i]);
}
}
}
const unsigned *
fs_generator::get_assembly()
{
prog_data->relocs = brw_get_shader_relocs(p, &prog_data->num_relocs);
return brw_get_program(p, &prog_data->program_size);
}
Reference in New Issue View Git Blame Copy Permalink