freedreno: move ir3 to common location

[mesa.git] / src / freedreno / ir3 / ir3_compiler_nir.c

/*
 * Copyright (C) 2015 Rob Clark <robclark@freedesktop.org>
 *
 * 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.
 *
 * Authors:
 *    Rob Clark <robclark@freedesktop.org>
 */

#include <stdarg.h>

#include "util/u_string.h"
#include "util/u_memory.h"
#include "util/u_math.h"

#include "ir3_compiler.h"
#include "ir3_shader.h"
#include "ir3_nir.h"

#include "instr-a3xx.h"
#include "ir3.h"

/* for conditionally setting boolean flag(s): */
#define COND(bool, val) ((bool) ? (val) : 0)

#define DBG(fmt, ...) \
                do { debug_printf("%s:%d: "fmt "\n", \
                                __FUNCTION__, __LINE__, ##__VA_ARGS__); } while (0)

struct ir3_context {
        struct ir3_compiler *compiler;

        struct nir_shader *s;

        struct nir_instr *cur_instr;  /* current instruction, just for debug */

        struct ir3 *ir;
        struct ir3_shader_variant *so;

        struct ir3_block *block;      /* the current block */
        struct ir3_block *in_block;   /* block created for shader inputs */

        nir_function_impl *impl;

        /* For fragment shaders, varyings are not actual shader inputs,
         * instead the hw passes a varying-coord which is used with
         * bary.f.
         *
         * But NIR doesn't know that, it still declares varyings as
         * inputs.  So we do all the input tracking normally and fix
         * things up after compile_instructions()
         *
         * NOTE that frag_vcoord is the hardware position (possibly it
         * is actually an index or tag or some such.. it is *not*
         * values that can be directly used for gl_FragCoord..)
         */
        struct ir3_instruction *frag_vcoord;

        /* for fragment shaders, for gl_FrontFacing and gl_FragCoord: */
        struct ir3_instruction *frag_face, *frag_coord;

        /* For vertex shaders, keep track of the system values sources */
        struct ir3_instruction *vertex_id, *basevertex, *instance_id;

        /* For fragment shaders: */
        struct ir3_instruction *samp_id, *samp_mask_in;

        /* Compute shader inputs: */
        struct ir3_instruction *local_invocation_id, *work_group_id;

        /* mapping from nir_register to defining instruction: */
        struct hash_table *def_ht;

        unsigned num_arrays;

        /* a common pattern for indirect addressing is to request the
         * same address register multiple times.  To avoid generating
         * duplicate instruction sequences (which our backend does not
         * try to clean up, since that should be done as the NIR stage)
         * we cache the address value generated for a given src value:
         *
         * Note that we have to cache these per alignment, since same
         * src used for an array of vec1 cannot be also used for an
         * array of vec4.
         */
        struct hash_table *addr_ht[4];

        /* last dst array, for indirect we need to insert a var-store.
         */
        struct ir3_instruction **last_dst;
        unsigned last_dst_n;

        /* maps nir_block to ir3_block, mostly for the purposes of
         * figuring out the blocks successors
         */
        struct hash_table *block_ht;

        /* on a4xx, bitmask of samplers which need astc+srgb workaround: */
        unsigned astc_srgb;

        unsigned samples;             /* bitmask of x,y sample shifts */

        unsigned max_texture_index;

        /* set if we encounter something we can't handle yet, so we
         * can bail cleanly and fallback to TGSI compiler f/e
         */
        bool error;
};

/* gpu pointer size in units of 32bit registers/slots */
static unsigned pointer_size(struct ir3_context *ctx)
{
        return (ctx->compiler->gpu_id >= 500) ? 2 : 1;
}

static struct ir3_instruction * create_immed(struct ir3_block *block, uint32_t val);
static struct ir3_block * get_block(struct ir3_context *ctx, const nir_block *nblock);


static struct ir3_context *
compile_init(struct ir3_compiler *compiler,
                struct ir3_shader_variant *so)
{
        struct ir3_context *ctx = rzalloc(NULL, struct ir3_context);

        if (compiler->gpu_id >= 400) {
                if (so->type == MESA_SHADER_VERTEX) {
                        ctx->astc_srgb = so->key.vastc_srgb;
                } else if (so->type == MESA_SHADER_FRAGMENT) {
                        ctx->astc_srgb = so->key.fastc_srgb;
                }

        } else {
                if (so->type == MESA_SHADER_VERTEX) {
                        ctx->samples = so->key.vsamples;
                } else if (so->type == MESA_SHADER_FRAGMENT) {
                        ctx->samples = so->key.fsamples;
                }
        }

        ctx->compiler = compiler;
        ctx->so = so;
        ctx->def_ht = _mesa_hash_table_create(ctx,
                        _mesa_hash_pointer, _mesa_key_pointer_equal);
        ctx->block_ht = _mesa_hash_table_create(ctx,
                        _mesa_hash_pointer, _mesa_key_pointer_equal);

        /* TODO: maybe generate some sort of bitmask of what key
         * lowers vs what shader has (ie. no need to lower
         * texture clamp lowering if no texture sample instrs)..
         * although should be done further up the stack to avoid
         * creating duplicate variants..
         */

        if (ir3_key_lowers_nir(&so->key)) {
                nir_shader *s = nir_shader_clone(ctx, so->shader->nir);
                ctx->s = ir3_optimize_nir(so->shader, s, &so->key);
        } else {
                /* fast-path for shader key that lowers nothing in NIR: */
                ctx->s = so->shader->nir;
        }

        /* this needs to be the last pass run, so do this here instead of
         * in ir3_optimize_nir():
         */
        NIR_PASS_V(ctx->s, nir_lower_locals_to_regs);
        NIR_PASS_V(ctx->s, nir_convert_from_ssa, true);

        if (ir3_shader_debug & IR3_DBG_DISASM) {
                printf("dump nir%dv%d: type=%d, k={cts=%u,hp=%u}",
                        so->shader->id, so->id, so->type,
                        so->key.color_two_side, so->key.half_precision);
                nir_print_shader(ctx->s, stdout);
        }

        if (shader_debug_enabled(so->type)) {
                fprintf(stderr, "NIR (final form) for %s shader:\n",
                        _mesa_shader_stage_to_string(so->type));
                nir_print_shader(ctx->s, stderr);
        }

        ir3_nir_scan_driver_consts(ctx->s, &so->const_layout);

        so->num_uniforms = ctx->s->num_uniforms;
        so->num_ubos = ctx->s->info.num_ubos;

        /* Layout of constant registers, each section aligned to vec4.  Note
         * that pointer size (ubo, etc) changes depending on generation.
         *
         *    user consts
         *    UBO addresses
         *    SSBO sizes
         *    if (vertex shader) {
         *        driver params (IR3_DP_*)
         *        if (stream_output.num_outputs > 0)
         *           stream-out addresses
         *    }
         *    immediates
         *
         * Immediates go last mostly because they are inserted in the CP pass
         * after the nir -> ir3 frontend.
         */
        unsigned constoff = align(ctx->s->num_uniforms, 4);
        unsigned ptrsz = pointer_size(ctx);

        memset(&so->constbase, ~0, sizeof(so->constbase));

        if (so->num_ubos > 0) {
                so->constbase.ubo = constoff;
                constoff += align(ctx->s->info.num_ubos * ptrsz, 4) / 4;
        }

        if (so->const_layout.ssbo_size.count > 0) {
                unsigned cnt = so->const_layout.ssbo_size.count;
                so->constbase.ssbo_sizes = constoff;
                constoff += align(cnt, 4) / 4;
        }

        if (so->const_layout.image_dims.count > 0) {
                unsigned cnt = so->const_layout.image_dims.count;
                so->constbase.image_dims = constoff;
                constoff += align(cnt, 4) / 4;
        }

        unsigned num_driver_params = 0;
        if (so->type == MESA_SHADER_VERTEX) {
                num_driver_params = IR3_DP_VS_COUNT;
        } else if (so->type == MESA_SHADER_COMPUTE) {
                num_driver_params = IR3_DP_CS_COUNT;
        }

        so->constbase.driver_param = constoff;
        constoff += align(num_driver_params, 4) / 4;

        if ((so->type == MESA_SHADER_VERTEX) &&
                        (compiler->gpu_id < 500) &&
                        so->shader->stream_output.num_outputs > 0) {
                so->constbase.tfbo = constoff;
                constoff += align(IR3_MAX_SO_BUFFERS * ptrsz, 4) / 4;
        }

        so->constbase.immediate = constoff;

        return ctx;
}

static void
compile_error(struct ir3_context *ctx, const char *format, ...)
{
        struct hash_table *errors = NULL;
        va_list ap;
        va_start(ap, format);
        if (ctx->cur_instr) {
                errors = _mesa_hash_table_create(NULL,
                                _mesa_hash_pointer,
                                _mesa_key_pointer_equal);
                char *msg = ralloc_vasprintf(errors, format, ap);
                _mesa_hash_table_insert(errors, ctx->cur_instr, msg);
        } else {
                _debug_vprintf(format, ap);
        }
        va_end(ap);
        nir_print_shader_annotated(ctx->s, stdout, errors);
        ralloc_free(errors);
        ctx->error = true;
        debug_assert(0);
}

#define compile_assert(ctx, cond) do { \
                if (!(cond)) compile_error((ctx), "failed assert: "#cond"\n"); \
        } while (0)

static void
compile_free(struct ir3_context *ctx)
{
        ralloc_free(ctx);
}

static void
declare_array(struct ir3_context *ctx, nir_register *reg)
{
        struct ir3_array *arr = rzalloc(ctx, struct ir3_array);
        arr->id = ++ctx->num_arrays;
        /* NOTE: sometimes we get non array regs, for example for arrays of
         * length 1.  See fs-const-array-of-struct-of-array.shader_test.  So
         * treat a non-array as if it was an array of length 1.
         *
         * It would be nice if there was a nir pass to convert arrays of
         * length 1 to ssa.
         */
        arr->length = reg->num_components * MAX2(1, reg->num_array_elems);
        compile_assert(ctx, arr->length > 0);
        arr->r = reg;
        list_addtail(&arr->node, &ctx->ir->array_list);
}

static struct ir3_array *
get_array(struct ir3_context *ctx, nir_register *reg)
{
        list_for_each_entry (struct ir3_array, arr, &ctx->ir->array_list, node) {
                if (arr->r == reg)
                        return arr;
        }
        compile_error(ctx, "bogus reg: %s\n", reg->name);
        return NULL;
}

/* relative (indirect) if address!=NULL */
static struct ir3_instruction *
create_array_load(struct ir3_context *ctx, struct ir3_array *arr, int n,
                struct ir3_instruction *address)
{
        struct ir3_block *block = ctx->block;
        struct ir3_instruction *mov;
        struct ir3_register *src;

        mov = ir3_instr_create(block, OPC_MOV);
        mov->cat1.src_type = TYPE_U32;
        mov->cat1.dst_type = TYPE_U32;
        mov->barrier_class = IR3_BARRIER_ARRAY_R;
        mov->barrier_conflict = IR3_BARRIER_ARRAY_W;
        ir3_reg_create(mov, 0, 0);
        src = ir3_reg_create(mov, 0, IR3_REG_ARRAY |
                        COND(address, IR3_REG_RELATIV));
        src->instr = arr->last_write;
        src->size  = arr->length;
        src->array.id = arr->id;
        src->array.offset = n;

        if (address)
                ir3_instr_set_address(mov, address);

        return mov;
}

/* relative (indirect) if address!=NULL */
static void
create_array_store(struct ir3_context *ctx, struct ir3_array *arr, int n,
                struct ir3_instruction *src, struct ir3_instruction *address)
{
        struct ir3_block *block = ctx->block;
        struct ir3_instruction *mov;
        struct ir3_register *dst;

        /* if not relative store, don't create an extra mov, since that
         * ends up being difficult for cp to remove.
         */
        if (!address) {
                dst = src->regs[0];

                src->barrier_class |= IR3_BARRIER_ARRAY_W;
                src->barrier_conflict |= IR3_BARRIER_ARRAY_R | IR3_BARRIER_ARRAY_W;

                dst->flags |= IR3_REG_ARRAY;
                dst->instr = arr->last_write;
                dst->size = arr->length;
                dst->array.id = arr->id;
                dst->array.offset = n;

                arr->last_write = src;

                array_insert(block, block->keeps, src);

                return;
        }

        mov = ir3_instr_create(block, OPC_MOV);
        mov->cat1.src_type = TYPE_U32;
        mov->cat1.dst_type = TYPE_U32;
        mov->barrier_class = IR3_BARRIER_ARRAY_W;
        mov->barrier_conflict = IR3_BARRIER_ARRAY_R | IR3_BARRIER_ARRAY_W;
        dst = ir3_reg_create(mov, 0, IR3_REG_ARRAY |
                        COND(address, IR3_REG_RELATIV));
        dst->instr = arr->last_write;
        dst->size  = arr->length;
        dst->array.id = arr->id;
        dst->array.offset = n;
        ir3_reg_create(mov, 0, IR3_REG_SSA)->instr = src;

        if (address)
                ir3_instr_set_address(mov, address);

        arr->last_write = mov;

        /* the array store may only matter to something in an earlier
         * block (ie. loops), but since arrays are not in SSA, depth
         * pass won't know this.. so keep all array stores:
         */
        array_insert(block, block->keeps, mov);
}

static inline type_t utype_for_size(unsigned bit_size)
{
        switch (bit_size) {
        case 32: return TYPE_U32;
        case 16: return TYPE_U16;
        case  8: return TYPE_U8;
        default: unreachable("bad bitsize"); return ~0;
        }
}

static inline type_t utype_src(nir_src src)
{ return utype_for_size(nir_src_bit_size(src)); }

static inline type_t utype_dst(nir_dest dst)
{ return utype_for_size(nir_dest_bit_size(dst)); }

/* allocate a n element value array (to be populated by caller) and
 * insert in def_ht
 */
static struct ir3_instruction **
get_dst_ssa(struct ir3_context *ctx, nir_ssa_def *dst, unsigned n)
{
        struct ir3_instruction **value =
                ralloc_array(ctx->def_ht, struct ir3_instruction *, n);
        _mesa_hash_table_insert(ctx->def_ht, dst, value);
        return value;
}

static struct ir3_instruction **
get_dst(struct ir3_context *ctx, nir_dest *dst, unsigned n)
{
        struct ir3_instruction **value;

        if (dst->is_ssa) {
                value = get_dst_ssa(ctx, &dst->ssa, n);
        } else {
                value = ralloc_array(ctx, struct ir3_instruction *, n);
        }

        /* NOTE: in non-ssa case, we don't really need to store last_dst
         * but this helps us catch cases where put_dst() call is forgotten
         */
        compile_assert(ctx, !ctx->last_dst);
        ctx->last_dst = value;
        ctx->last_dst_n = n;

        return value;
}

static struct ir3_instruction * get_addr(struct ir3_context *ctx, struct ir3_instruction *src, int align);

static struct ir3_instruction * const *
get_src(struct ir3_context *ctx, nir_src *src)
{
        if (src->is_ssa) {
                struct hash_entry *entry;
                entry = _mesa_hash_table_search(ctx->def_ht, src->ssa);
                compile_assert(ctx, entry);
                return entry->data;
        } else {
                nir_register *reg = src->reg.reg;
                struct ir3_array *arr = get_array(ctx, reg);
                unsigned num_components = arr->r->num_components;
                struct ir3_instruction *addr = NULL;
                struct ir3_instruction **value =
                        ralloc_array(ctx, struct ir3_instruction *, num_components);

                if (src->reg.indirect)
                        addr = get_addr(ctx, get_src(ctx, src->reg.indirect)[0],
                                        reg->num_components);

                for (unsigned i = 0; i < num_components; i++) {
                        unsigned n = src->reg.base_offset * reg->num_components + i;
                        compile_assert(ctx, n < arr->length);
                        value[i] = create_array_load(ctx, arr, n, addr);
                }

                return value;
        }
}

static void
put_dst(struct ir3_context *ctx, nir_dest *dst)
{
        unsigned bit_size = nir_dest_bit_size(*dst);

        if (bit_size < 32) {
                for (unsigned i = 0; i < ctx->last_dst_n; i++) {
                        struct ir3_instruction *dst = ctx->last_dst[i];
                        dst->regs[0]->flags |= IR3_REG_HALF;
                        if (ctx->last_dst[i]->opc == OPC_META_FO)
                                dst->regs[1]->instr->regs[0]->flags |= IR3_REG_HALF;
                }
        }

        if (!dst->is_ssa) {
                nir_register *reg = dst->reg.reg;
                struct ir3_array *arr = get_array(ctx, reg);
                unsigned num_components = ctx->last_dst_n;
                struct ir3_instruction *addr = NULL;

                if (dst->reg.indirect)
                        addr = get_addr(ctx, get_src(ctx, dst->reg.indirect)[0],
                                        reg->num_components);

                for (unsigned i = 0; i < num_components; i++) {
                        unsigned n = dst->reg.base_offset * reg->num_components + i;
                        compile_assert(ctx, n < arr->length);
                        if (!ctx->last_dst[i])
                                continue;
                        create_array_store(ctx, arr, n, ctx->last_dst[i], addr);
                }

                ralloc_free(ctx->last_dst);
        }
        ctx->last_dst = NULL;
        ctx->last_dst_n = 0;
}

static struct ir3_instruction *
create_immed_typed(struct ir3_block *block, uint32_t val, type_t type)
{
        struct ir3_instruction *mov;
        unsigned flags = (type_size(type) < 32) ? IR3_REG_HALF : 0;

        mov = ir3_instr_create(block, OPC_MOV);
        mov->cat1.src_type = type;
        mov->cat1.dst_type = type;
        ir3_reg_create(mov, 0, flags);
        ir3_reg_create(mov, 0, IR3_REG_IMMED)->uim_val = val;

        return mov;
}

static struct ir3_instruction *
create_immed(struct ir3_block *block, uint32_t val)
{
        return create_immed_typed(block, val, TYPE_U32);
}

static struct ir3_instruction *
create_addr(struct ir3_block *block, struct ir3_instruction *src, int align)
{
        struct ir3_instruction *instr, *immed;

        /* TODO in at least some cases, the backend could probably be
         * made clever enough to propagate IR3_REG_HALF..
         */
        instr = ir3_COV(block, src, TYPE_U32, TYPE_S16);
        instr->regs[0]->flags |= IR3_REG_HALF;

        switch(align){
        case 1:
                /* src *= 1: */
                break;
        case 2:
                /* src *= 2     => src <<= 1: */
                immed = create_immed(block, 1);
                immed->regs[0]->flags |= IR3_REG_HALF;

                instr = ir3_SHL_B(block, instr, 0, immed, 0);
                instr->regs[0]->flags |= IR3_REG_HALF;
                instr->regs[1]->flags |= IR3_REG_HALF;
                break;
        case 3:
                /* src *= 3: */
                immed = create_immed(block, 3);
                immed->regs[0]->flags |= IR3_REG_HALF;

                instr = ir3_MULL_U(block, instr, 0, immed, 0);
                instr->regs[0]->flags |= IR3_REG_HALF;
                instr->regs[1]->flags |= IR3_REG_HALF;
                break;
        case 4:
                /* src *= 4 => src <<= 2: */
                immed = create_immed(block, 2);
                immed->regs[0]->flags |= IR3_REG_HALF;

                instr = ir3_SHL_B(block, instr, 0, immed, 0);
                instr->regs[0]->flags |= IR3_REG_HALF;
                instr->regs[1]->flags |= IR3_REG_HALF;
                break;
        default:
                unreachable("bad align");
                return NULL;
        }

        instr = ir3_MOV(block, instr, TYPE_S16);
        instr->regs[0]->num = regid(REG_A0, 0);
        instr->regs[0]->flags |= IR3_REG_HALF;
        instr->regs[1]->flags |= IR3_REG_HALF;

        return instr;
}

/* caches addr values to avoid generating multiple cov/shl/mova
 * sequences for each use of a given NIR level src as address
 */
static struct ir3_instruction *
get_addr(struct ir3_context *ctx, struct ir3_instruction *src, int align)
{
        struct ir3_instruction *addr;
        unsigned idx = align - 1;

        compile_assert(ctx, idx < ARRAY_SIZE(ctx->addr_ht));

        if (!ctx->addr_ht[idx]) {
                ctx->addr_ht[idx] = _mesa_hash_table_create(ctx,
                                _mesa_hash_pointer, _mesa_key_pointer_equal);
        } else {
                struct hash_entry *entry;
                entry = _mesa_hash_table_search(ctx->addr_ht[idx], src);
                if (entry)
                        return entry->data;
        }

        addr = create_addr(ctx->block, src, align);
        _mesa_hash_table_insert(ctx->addr_ht[idx], src, addr);

        return addr;
}

static struct ir3_instruction *
get_predicate(struct ir3_context *ctx, struct ir3_instruction *src)
{
        struct ir3_block *b = ctx->block;
        struct ir3_instruction *cond;

        /* NOTE: only cmps.*.* can write p0.x: */
        cond = ir3_CMPS_S(b, src, 0, create_immed(b, 0), 0);
        cond->cat2.condition = IR3_COND_NE;

        /* condition always goes in predicate register: */
        cond->regs[0]->num = regid(REG_P0, 0);

        return cond;
}

static struct ir3_instruction *
create_uniform(struct ir3_context *ctx, unsigned n)
{
        struct ir3_instruction *mov;

        mov = ir3_instr_create(ctx->block, OPC_MOV);
        /* TODO get types right? */
        mov->cat1.src_type = TYPE_F32;
        mov->cat1.dst_type = TYPE_F32;
        ir3_reg_create(mov, 0, 0);
        ir3_reg_create(mov, n, IR3_REG_CONST);

        return mov;
}

static struct ir3_instruction *
create_uniform_indirect(struct ir3_context *ctx, int n,
                struct ir3_instruction *address)
{
        struct ir3_instruction *mov;

        mov = ir3_instr_create(ctx->block, OPC_MOV);
        mov->cat1.src_type = TYPE_U32;
        mov->cat1.dst_type = TYPE_U32;
        ir3_reg_create(mov, 0, 0);
        ir3_reg_create(mov, 0, IR3_REG_CONST | IR3_REG_RELATIV)->array.offset = n;

        ir3_instr_set_address(mov, address);

        return mov;
}

static struct ir3_instruction *
create_collect(struct ir3_context *ctx, struct ir3_instruction *const *arr,
                unsigned arrsz)
{
        struct ir3_block *block = ctx->block;
        struct ir3_instruction *collect;

        if (arrsz == 0)
                return NULL;

        unsigned flags = arr[0]->regs[0]->flags & IR3_REG_HALF;

        collect = ir3_instr_create2(block, OPC_META_FI, 1 + arrsz);
        ir3_reg_create(collect, 0, flags);     /* dst */
        for (unsigned i = 0; i < arrsz; i++) {
                struct ir3_instruction *elem = arr[i];

                /* Since arrays are pre-colored in RA, we can't assume that
                 * things will end up in the right place.  (Ie. if a collect
                 * joins elements from two different arrays.)  So insert an
                 * extra mov.
                 *
                 * We could possibly skip this if all the collected elements
                 * are contiguous elements in a single array.. not sure how
                 * likely that is to happen.
                 *
                 * Fixes a problem with glamor shaders, that in effect do
                 * something like:
                 *
                 *   if (foo)
                 *     texcoord = ..
                 *   else
                 *     texcoord = ..
                 *   color = texture2D(tex, texcoord);
                 *
                 * In this case, texcoord will end up as nir registers (which
                 * translate to ir3 array's of length 1.  And we can't assume
                 * the two (or more) arrays will get allocated in consecutive
                 * scalar registers.
                 *
                 */
                if (elem->regs[0]->flags & IR3_REG_ARRAY) {
                        type_t type = (flags & IR3_REG_HALF) ? TYPE_U16 : TYPE_U32;
                        elem = ir3_MOV(block, elem, type);
                }

                compile_assert(ctx, (elem->regs[0]->flags & IR3_REG_HALF) == flags);
                ir3_reg_create(collect, 0, IR3_REG_SSA | flags)->instr = elem;
        }

        return collect;
}

static struct ir3_instruction *
create_indirect_load(struct ir3_context *ctx, unsigned arrsz, int n,
                struct ir3_instruction *address, struct ir3_instruction *collect)
{
        struct ir3_block *block = ctx->block;
        struct ir3_instruction *mov;
        struct ir3_register *src;

        mov = ir3_instr_create(block, OPC_MOV);
        mov->cat1.src_type = TYPE_U32;
        mov->cat1.dst_type = TYPE_U32;
        ir3_reg_create(mov, 0, 0);
        src = ir3_reg_create(mov, 0, IR3_REG_SSA | IR3_REG_RELATIV);
        src->instr = collect;
        src->size  = arrsz;
        src->array.offset = n;

        ir3_instr_set_address(mov, address);

        return mov;
}

static struct ir3_instruction *
create_input_compmask(struct ir3_context *ctx, unsigned n, unsigned compmask)
{
        struct ir3_instruction *in;

        in = ir3_instr_create(ctx->in_block, OPC_META_INPUT);
        in->inout.block = ctx->in_block;
        ir3_reg_create(in, n, 0);

        in->regs[0]->wrmask = compmask;

        return in;
}

static struct ir3_instruction *
create_input(struct ir3_context *ctx, unsigned n)
{
        return create_input_compmask(ctx, n, 0x1);
}

static struct ir3_instruction *
create_frag_input(struct ir3_context *ctx, bool use_ldlv)
{
        struct ir3_block *block = ctx->block;
        struct ir3_instruction *instr;
        /* actual inloc is assigned and fixed up later: */
        struct ir3_instruction *inloc = create_immed(block, 0);

        if (use_ldlv) {
                instr = ir3_LDLV(block, inloc, 0, create_immed(block, 1), 0);
                instr->cat6.type = TYPE_U32;
                instr->cat6.iim_val = 1;
        } else {
                instr = ir3_BARY_F(block, inloc, 0, ctx->frag_vcoord, 0);
                instr->regs[2]->wrmask = 0x3;
        }

        return instr;
}

static struct ir3_instruction *
create_driver_param(struct ir3_context *ctx, enum ir3_driver_param dp)
{
        /* first four vec4 sysval's reserved for UBOs: */
        /* NOTE: dp is in scalar, but there can be >4 dp components: */
        unsigned n = ctx->so->constbase.driver_param;
        unsigned r = regid(n + dp / 4, dp % 4);
        return create_uniform(ctx, r);
}

/* helper for instructions that produce multiple consecutive scalar
 * outputs which need to have a split/fanout meta instruction inserted
 */
static void
split_dest(struct ir3_block *block, struct ir3_instruction **dst,
                struct ir3_instruction *src, unsigned base, unsigned n)
{
        struct ir3_instruction *prev = NULL;

        if ((n == 1) && (src->regs[0]->wrmask == 0x1)) {
                dst[0] = src;
                return;
        }

        for (int i = 0, j = 0; i < n; i++) {
                struct ir3_instruction *split = ir3_instr_create(block, OPC_META_FO);
                ir3_reg_create(split, 0, IR3_REG_SSA);
                ir3_reg_create(split, 0, IR3_REG_SSA)->instr = src;
                split->fo.off = i + base;

                if (prev) {
                        split->cp.left = prev;
                        split->cp.left_cnt++;
                        prev->cp.right = split;
                        prev->cp.right_cnt++;
                }
                prev = split;

                if (src->regs[0]->wrmask & (1 << (i + base)))
                        dst[j++] = split;
        }
}

/*
 * Adreno uses uint rather than having dedicated bool type,
 * which (potentially) requires some conversion, in particular
 * when using output of an bool instr to int input, or visa
 * versa.
 *
 *         | Adreno  |  NIR  |
 *  -------+---------+-------+-
 *   true  |    1    |  ~0   |
 *   false |    0    |   0   |
 *
 * To convert from an adreno bool (uint) to nir, use:
 *
 *    absneg.s dst, (neg)src
 *
 * To convert back in the other direction:
 *
 *    absneg.s dst, (abs)arc
 *
 * The CP step can clean up the absneg.s that cancel each other
 * out, and with a slight bit of extra cleverness (to recognize
 * the instructions which produce either a 0 or 1) can eliminate
 * the absneg.s's completely when an instruction that wants
 * 0/1 consumes the result.  For example, when a nir 'bcsel'
 * consumes the result of 'feq'.  So we should be able to get by
 * without a boolean resolve step, and without incuring any
 * extra penalty in instruction count.
 */

/* NIR bool -> native (adreno): */
static struct ir3_instruction *
ir3_b2n(struct ir3_block *block, struct ir3_instruction *instr)
{
        return ir3_ABSNEG_S(block, instr, IR3_REG_SABS);
}

/* native (adreno) -> NIR bool: */
static struct ir3_instruction *
ir3_n2b(struct ir3_block *block, struct ir3_instruction *instr)
{
        return ir3_ABSNEG_S(block, instr, IR3_REG_SNEG);
}

/*
 * alu/sfu instructions:
 */

static struct ir3_instruction *
create_cov(struct ir3_context *ctx, struct ir3_instruction *src,
                unsigned src_bitsize, nir_op op)
{
        type_t src_type, dst_type;

        switch (op) {
        case nir_op_f2f32:
        case nir_op_f2f16_rtne:
        case nir_op_f2f16_rtz:
        case nir_op_f2f16:
        case nir_op_f2i32:
        case nir_op_f2i16:
        case nir_op_f2i8:
        case nir_op_f2u32:
        case nir_op_f2u16:
        case nir_op_f2u8:
                switch (src_bitsize) {
                case 32:
                        src_type = TYPE_F32;
                        break;
                case 16:
                        src_type = TYPE_F16;
                        break;
                default:
                        compile_error(ctx, "invalid src bit size: %u", src_bitsize);
                }
                break;

        case nir_op_i2f32:
        case nir_op_i2f16:
        case nir_op_i2i32:
        case nir_op_i2i16:
        case nir_op_i2i8:
                switch (src_bitsize) {
                case 32:
                        src_type = TYPE_S32;
                        break;
                case 16:
                        src_type = TYPE_S16;
                        break;
                case 8:
                        src_type = TYPE_S8;
                        break;
                default:
                        compile_error(ctx, "invalid src bit size: %u", src_bitsize);
                }
                break;

        case nir_op_u2f32:
        case nir_op_u2f16:
        case nir_op_u2u32:
        case nir_op_u2u16:
        case nir_op_u2u8:
                switch (src_bitsize) {
                case 32:
                        src_type = TYPE_U32;
                        break;
                case 16:
                        src_type = TYPE_U16;
                        break;
                case 8:
                        src_type = TYPE_U8;
                        break;
                default:
                        compile_error(ctx, "invalid src bit size: %u", src_bitsize);
                }
                break;

        default:
                compile_error(ctx, "invalid conversion op: %u", op);
        }

        switch (op) {
        case nir_op_f2f32:
        case nir_op_i2f32:
        case nir_op_u2f32:
                dst_type = TYPE_F32;
                break;

        case nir_op_f2f16_rtne:
        case nir_op_f2f16_rtz:
        case nir_op_f2f16:
                /* TODO how to handle rounding mode? */
        case nir_op_i2f16:
        case nir_op_u2f16:
                dst_type = TYPE_F16;
                break;

        case nir_op_f2i32:
        case nir_op_i2i32:
                dst_type = TYPE_S32;
                break;

        case nir_op_f2i16:
        case nir_op_i2i16:
                dst_type = TYPE_S16;
                break;

        case nir_op_f2i8:
        case nir_op_i2i8:
                dst_type = TYPE_S8;
                break;

        case nir_op_f2u32:
        case nir_op_u2u32:
                dst_type = TYPE_U32;
                break;

        case nir_op_f2u16:
        case nir_op_u2u16:
                dst_type = TYPE_U16;
                break;

        case nir_op_f2u8:
        case nir_op_u2u8:
                dst_type = TYPE_U8;
                break;

        default:
                compile_error(ctx, "invalid conversion op: %u", op);
        }

        return ir3_COV(ctx->block, src, src_type, dst_type);
}

static void
emit_alu(struct ir3_context *ctx, nir_alu_instr *alu)
{
        const nir_op_info *info = &nir_op_infos[alu->op];
        struct ir3_instruction **dst, *src[info->num_inputs];
        unsigned bs[info->num_inputs];     /* bit size */
        struct ir3_block *b = ctx->block;
        unsigned dst_sz, wrmask;

        if (alu->dest.dest.is_ssa) {
                dst_sz = alu->dest.dest.ssa.num_components;
                wrmask = (1 << dst_sz) - 1;
        } else {
                dst_sz = alu->dest.dest.reg.reg->num_components;
                wrmask = alu->dest.write_mask;
        }

        dst = get_dst(ctx, &alu->dest.dest, dst_sz);

        /* Vectors are special in that they have non-scalarized writemasks,
         * and just take the first swizzle channel for each argument in
         * order into each writemask channel.
         */
        if ((alu->op == nir_op_vec2) ||
                        (alu->op == nir_op_vec3) ||
                        (alu->op == nir_op_vec4)) {

                for (int i = 0; i < info->num_inputs; i++) {
                        nir_alu_src *asrc = &alu->src[i];

                        compile_assert(ctx, !asrc->abs);
                        compile_assert(ctx, !asrc->negate);

                        src[i] = get_src(ctx, &asrc->src)[asrc->swizzle[0]];
                        if (!src[i])
                                src[i] = create_immed(ctx->block, 0);
                        dst[i] = ir3_MOV(b, src[i], TYPE_U32);
                }

                put_dst(ctx, &alu->dest.dest);
                return;
        }

        /* We also get mov's with more than one component for mov's so
         * handle those specially:
         */
        if ((alu->op == nir_op_imov) || (alu->op == nir_op_fmov)) {
                type_t type = (alu->op == nir_op_imov) ? TYPE_U32 : TYPE_F32;
                nir_alu_src *asrc = &alu->src[0];
                struct ir3_instruction *const *src0 = get_src(ctx, &asrc->src);

                for (unsigned i = 0; i < dst_sz; i++) {
                        if (wrmask & (1 << i)) {
                                dst[i] = ir3_MOV(b, src0[asrc->swizzle[i]], type);
                        } else {
                                dst[i] = NULL;
                        }
                }

                put_dst(ctx, &alu->dest.dest);
                return;
        }

        /* General case: We can just grab the one used channel per src. */
        for (int i = 0; i < info->num_inputs; i++) {
                unsigned chan = ffs(alu->dest.write_mask) - 1;
                nir_alu_src *asrc = &alu->src[i];

                compile_assert(ctx, !asrc->abs);
                compile_assert(ctx, !asrc->negate);

                src[i] = get_src(ctx, &asrc->src)[asrc->swizzle[chan]];
                bs[i] = nir_src_bit_size(asrc->src);

                compile_assert(ctx, src[i]);
        }

        switch (alu->op) {
        case nir_op_f2f32:
        case nir_op_f2f16_rtne:
        case nir_op_f2f16_rtz:
        case nir_op_f2f16:
        case nir_op_f2i32:
        case nir_op_f2i16:
        case nir_op_f2i8:
        case nir_op_f2u32:
        case nir_op_f2u16:
        case nir_op_f2u8:
        case nir_op_i2f32:
        case nir_op_i2f16:
        case nir_op_i2i32:
        case nir_op_i2i16:
        case nir_op_i2i8:
        case nir_op_u2f32:
        case nir_op_u2f16:
        case nir_op_u2u32:
        case nir_op_u2u16:
        case nir_op_u2u8:
                dst[0] = create_cov(ctx, src[0], bs[0], alu->op);
                break;
        case nir_op_f2b:
                dst[0] = ir3_CMPS_F(b, src[0], 0, create_immed(b, fui(0.0)), 0);
                dst[0]->cat2.condition = IR3_COND_NE;
                dst[0] = ir3_n2b(b, dst[0]);
                break;
        case nir_op_b2f:
                dst[0] = ir3_COV(b, ir3_b2n(b, src[0]), TYPE_U32, TYPE_F32);
                break;
        case nir_op_b2i:
                dst[0] = ir3_b2n(b, src[0]);
                break;
        case nir_op_i2b:
                dst[0] = ir3_CMPS_S(b, src[0], 0, create_immed(b, 0), 0);
                dst[0]->cat2.condition = IR3_COND_NE;
                dst[0] = ir3_n2b(b, dst[0]);
                break;

        case nir_op_fneg:
                dst[0] = ir3_ABSNEG_F(b, src[0], IR3_REG_FNEG);
                break;
        case nir_op_fabs:
                dst[0] = ir3_ABSNEG_F(b, src[0], IR3_REG_FABS);
                break;
        case nir_op_fmax:
                dst[0] = ir3_MAX_F(b, src[0], 0, src[1], 0);
                break;
        case nir_op_fmin:
                dst[0] = ir3_MIN_F(b, src[0], 0, src[1], 0);
                break;
        case nir_op_fsat:
                /* if there is just a single use of the src, and it supports
                 * (sat) bit, we can just fold the (sat) flag back to the
                 * src instruction and create a mov.  This is easier for cp
                 * to eliminate.
                 *
                 * TODO probably opc_cat==4 is ok too
                 */
                if (alu->src[0].src.is_ssa &&
                                (list_length(&alu->src[0].src.ssa->uses) == 1) &&
                                ((opc_cat(src[0]->opc) == 2) || (opc_cat(src[0]->opc) == 3))) {
                        src[0]->flags |= IR3_INSTR_SAT;
                        dst[0] = ir3_MOV(b, src[0], TYPE_U32);
                } else {
                        /* otherwise generate a max.f that saturates.. blob does
                         * similar (generating a cat2 mov using max.f)
                         */
                        dst[0] = ir3_MAX_F(b, src[0], 0, src[0], 0);
                        dst[0]->flags |= IR3_INSTR_SAT;
                }
                break;
        case nir_op_fmul:
                dst[0] = ir3_MUL_F(b, src[0], 0, src[1], 0);
                break;
        case nir_op_fadd:
                dst[0] = ir3_ADD_F(b, src[0], 0, src[1], 0);
                break;
        case nir_op_fsub:
                dst[0] = ir3_ADD_F(b, src[0], 0, src[1], IR3_REG_FNEG);
                break;
        case nir_op_ffma:
                dst[0] = ir3_MAD_F32(b, src[0], 0, src[1], 0, src[2], 0);
                break;
        case nir_op_fddx:
                dst[0] = ir3_DSX(b, src[0], 0);
                dst[0]->cat5.type = TYPE_F32;
                break;
        case nir_op_fddy:
                dst[0] = ir3_DSY(b, src[0], 0);
                dst[0]->cat5.type = TYPE_F32;
                break;
                break;
        case nir_op_flt:
                dst[0] = ir3_CMPS_F(b, src[0], 0, src[1], 0);
                dst[0]->cat2.condition = IR3_COND_LT;
                dst[0] = ir3_n2b(b, dst[0]);
                break;
        case nir_op_fge:
                dst[0] = ir3_CMPS_F(b, src[0], 0, src[1], 0);
                dst[0]->cat2.condition = IR3_COND_GE;
                dst[0] = ir3_n2b(b, dst[0]);
                break;
        case nir_op_feq:
                dst[0] = ir3_CMPS_F(b, src[0], 0, src[1], 0);
                dst[0]->cat2.condition = IR3_COND_EQ;
                dst[0] = ir3_n2b(b, dst[0]);
                break;
        case nir_op_fne:
                dst[0] = ir3_CMPS_F(b, src[0], 0, src[1], 0);
                dst[0]->cat2.condition = IR3_COND_NE;
                dst[0] = ir3_n2b(b, dst[0]);
                break;
        case nir_op_fceil:
                dst[0] = ir3_CEIL_F(b, src[0], 0);
                break;
        case nir_op_ffloor:
                dst[0] = ir3_FLOOR_F(b, src[0], 0);
                break;
        case nir_op_ftrunc:
                dst[0] = ir3_TRUNC_F(b, src[0], 0);
                break;
        case nir_op_fround_even:
                dst[0] = ir3_RNDNE_F(b, src[0], 0);
                break;
        case nir_op_fsign:
                dst[0] = ir3_SIGN_F(b, src[0], 0);
                break;

        case nir_op_fsin:
                dst[0] = ir3_SIN(b, src[0], 0);
                break;
        case nir_op_fcos:
                dst[0] = ir3_COS(b, src[0], 0);
                break;
        case nir_op_frsq:
                dst[0] = ir3_RSQ(b, src[0], 0);
                break;
        case nir_op_frcp:
                dst[0] = ir3_RCP(b, src[0], 0);
                break;
        case nir_op_flog2:
                dst[0] = ir3_LOG2(b, src[0], 0);
                break;
        case nir_op_fexp2:
                dst[0] = ir3_EXP2(b, src[0], 0);
                break;
        case nir_op_fsqrt:
                dst[0] = ir3_SQRT(b, src[0], 0);
                break;

        case nir_op_iabs:
                dst[0] = ir3_ABSNEG_S(b, src[0], IR3_REG_SABS);
                break;
        case nir_op_iadd:
                dst[0] = ir3_ADD_U(b, src[0], 0, src[1], 0);
                break;
        case nir_op_iand:
                dst[0] = ir3_AND_B(b, src[0], 0, src[1], 0);
                break;
        case nir_op_imax:
                dst[0] = ir3_MAX_S(b, src[0], 0, src[1], 0);
                break;
        case nir_op_umax:
                dst[0] = ir3_MAX_U(b, src[0], 0, src[1], 0);
                break;
        case nir_op_imin:
                dst[0] = ir3_MIN_S(b, src[0], 0, src[1], 0);
                break;
        case nir_op_umin:
                dst[0] = ir3_MIN_U(b, src[0], 0, src[1], 0);
                break;
        case nir_op_imul:
                /*
                 * dst = (al * bl) + (ah * bl << 16) + (al * bh << 16)
                 *   mull.u tmp0, a, b           ; mul low, i.e. al * bl
                 *   madsh.m16 tmp1, a, b, tmp0  ; mul-add shift high mix, i.e. ah * bl << 16
                 *   madsh.m16 dst, b, a, tmp1   ; i.e. al * bh << 16
                 */
                dst[0] = ir3_MADSH_M16(b, src[1], 0, src[0], 0,
                                        ir3_MADSH_M16(b, src[0], 0, src[1], 0,
                                                ir3_MULL_U(b, src[0], 0, src[1], 0), 0), 0);
                break;
        case nir_op_ineg:
                dst[0] = ir3_ABSNEG_S(b, src[0], IR3_REG_SNEG);
                break;
        case nir_op_inot:
                dst[0] = ir3_NOT_B(b, src[0], 0);
                break;
        case nir_op_ior:
                dst[0] = ir3_OR_B(b, src[0], 0, src[1], 0);
                break;
        case nir_op_ishl:
                dst[0] = ir3_SHL_B(b, src[0], 0, src[1], 0);
                break;
        case nir_op_ishr:
                dst[0] = ir3_ASHR_B(b, src[0], 0, src[1], 0);
                break;
        case nir_op_isign: {
                /* maybe this would be sane to lower in nir.. */
                struct ir3_instruction *neg, *pos;

                neg = ir3_CMPS_S(b, src[0], 0, create_immed(b, 0), 0);
                neg->cat2.condition = IR3_COND_LT;

                pos = ir3_CMPS_S(b, src[0], 0, create_immed(b, 0), 0);
                pos->cat2.condition = IR3_COND_GT;

                dst[0] = ir3_SUB_U(b, pos, 0, neg, 0);

                break;
        }
        case nir_op_isub:
                dst[0] = ir3_SUB_U(b, src[0], 0, src[1], 0);
                break;
        case nir_op_ixor:
                dst[0] = ir3_XOR_B(b, src[0], 0, src[1], 0);
                break;
        case nir_op_ushr:
                dst[0] = ir3_SHR_B(b, src[0], 0, src[1], 0);
                break;
        case nir_op_ilt:
                dst[0] = ir3_CMPS_S(b, src[0], 0, src[1], 0);
                dst[0]->cat2.condition = IR3_COND_LT;
                dst[0] = ir3_n2b(b, dst[0]);
                break;
        case nir_op_ige:
                dst[0] = ir3_CMPS_S(b, src[0], 0, src[1], 0);
                dst[0]->cat2.condition = IR3_COND_GE;
                dst[0] = ir3_n2b(b, dst[0]);
                break;
        case nir_op_ieq:
                dst[0] = ir3_CMPS_S(b, src[0], 0, src[1], 0);
                dst[0]->cat2.condition = IR3_COND_EQ;
                dst[0] = ir3_n2b(b, dst[0]);
                break;
        case nir_op_ine:
                dst[0] = ir3_CMPS_S(b, src[0], 0, src[1], 0);
                dst[0]->cat2.condition = IR3_COND_NE;
                dst[0] = ir3_n2b(b, dst[0]);
                break;
        case nir_op_ult:
                dst[0] = ir3_CMPS_U(b, src[0], 0, src[1], 0);
                dst[0]->cat2.condition = IR3_COND_LT;
                dst[0] = ir3_n2b(b, dst[0]);
                break;
        case nir_op_uge:
                dst[0] = ir3_CMPS_U(b, src[0], 0, src[1], 0);
                dst[0]->cat2.condition = IR3_COND_GE;
                dst[0] = ir3_n2b(b, dst[0]);
                break;

        case nir_op_bcsel: {
                struct ir3_instruction *cond = ir3_b2n(b, src[0]);
                compile_assert(ctx, bs[1] == bs[2]);
                /* the boolean condition is 32b even if src[1] and src[2] are
                 * half-precision, but sel.b16 wants all three src's to be the
                 * same type.
                 */
                if (bs[1] < 32)
                        cond = ir3_COV(b, cond, TYPE_U32, TYPE_U16);
                dst[0] = ir3_SEL_B32(b, src[1], 0, cond, 0, src[2], 0);
                break;
        }
        case nir_op_bit_count:
                dst[0] = ir3_CBITS_B(b, src[0], 0);
                break;
        case nir_op_ifind_msb: {
                struct ir3_instruction *cmp;
                dst[0] = ir3_CLZ_S(b, src[0], 0);
                cmp = ir3_CMPS_S(b, dst[0], 0, create_immed(b, 0), 0);
                cmp->cat2.condition = IR3_COND_GE;
                dst[0] = ir3_SEL_B32(b,
                                ir3_SUB_U(b, create_immed(b, 31), 0, dst[0], 0), 0,
                                cmp, 0, dst[0], 0);
                break;
        }
        case nir_op_ufind_msb:
                dst[0] = ir3_CLZ_B(b, src[0], 0);
                dst[0] = ir3_SEL_B32(b,
                                ir3_SUB_U(b, create_immed(b, 31), 0, dst[0], 0), 0,
                                src[0], 0, dst[0], 0);
                break;
        case nir_op_find_lsb:
                dst[0] = ir3_BFREV_B(b, src[0], 0);
                dst[0] = ir3_CLZ_B(b, dst[0], 0);
                break;
        case nir_op_bitfield_reverse:
                dst[0] = ir3_BFREV_B(b, src[0], 0);
                break;

        default:
                compile_error(ctx, "Unhandled ALU op: %s\n",
                                nir_op_infos[alu->op].name);
                break;
        }

        put_dst(ctx, &alu->dest.dest);
}

/* handles direct/indirect UBO reads: */
static void
emit_intrinsic_load_ubo(struct ir3_context *ctx, nir_intrinsic_instr *intr,
                struct ir3_instruction **dst)
{
        struct ir3_block *b = ctx->block;
        struct ir3_instruction *base_lo, *base_hi, *addr, *src0, *src1;
        nir_const_value *const_offset;
        /* UBO addresses are the first driver params: */
        unsigned ubo = regid(ctx->so->constbase.ubo, 0);
        const unsigned ptrsz = pointer_size(ctx);

        int off = 0;

        /* First src is ubo index, which could either be an immed or not: */
        src0 = get_src(ctx, &intr->src[0])[0];
        if (is_same_type_mov(src0) &&
                        (src0->regs[1]->flags & IR3_REG_IMMED)) {
                base_lo = create_uniform(ctx, ubo + (src0->regs[1]->iim_val * ptrsz));
                base_hi = create_uniform(ctx, ubo + (src0->regs[1]->iim_val * ptrsz) + 1);
        } else {
                base_lo = create_uniform_indirect(ctx, ubo, get_addr(ctx, src0, 4));
                base_hi = create_uniform_indirect(ctx, ubo + 1, get_addr(ctx, src0, 4));
        }

        /* note: on 32bit gpu's base_hi is ignored and DCE'd */
        addr = base_lo;

        const_offset = nir_src_as_const_value(intr->src[1]);
        if (const_offset) {
                off += const_offset->u32[0];
        } else {
                /* For load_ubo_indirect, second src is indirect offset: */
                src1 = get_src(ctx, &intr->src[1])[0];

                /* and add offset to addr: */
                addr = ir3_ADD_S(b, addr, 0, src1, 0);
        }

        /* if offset is to large to encode in the ldg, split it out: */
        if ((off + (intr->num_components * 4)) > 1024) {
                /* split out the minimal amount to improve the odds that
                 * cp can fit the immediate in the add.s instruction:
                 */
                unsigned off2 = off + (intr->num_components * 4) - 1024;
                addr = ir3_ADD_S(b, addr, 0, create_immed(b, off2), 0);
                off -= off2;
        }

        if (ptrsz == 2) {
                struct ir3_instruction *carry;

                /* handle 32b rollover, ie:
                 *   if (addr < base_lo)
                 *      base_hi++
                 */
                carry = ir3_CMPS_U(b, addr, 0, base_lo, 0);
                carry->cat2.condition = IR3_COND_LT;
                base_hi = ir3_ADD_S(b, base_hi, 0, carry, 0);

                addr = create_collect(ctx, (struct ir3_instruction*[]){ addr, base_hi }, 2);
        }

        for (int i = 0; i < intr->num_components; i++) {
                struct ir3_instruction *load =
                                ir3_LDG(b, addr, 0, create_immed(b, 1), 0);
                load->cat6.type = TYPE_U32;
                load->cat6.src_offset = off + i * 4;     /* byte offset */
                dst[i] = load;
        }
}

/* src[] = { buffer_index, offset }. No const_index */
static void
emit_intrinsic_load_ssbo(struct ir3_context *ctx, nir_intrinsic_instr *intr,
                struct ir3_instruction **dst)
{
        struct ir3_block *b = ctx->block;
        struct ir3_instruction *ldgb, *src0, *src1, *offset;
        nir_const_value *const_offset;

        /* can this be non-const buffer_index?  how do we handle that? */
        const_offset = nir_src_as_const_value(intr->src[0]);
        compile_assert(ctx, const_offset);

        offset = get_src(ctx, &intr->src[1])[0];

        /* src0 is uvec2(offset*4, 0), src1 is offset.. nir already *= 4: */
        src0 = create_collect(ctx, (struct ir3_instruction*[]){
                offset,
                create_immed(b, 0),
        }, 2);
        src1 = ir3_SHR_B(b, offset, 0, create_immed(b, 2), 0);

        ldgb = ir3_LDGB(b, create_immed(b, const_offset->u32[0]), 0,
                        src0, 0, src1, 0);
        ldgb->regs[0]->wrmask = MASK(intr->num_components);
        ldgb->cat6.iim_val = intr->num_components;
        ldgb->cat6.d = 4;
        ldgb->cat6.type = TYPE_U32;
        ldgb->barrier_class = IR3_BARRIER_BUFFER_R;
        ldgb->barrier_conflict = IR3_BARRIER_BUFFER_W;

        split_dest(b, dst, ldgb, 0, intr->num_components);
}

/* src[] = { value, block_index, offset }. const_index[] = { write_mask } */
static void
emit_intrinsic_store_ssbo(struct ir3_context *ctx, nir_intrinsic_instr *intr)
{
        struct ir3_block *b = ctx->block;
        struct ir3_instruction *stgb, *src0, *src1, *src2, *offset;
        nir_const_value *const_offset;
        /* TODO handle wrmask properly, see _store_shared().. but I think
         * it is more a PITA than that, since blob ends up loading the
         * masked components and writing them back out.
         */
        unsigned wrmask = intr->const_index[0];
        unsigned ncomp = ffs(~wrmask) - 1;

        /* can this be non-const buffer_index?  how do we handle that? */
        const_offset = nir_src_as_const_value(intr->src[1]);
        compile_assert(ctx, const_offset);

        offset = get_src(ctx, &intr->src[2])[0];

        /* src0 is value, src1 is offset, src2 is uvec2(offset*4, 0)..
         * nir already *= 4:
         */
        src0 = create_collect(ctx, get_src(ctx, &intr->src[0]), ncomp);
        src1 = ir3_SHR_B(b, offset, 0, create_immed(b, 2), 0);
        src2 = create_collect(ctx, (struct ir3_instruction*[]){
                offset,
                create_immed(b, 0),
        }, 2);

        stgb = ir3_STGB(b, create_immed(b, const_offset->u32[0]), 0,
                        src0, 0, src1, 0, src2, 0);
        stgb->cat6.iim_val = ncomp;
        stgb->cat6.d = 4;
        stgb->cat6.type = TYPE_U32;
        stgb->barrier_class = IR3_BARRIER_BUFFER_W;
        stgb->barrier_conflict = IR3_BARRIER_BUFFER_R | IR3_BARRIER_BUFFER_W;

        array_insert(b, b->keeps, stgb);
}

/* src[] = { block_index } */
static void
emit_intrinsic_ssbo_size(struct ir3_context *ctx, nir_intrinsic_instr *intr,
                struct ir3_instruction **dst)
{
        /* SSBO size stored as a const starting at ssbo_sizes: */
        unsigned blk_idx = nir_src_as_const_value(intr->src[0])->u32[0];
        unsigned idx = regid(ctx->so->constbase.ssbo_sizes, 0) +
                ctx->so->const_layout.ssbo_size.off[blk_idx];

        debug_assert(ctx->so->const_layout.ssbo_size.mask & (1 << blk_idx));

        dst[0] = create_uniform(ctx, idx);
}

/*
 * SSBO atomic intrinsics
 *
 * All of the SSBO atomic memory operations read a value from memory,
 * compute a new value using one of the operations below, write the new
 * value to memory, and return the original value read.
 *
 * All operations take 3 sources except CompSwap that takes 4. These
 * sources represent:
 *
 * 0: The SSBO buffer index.
 * 1: The offset into the SSBO buffer of the variable that the atomic
 *    operation will operate on.
 * 2: The data parameter to the atomic function (i.e. the value to add
 *    in ssbo_atomic_add, etc).
 * 3: For CompSwap only: the second data parameter.
 */
static struct ir3_instruction *
emit_intrinsic_atomic_ssbo(struct ir3_context *ctx, nir_intrinsic_instr *intr)
{
        struct ir3_block *b = ctx->block;
        struct ir3_instruction *atomic, *ssbo, *src0, *src1, *src2, *offset;
        nir_const_value *const_offset;
        type_t type = TYPE_U32;

        /* can this be non-const buffer_index?  how do we handle that? */
        const_offset = nir_src_as_const_value(intr->src[0]);
        compile_assert(ctx, const_offset);
        ssbo = create_immed(b, const_offset->u32[0]);

        offset = get_src(ctx, &intr->src[1])[0];

        /* src0 is data (or uvec2(data, compare))
         * src1 is offset
         * src2 is uvec2(offset*4, 0) (appears to be 64b byte offset)
         *
         * Note that nir already multiplies the offset by four
         */
        src0 = get_src(ctx, &intr->src[2])[0];
        src1 = ir3_SHR_B(b, offset, 0, create_immed(b, 2), 0);
        src2 = create_collect(ctx, (struct ir3_instruction*[]){
                offset,
                create_immed(b, 0),
        }, 2);

        switch (intr->intrinsic) {
        case nir_intrinsic_ssbo_atomic_add:
                atomic = ir3_ATOMIC_ADD_G(b, ssbo, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_ssbo_atomic_imin:
                atomic = ir3_ATOMIC_MIN_G(b, ssbo, 0, src0, 0, src1, 0, src2, 0);
                type = TYPE_S32;
                break;
        case nir_intrinsic_ssbo_atomic_umin:
                atomic = ir3_ATOMIC_MIN_G(b, ssbo, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_ssbo_atomic_imax:
                atomic = ir3_ATOMIC_MAX_G(b, ssbo, 0, src0, 0, src1, 0, src2, 0);
                type = TYPE_S32;
                break;
        case nir_intrinsic_ssbo_atomic_umax:
                atomic = ir3_ATOMIC_MAX_G(b, ssbo, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_ssbo_atomic_and:
                atomic = ir3_ATOMIC_AND_G(b, ssbo, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_ssbo_atomic_or:
                atomic = ir3_ATOMIC_OR_G(b, ssbo, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_ssbo_atomic_xor:
                atomic = ir3_ATOMIC_XOR_G(b, ssbo, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_ssbo_atomic_exchange:
                atomic = ir3_ATOMIC_XCHG_G(b, ssbo, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_ssbo_atomic_comp_swap:
                /* for cmpxchg, src0 is [ui]vec2(data, compare): */
                src0 = create_collect(ctx, (struct ir3_instruction*[]){
                        get_src(ctx, &intr->src[3])[0],
                        src0,
                }, 2);
                atomic = ir3_ATOMIC_CMPXCHG_G(b, ssbo, 0, src0, 0, src1, 0, src2, 0);
                break;
        default:
                unreachable("boo");
        }

        atomic->cat6.iim_val = 1;
        atomic->cat6.d = 4;
        atomic->cat6.type = type;
        atomic->barrier_class = IR3_BARRIER_BUFFER_W;
        atomic->barrier_conflict = IR3_BARRIER_BUFFER_R | IR3_BARRIER_BUFFER_W;

        /* even if nothing consume the result, we can't DCE the instruction: */
        array_insert(b, b->keeps, atomic);

        return atomic;
}

/* src[] = { offset }. const_index[] = { base } */
static void
emit_intrinsic_load_shared(struct ir3_context *ctx, nir_intrinsic_instr *intr,
                struct ir3_instruction **dst)
{
        struct ir3_block *b = ctx->block;
        struct ir3_instruction *ldl, *offset;
        unsigned base;

        offset = get_src(ctx, &intr->src[0])[0];
        base   = nir_intrinsic_base(intr);

        ldl = ir3_LDL(b, offset, 0, create_immed(b, intr->num_components), 0);
        ldl->cat6.src_offset = base;
        ldl->cat6.type = utype_dst(intr->dest);
        ldl->regs[0]->wrmask = MASK(intr->num_components);

        ldl->barrier_class = IR3_BARRIER_SHARED_R;
        ldl->barrier_conflict = IR3_BARRIER_SHARED_W;

        split_dest(b, dst, ldl, 0, intr->num_components);
}

/* src[] = { value, offset }. const_index[] = { base, write_mask } */
static void
emit_intrinsic_store_shared(struct ir3_context *ctx, nir_intrinsic_instr *intr)
{
        struct ir3_block *b = ctx->block;
        struct ir3_instruction *stl, *offset;
        struct ir3_instruction * const *value;
        unsigned base, wrmask;

        value  = get_src(ctx, &intr->src[0]);
        offset = get_src(ctx, &intr->src[1])[0];

        base   = nir_intrinsic_base(intr);
        wrmask = nir_intrinsic_write_mask(intr);

        /* Combine groups of consecutive enabled channels in one write
         * message. We use ffs to find the first enabled channel and then ffs on
         * the bit-inverse, down-shifted writemask to determine the length of
         * the block of enabled bits.
         *
         * (trick stolen from i965's fs_visitor::nir_emit_cs_intrinsic())
         */
        while (wrmask) {
                unsigned first_component = ffs(wrmask) - 1;
                unsigned length = ffs(~(wrmask >> first_component)) - 1;

                stl = ir3_STL(b, offset, 0,
                        create_collect(ctx, &value[first_component], length), 0,
                        create_immed(b, length), 0);
                stl->cat6.dst_offset = first_component + base;
                stl->cat6.type = utype_src(intr->src[0]);
                stl->barrier_class = IR3_BARRIER_SHARED_W;
                stl->barrier_conflict = IR3_BARRIER_SHARED_R | IR3_BARRIER_SHARED_W;

                array_insert(b, b->keeps, stl);

                /* Clear the bits in the writemask that we just wrote, then try
                 * again to see if more channels are left.
                 */
                wrmask &= (15 << (first_component + length));
        }
}

/*
 * CS shared variable atomic intrinsics
 *
 * All of the shared variable atomic memory operations read a value from
 * memory, compute a new value using one of the operations below, write the
 * new value to memory, and return the original value read.
 *
 * All operations take 2 sources except CompSwap that takes 3. These
 * sources represent:
 *
 * 0: The offset into the shared variable storage region that the atomic
 *    operation will operate on.
 * 1: The data parameter to the atomic function (i.e. the value to add
 *    in shared_atomic_add, etc).
 * 2: For CompSwap only: the second data parameter.
 */
static struct ir3_instruction *
emit_intrinsic_atomic_shared(struct ir3_context *ctx, nir_intrinsic_instr *intr)
{
        struct ir3_block *b = ctx->block;
        struct ir3_instruction *atomic, *src0, *src1;
        type_t type = TYPE_U32;

        src0 = get_src(ctx, &intr->src[0])[0];   /* offset */
        src1 = get_src(ctx, &intr->src[1])[0];   /* value */

        switch (intr->intrinsic) {
        case nir_intrinsic_shared_atomic_add:
                atomic = ir3_ATOMIC_ADD(b, src0, 0, src1, 0);
                break;
        case nir_intrinsic_shared_atomic_imin:
                atomic = ir3_ATOMIC_MIN(b, src0, 0, src1, 0);
                type = TYPE_S32;
                break;
        case nir_intrinsic_shared_atomic_umin:
                atomic = ir3_ATOMIC_MIN(b, src0, 0, src1, 0);
                break;
        case nir_intrinsic_shared_atomic_imax:
                atomic = ir3_ATOMIC_MAX(b, src0, 0, src1, 0);
                type = TYPE_S32;
                break;
        case nir_intrinsic_shared_atomic_umax:
                atomic = ir3_ATOMIC_MAX(b, src0, 0, src1, 0);
                break;
        case nir_intrinsic_shared_atomic_and:
                atomic = ir3_ATOMIC_AND(b, src0, 0, src1, 0);
                break;
        case nir_intrinsic_shared_atomic_or:
                atomic = ir3_ATOMIC_OR(b, src0, 0, src1, 0);
                break;
        case nir_intrinsic_shared_atomic_xor:
                atomic = ir3_ATOMIC_XOR(b, src0, 0, src1, 0);
                break;
        case nir_intrinsic_shared_atomic_exchange:
                atomic = ir3_ATOMIC_XCHG(b, src0, 0, src1, 0);
                break;
        case nir_intrinsic_shared_atomic_comp_swap:
                /* for cmpxchg, src1 is [ui]vec2(data, compare): */
                src1 = create_collect(ctx, (struct ir3_instruction*[]){
                        get_src(ctx, &intr->src[2])[0],
                        src1,
                }, 2);
                atomic = ir3_ATOMIC_CMPXCHG(b, src0, 0, src1, 0);
                break;
        default:
                unreachable("boo");
        }

        atomic->cat6.iim_val = 1;
        atomic->cat6.d = 1;
        atomic->cat6.type = type;
        atomic->barrier_class = IR3_BARRIER_SHARED_W;
        atomic->barrier_conflict = IR3_BARRIER_SHARED_R | IR3_BARRIER_SHARED_W;

        /* even if nothing consume the result, we can't DCE the instruction: */
        array_insert(b, b->keeps, atomic);

        return atomic;
}

/* Images get mapped into SSBO/image state (for store/atomic) and texture
 * state block (for load).  To simplify things, invert the image id and
 * map it from end of state block, ie. image 0 becomes num-1, image 1
 * becomes num-2, etc.  This potentially avoids needing to re-emit texture
 * state when switching shaders.
 *
 * TODO is max # of samplers and SSBOs the same.  This shouldn't be hard-
 * coded.  Also, since all the gl shader stages (ie. everything but CS)
 * share the same SSBO/image state block, this might require some more
 * logic if we supported images in anything other than FS..
 */
static unsigned
get_image_slot(struct ir3_context *ctx, nir_deref_instr *deref)
{
        unsigned int loc = 0;
        unsigned inner_size = 1;

        while (deref->deref_type != nir_deref_type_var) {
                assert(deref->deref_type == nir_deref_type_array);
                nir_const_value *const_index = nir_src_as_const_value(deref->arr.index);
                assert(const_index);

                /* Go to the next instruction */
                deref = nir_deref_instr_parent(deref);

                assert(glsl_type_is_array(deref->type));
                const unsigned array_len = glsl_get_length(deref->type);
                loc += MIN2(const_index->u32[0], array_len - 1) * inner_size;

                /* Update the inner size */
                inner_size *= array_len;
        }

        loc += deref->var->data.driver_location;

        /* TODO figure out real limit per generation, and don't hardcode: */
        const unsigned max_samplers = 16;
        return max_samplers - loc - 1;
}

/* see tex_info() for equiv logic for texture instructions.. it would be
 * nice if this could be better unified..
 */
static unsigned
get_image_coords(const nir_variable *var, unsigned *flagsp)
{
        const struct glsl_type *type = glsl_without_array(var->type);
        unsigned coords, flags = 0;

        switch (glsl_get_sampler_dim(type)) {
        case GLSL_SAMPLER_DIM_1D:
        case GLSL_SAMPLER_DIM_BUF:
                coords = 1;
                break;
        case GLSL_SAMPLER_DIM_2D:
        case GLSL_SAMPLER_DIM_RECT:
        case GLSL_SAMPLER_DIM_EXTERNAL:
        case GLSL_SAMPLER_DIM_MS:
                coords = 2;
                break;
        case GLSL_SAMPLER_DIM_3D:
        case GLSL_SAMPLER_DIM_CUBE:
                flags |= IR3_INSTR_3D;
                coords = 3;
                break;
        default:
                unreachable("bad sampler dim");
                return 0;
        }

        if (glsl_sampler_type_is_array(type)) {
                /* note: unlike tex_info(), adjust # of coords to include array idx: */
                coords++;
                flags |= IR3_INSTR_A;
        }

        if (flagsp)
                *flagsp = flags;

        return coords;
}

static type_t
get_image_type(const nir_variable *var)
{
        switch (glsl_get_sampler_result_type(glsl_without_array(var->type))) {
        case GLSL_TYPE_UINT:
                return TYPE_U32;
        case GLSL_TYPE_INT:
                return TYPE_S32;
        case GLSL_TYPE_FLOAT:
                return TYPE_F32;
        default:
                unreachable("bad sampler type.");
                return 0;
        }
}

static struct ir3_instruction *
get_image_offset(struct ir3_context *ctx, const nir_variable *var,
                struct ir3_instruction * const *coords, bool byteoff)
{
        struct ir3_block *b = ctx->block;
        struct ir3_instruction *offset;
        unsigned ncoords = get_image_coords(var, NULL);

        /* to calculate the byte offset (yes, uggg) we need (up to) three
         * const values to know the bytes per pixel, and y and z stride:
         */
        unsigned cb = regid(ctx->so->constbase.image_dims, 0) +
                ctx->so->const_layout.image_dims.off[var->data.driver_location];

        debug_assert(ctx->so->const_layout.image_dims.mask &
                        (1 << var->data.driver_location));

        /* offset = coords.x * bytes_per_pixel: */
        offset = ir3_MUL_S(b, coords[0], 0, create_uniform(ctx, cb + 0), 0);
        if (ncoords > 1) {
                /* offset += coords.y * y_pitch: */
                offset = ir3_MAD_S24(b, create_uniform(ctx, cb + 1), 0,
                                coords[1], 0, offset, 0);
        }
        if (ncoords > 2) {
                /* offset += coords.z * z_pitch: */
                offset = ir3_MAD_S24(b, create_uniform(ctx, cb + 2), 0,
                                coords[2], 0, offset, 0);
        }

        if (!byteoff) {
                /* Some cases, like atomics, seem to use dword offset instead
                 * of byte offsets.. blob just puts an extra shr.b in there
                 * in those cases:
                 */
                offset = ir3_SHR_B(b, offset, 0, create_immed(b, 2), 0);
        }

        return create_collect(ctx, (struct ir3_instruction*[]){
                offset,
                create_immed(b, 0),
        }, 2);
}

/* src[] = { deref, coord, sample_index }. const_index[] = {} */
static void
emit_intrinsic_load_image(struct ir3_context *ctx, nir_intrinsic_instr *intr,
                struct ir3_instruction **dst)
{
        struct ir3_block *b = ctx->block;
        const nir_variable *var = nir_intrinsic_get_var(intr, 0);
        struct ir3_instruction *sam;
        struct ir3_instruction * const *src0 = get_src(ctx, &intr->src[1]);
        struct ir3_instruction *coords[4];
        unsigned flags, ncoords = get_image_coords(var, &flags);
        unsigned tex_idx = get_image_slot(ctx, nir_src_as_deref(intr->src[0]));
        type_t type = get_image_type(var);

        /* hmm, this seems a bit odd, but it is what blob does and (at least
         * a5xx) just faults on bogus addresses otherwise:
         */
        if (flags & IR3_INSTR_3D) {
                flags &= ~IR3_INSTR_3D;
                flags |= IR3_INSTR_A;
        }

        for (unsigned i = 0; i < ncoords; i++)
                coords[i] = src0[i];

        if (ncoords == 1)
                coords[ncoords++] = create_immed(b, 0);

        sam = ir3_SAM(b, OPC_ISAM, type, 0b1111, flags,
                        tex_idx, tex_idx, create_collect(ctx, coords, ncoords), NULL);

        sam->barrier_class = IR3_BARRIER_IMAGE_R;
        sam->barrier_conflict = IR3_BARRIER_IMAGE_W;

        split_dest(b, dst, sam, 0, 4);
}

/* src[] = { deref, coord, sample_index, value }. const_index[] = {} */
static void
emit_intrinsic_store_image(struct ir3_context *ctx, nir_intrinsic_instr *intr)
{
        struct ir3_block *b = ctx->block;
        const nir_variable *var = nir_intrinsic_get_var(intr, 0);
        struct ir3_instruction *stib, *offset;
        struct ir3_instruction * const *value = get_src(ctx, &intr->src[3]);
        struct ir3_instruction * const *coords = get_src(ctx, &intr->src[1]);
        unsigned ncoords = get_image_coords(var, NULL);
        unsigned tex_idx = get_image_slot(ctx, nir_src_as_deref(intr->src[0]));

        /* src0 is value
         * src1 is coords
         * src2 is 64b byte offset
         */

        offset = get_image_offset(ctx, var, coords, true);

        /* NOTE: stib seems to take byte offset, but stgb.typed can be used
         * too and takes a dword offset.. not quite sure yet why blob uses
         * one over the other in various cases.
         */

        stib = ir3_STIB(b, create_immed(b, tex_idx), 0,
                        create_collect(ctx, value, 4), 0,
                        create_collect(ctx, coords, ncoords), 0,
                        offset, 0);
        stib->cat6.iim_val = 4;
        stib->cat6.d = ncoords;
        stib->cat6.type = get_image_type(var);
        stib->cat6.typed = true;
        stib->barrier_class = IR3_BARRIER_IMAGE_W;
        stib->barrier_conflict = IR3_BARRIER_IMAGE_R | IR3_BARRIER_IMAGE_W;

        array_insert(b, b->keeps, stib);
}

static void
emit_intrinsic_image_size(struct ir3_context *ctx, nir_intrinsic_instr *intr,
                struct ir3_instruction **dst)
{
        struct ir3_block *b = ctx->block;
        const nir_variable *var = nir_intrinsic_get_var(intr, 0);
        unsigned tex_idx = get_image_slot(ctx, nir_src_as_deref(intr->src[0]));
        struct ir3_instruction *sam, *lod;
        unsigned flags, ncoords = get_image_coords(var, &flags);

        lod = create_immed(b, 0);
        sam = ir3_SAM(b, OPC_GETSIZE, TYPE_U32, 0b1111, flags,
                        tex_idx, tex_idx, lod, NULL);

        /* Array size actually ends up in .w rather than .z. This doesn't
         * matter for miplevel 0, but for higher mips the value in z is
         * minified whereas w stays. Also, the value in TEX_CONST_3_DEPTH is
         * returned, which means that we have to add 1 to it for arrays for
         * a3xx.
         *
         * Note use a temporary dst and then copy, since the size of the dst
         * array that is passed in is based on nir's understanding of the
         * result size, not the hardware's
         */
        struct ir3_instruction *tmp[4];

        split_dest(b, tmp, sam, 0, 4);

        /* get_size instruction returns size in bytes instead of texels
         * for imageBuffer, so we need to divide it by the pixel size
         * of the image format.
         *
         * TODO: This is at least true on a5xx. Check other gens.
         */
        enum glsl_sampler_dim dim =
                glsl_get_sampler_dim(glsl_without_array(var->type));
        if (dim == GLSL_SAMPLER_DIM_BUF) {
                /* Since all the possible values the divisor can take are
                 * power-of-two (4, 8, or 16), the division is implemented
                 * as a shift-right.
                 * During shader setup, the log2 of the image format's
                 * bytes-per-pixel should have been emitted in 2nd slot of
                 * image_dims. See ir3_shader::emit_image_dims().
                 */
                unsigned cb = regid(ctx->so->constbase.image_dims, 0) +
                        ctx->so->const_layout.image_dims.off[var->data.driver_location];
                struct ir3_instruction *aux = create_uniform(ctx, cb + 1);

                tmp[0] = ir3_SHR_B(b, tmp[0], 0, aux, 0);
        }

        for (unsigned i = 0; i < ncoords; i++)
                dst[i] = tmp[i];

        if (flags & IR3_INSTR_A) {
                if (ctx->compiler->levels_add_one) {
                        dst[ncoords-1] = ir3_ADD_U(b, tmp[3], 0, create_immed(b, 1), 0);
                } else {
                        dst[ncoords-1] = ir3_MOV(b, tmp[3], TYPE_U32);
                }
        }
}

/* src[] = { deref, coord, sample_index, value, compare }. const_index[] = {} */
static struct ir3_instruction *
emit_intrinsic_atomic_image(struct ir3_context *ctx, nir_intrinsic_instr *intr)
{
        struct ir3_block *b = ctx->block;
        const nir_variable *var = nir_intrinsic_get_var(intr, 0);
        struct ir3_instruction *atomic, *image, *src0, *src1, *src2;
        struct ir3_instruction * const *coords = get_src(ctx, &intr->src[1]);
        unsigned ncoords = get_image_coords(var, NULL);

        image = create_immed(b, get_image_slot(ctx, nir_src_as_deref(intr->src[0])));

        /* src0 is value (or uvec2(value, compare))
         * src1 is coords
         * src2 is 64b byte offset
         */
        src0 = get_src(ctx, &intr->src[3])[0];
        src1 = create_collect(ctx, coords, ncoords);
        src2 = get_image_offset(ctx, var, coords, false);

        switch (intr->intrinsic) {
        case nir_intrinsic_image_deref_atomic_add:
                atomic = ir3_ATOMIC_ADD_G(b, image, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_image_deref_atomic_min:
                atomic = ir3_ATOMIC_MIN_G(b, image, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_image_deref_atomic_max:
                atomic = ir3_ATOMIC_MAX_G(b, image, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_image_deref_atomic_and:
                atomic = ir3_ATOMIC_AND_G(b, image, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_image_deref_atomic_or:
                atomic = ir3_ATOMIC_OR_G(b, image, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_image_deref_atomic_xor:
                atomic = ir3_ATOMIC_XOR_G(b, image, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_image_deref_atomic_exchange:
                atomic = ir3_ATOMIC_XCHG_G(b, image, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_image_deref_atomic_comp_swap:
                /* for cmpxchg, src0 is [ui]vec2(data, compare): */
                src0 = create_collect(ctx, (struct ir3_instruction*[]){
                        get_src(ctx, &intr->src[4])[0],
                        src0,
                }, 2);
                atomic = ir3_ATOMIC_CMPXCHG_G(b, image, 0, src0, 0, src1, 0, src2, 0);
                break;
        default:
                unreachable("boo");
        }

        atomic->cat6.iim_val = 1;
        atomic->cat6.d = ncoords;
        atomic->cat6.type = get_image_type(var);
        atomic->cat6.typed = true;
        atomic->barrier_class = IR3_BARRIER_IMAGE_W;
        atomic->barrier_conflict = IR3_BARRIER_IMAGE_R | IR3_BARRIER_IMAGE_W;

        /* even if nothing consume the result, we can't DCE the instruction: */
        array_insert(b, b->keeps, atomic);

        return atomic;
}

static void
emit_intrinsic_barrier(struct ir3_context *ctx, nir_intrinsic_instr *intr)
{
        struct ir3_block *b = ctx->block;
        struct ir3_instruction *barrier;

        switch (intr->intrinsic) {
        case nir_intrinsic_barrier:
                barrier = ir3_BAR(b);
                barrier->cat7.g = true;
                barrier->cat7.l = true;
                barrier->flags = IR3_INSTR_SS | IR3_INSTR_SY;
                barrier->barrier_class = IR3_BARRIER_EVERYTHING;
                break;
        case nir_intrinsic_memory_barrier:
                barrier = ir3_FENCE(b);
                barrier->cat7.g = true;
                barrier->cat7.r = true;
                barrier->cat7.w = true;
                barrier->barrier_class = IR3_BARRIER_IMAGE_W |
                                IR3_BARRIER_BUFFER_W;
                barrier->barrier_conflict =
                                IR3_BARRIER_IMAGE_R | IR3_BARRIER_IMAGE_W |
                                IR3_BARRIER_BUFFER_R | IR3_BARRIER_BUFFER_W;
                break;
        case nir_intrinsic_memory_barrier_atomic_counter:
        case nir_intrinsic_memory_barrier_buffer:
                barrier = ir3_FENCE(b);
                barrier->cat7.g = true;
                barrier->cat7.r = true;
                barrier->cat7.w = true;
                barrier->barrier_class = IR3_BARRIER_BUFFER_W;
                barrier->barrier_conflict = IR3_BARRIER_BUFFER_R |
                                IR3_BARRIER_BUFFER_W;
                break;
        case nir_intrinsic_memory_barrier_image:
                // TODO double check if this should have .g set
                barrier = ir3_FENCE(b);
                barrier->cat7.g = true;
                barrier->cat7.r = true;
                barrier->cat7.w = true;
                barrier->barrier_class = IR3_BARRIER_IMAGE_W;
                barrier->barrier_conflict = IR3_BARRIER_IMAGE_R |
                                IR3_BARRIER_IMAGE_W;
                break;
        case nir_intrinsic_memory_barrier_shared:
                barrier = ir3_FENCE(b);
                barrier->cat7.g = true;
                barrier->cat7.l = true;
                barrier->cat7.r = true;
                barrier->cat7.w = true;
                barrier->barrier_class = IR3_BARRIER_SHARED_W;
                barrier->barrier_conflict = IR3_BARRIER_SHARED_R |
                                IR3_BARRIER_SHARED_W;
                break;
        case nir_intrinsic_group_memory_barrier:
                barrier = ir3_FENCE(b);
                barrier->cat7.g = true;
                barrier->cat7.l = true;
                barrier->cat7.r = true;
                barrier->cat7.w = true;
                barrier->barrier_class = IR3_BARRIER_SHARED_W |
                                IR3_BARRIER_IMAGE_W |
                                IR3_BARRIER_BUFFER_W;
                barrier->barrier_conflict =
                                IR3_BARRIER_SHARED_R | IR3_BARRIER_SHARED_W |
                                IR3_BARRIER_IMAGE_R | IR3_BARRIER_IMAGE_W |
                                IR3_BARRIER_BUFFER_R | IR3_BARRIER_BUFFER_W;
                break;
        default:
                unreachable("boo");
        }

        /* make sure barrier doesn't get DCE'd */
        array_insert(b, b->keeps, barrier);
}

static void add_sysval_input_compmask(struct ir3_context *ctx,
                gl_system_value slot, unsigned compmask,
                struct ir3_instruction *instr)
{
        struct ir3_shader_variant *so = ctx->so;
        unsigned r = regid(so->inputs_count, 0);
        unsigned n = so->inputs_count++;

        so->inputs[n].sysval = true;
        so->inputs[n].slot = slot;
        so->inputs[n].compmask = compmask;
        so->inputs[n].regid = r;
        so->inputs[n].interpolate = INTERP_MODE_FLAT;
        so->total_in++;

        ctx->ir->ninputs = MAX2(ctx->ir->ninputs, r + 1);
        ctx->ir->inputs[r] = instr;
}

static void add_sysval_input(struct ir3_context *ctx, gl_system_value slot,
                struct ir3_instruction *instr)
{
        add_sysval_input_compmask(ctx, slot, 0x1, instr);
}

static void
emit_intrinsic(struct ir3_context *ctx, nir_intrinsic_instr *intr)
{
        const nir_intrinsic_info *info = &nir_intrinsic_infos[intr->intrinsic];
        struct ir3_instruction **dst;
        struct ir3_instruction * const *src;
        struct ir3_block *b = ctx->block;
        nir_const_value *const_offset;
        int idx, comp;

        if (info->has_dest) {
                unsigned n = nir_intrinsic_dest_components(intr);
                dst = get_dst(ctx, &intr->dest, n);
        } else {
                dst = NULL;
        }

        switch (intr->intrinsic) {
        case nir_intrinsic_load_uniform:
                idx = nir_intrinsic_base(intr);
                const_offset = nir_src_as_const_value(intr->src[0]);
                if (const_offset) {
                        idx += const_offset->u32[0];
                        for (int i = 0; i < intr->num_components; i++) {
                                unsigned n = idx * 4 + i;
                                dst[i] = create_uniform(ctx, n);
                        }
                } else {
                        src = get_src(ctx, &intr->src[0]);
                        for (int i = 0; i < intr->num_components; i++) {
                                int n = idx * 4 + i;
                                dst[i] = create_uniform_indirect(ctx, n,
                                                get_addr(ctx, src[0], 4));
                        }
                        /* NOTE: if relative addressing is used, we set
                         * constlen in the compiler (to worst-case value)
                         * since we don't know in the assembler what the max
                         * addr reg value can be:
                         */
                        ctx->so->constlen = ctx->s->num_uniforms;
                }
                break;
        case nir_intrinsic_load_ubo:
                emit_intrinsic_load_ubo(ctx, intr, dst);
                break;
        case nir_intrinsic_load_input:
                idx = nir_intrinsic_base(intr);
                comp = nir_intrinsic_component(intr);
                const_offset = nir_src_as_const_value(intr->src[0]);
                if (const_offset) {
                        idx += const_offset->u32[0];
                        for (int i = 0; i < intr->num_components; i++) {
                                unsigned n = idx * 4 + i + comp;
                                dst[i] = ctx->ir->inputs[n];
                        }
                } else {
                        src = get_src(ctx, &intr->src[0]);
                        struct ir3_instruction *collect =
                                        create_collect(ctx, ctx->ir->inputs, ctx->ir->ninputs);
                        struct ir3_instruction *addr = get_addr(ctx, src[0], 4);
                        for (int i = 0; i < intr->num_components; i++) {
                                unsigned n = idx * 4 + i + comp;
                                dst[i] = create_indirect_load(ctx, ctx->ir->ninputs,
                                                n, addr, collect);
                        }
                }
                break;
        case nir_intrinsic_load_ssbo:
                emit_intrinsic_load_ssbo(ctx, intr, dst);
                break;
        case nir_intrinsic_store_ssbo:
                emit_intrinsic_store_ssbo(ctx, intr);
                break;
        case nir_intrinsic_get_buffer_size:
                emit_intrinsic_ssbo_size(ctx, intr, dst);
                break;
        case nir_intrinsic_ssbo_atomic_add:
        case nir_intrinsic_ssbo_atomic_imin:
        case nir_intrinsic_ssbo_atomic_umin:
        case nir_intrinsic_ssbo_atomic_imax:
        case nir_intrinsic_ssbo_atomic_umax:
        case nir_intrinsic_ssbo_atomic_and:
        case nir_intrinsic_ssbo_atomic_or:
        case nir_intrinsic_ssbo_atomic_xor:
        case nir_intrinsic_ssbo_atomic_exchange:
        case nir_intrinsic_ssbo_atomic_comp_swap:
                dst[0] = emit_intrinsic_atomic_ssbo(ctx, intr);
                break;
        case nir_intrinsic_load_shared:
                emit_intrinsic_load_shared(ctx, intr, dst);
                break;
        case nir_intrinsic_store_shared:
                emit_intrinsic_store_shared(ctx, intr);
                break;
        case nir_intrinsic_shared_atomic_add:
        case nir_intrinsic_shared_atomic_imin:
        case nir_intrinsic_shared_atomic_umin:
        case nir_intrinsic_shared_atomic_imax:
        case nir_intrinsic_shared_atomic_umax:
        case nir_intrinsic_shared_atomic_and:
        case nir_intrinsic_shared_atomic_or:
        case nir_intrinsic_shared_atomic_xor:
        case nir_intrinsic_shared_atomic_exchange:
        case nir_intrinsic_shared_atomic_comp_swap:
                dst[0] = emit_intrinsic_atomic_shared(ctx, intr);
                break;
        case nir_intrinsic_image_deref_load:
                emit_intrinsic_load_image(ctx, intr, dst);
                break;
        case nir_intrinsic_image_deref_store:
                emit_intrinsic_store_image(ctx, intr);
                break;
        case nir_intrinsic_image_deref_size:
                emit_intrinsic_image_size(ctx, intr, dst);
                break;
        case nir_intrinsic_image_deref_atomic_add:
        case nir_intrinsic_image_deref_atomic_min:
        case nir_intrinsic_image_deref_atomic_max:
        case nir_intrinsic_image_deref_atomic_and:
        case nir_intrinsic_image_deref_atomic_or:
        case nir_intrinsic_image_deref_atomic_xor:
        case nir_intrinsic_image_deref_atomic_exchange:
        case nir_intrinsic_image_deref_atomic_comp_swap:
                dst[0] = emit_intrinsic_atomic_image(ctx, intr);
                break;
        case nir_intrinsic_barrier:
        case nir_intrinsic_memory_barrier:
        case nir_intrinsic_group_memory_barrier:
        case nir_intrinsic_memory_barrier_atomic_counter:
        case nir_intrinsic_memory_barrier_buffer:
        case nir_intrinsic_memory_barrier_image:
        case nir_intrinsic_memory_barrier_shared:
                emit_intrinsic_barrier(ctx, intr);
                /* note that blk ptr no longer valid, make that obvious: */
                b = NULL;
                break;
        case nir_intrinsic_store_output:
                idx = nir_intrinsic_base(intr);
                comp = nir_intrinsic_component(intr);
                const_offset = nir_src_as_const_value(intr->src[1]);
                compile_assert(ctx, const_offset != NULL);
                idx += const_offset->u32[0];

                src = get_src(ctx, &intr->src[0]);
                for (int i = 0; i < intr->num_components; i++) {
                        unsigned n = idx * 4 + i + comp;
                        ctx->ir->outputs[n] = src[i];
                }
                break;
        case nir_intrinsic_load_base_vertex:
        case nir_intrinsic_load_first_vertex:
                if (!ctx->basevertex) {
                        ctx->basevertex = create_driver_param(ctx, IR3_DP_VTXID_BASE);
                        add_sysval_input(ctx, SYSTEM_VALUE_FIRST_VERTEX, ctx->basevertex);
                }
                dst[0] = ctx->basevertex;
                break;
        case nir_intrinsic_load_vertex_id_zero_base:
        case nir_intrinsic_load_vertex_id:
                if (!ctx->vertex_id) {
                        gl_system_value sv = (intr->intrinsic == nir_intrinsic_load_vertex_id) ?
                                SYSTEM_VALUE_VERTEX_ID : SYSTEM_VALUE_VERTEX_ID_ZERO_BASE;
                        ctx->vertex_id = create_input(ctx, 0);
                        add_sysval_input(ctx, sv, ctx->vertex_id);
                }
                dst[0] = ctx->vertex_id;
                break;
        case nir_intrinsic_load_instance_id:
                if (!ctx->instance_id) {
                        ctx->instance_id = create_input(ctx, 0);
                        add_sysval_input(ctx, SYSTEM_VALUE_INSTANCE_ID,
                                        ctx->instance_id);
                }
                dst[0] = ctx->instance_id;
                break;
        case nir_intrinsic_load_sample_id:
        case nir_intrinsic_load_sample_id_no_per_sample:
                if (!ctx->samp_id) {
                        ctx->samp_id = create_input(ctx, 0);
                        ctx->samp_id->regs[0]->flags |= IR3_REG_HALF;
                        add_sysval_input(ctx, SYSTEM_VALUE_SAMPLE_ID,
                                        ctx->samp_id);
                }
                dst[0] = ir3_COV(b, ctx->samp_id, TYPE_U16, TYPE_U32);
                break;
        case nir_intrinsic_load_sample_mask_in:
                if (!ctx->samp_mask_in) {
                        ctx->samp_mask_in = create_input(ctx, 0);
                        add_sysval_input(ctx, SYSTEM_VALUE_SAMPLE_MASK_IN,
                                        ctx->samp_mask_in);
                }
                dst[0] = ctx->samp_mask_in;
                break;
        case nir_intrinsic_load_user_clip_plane:
                idx = nir_intrinsic_ucp_id(intr);
                for (int i = 0; i < intr->num_components; i++) {
                        unsigned n = idx * 4 + i;
                        dst[i] = create_driver_param(ctx, IR3_DP_UCP0_X + n);
                }
                break;
        case nir_intrinsic_load_front_face:
                if (!ctx->frag_face) {
                        ctx->so->frag_face = true;
                        ctx->frag_face = create_input(ctx, 0);
                        add_sysval_input(ctx, SYSTEM_VALUE_FRONT_FACE, ctx->frag_face);
                        ctx->frag_face->regs[0]->flags |= IR3_REG_HALF;
                }
                /* for fragface, we get -1 for back and 0 for front. However this is
                 * the inverse of what nir expects (where ~0 is true).
                 */
                dst[0] = ir3_COV(b, ctx->frag_face, TYPE_S16, TYPE_S32);
                dst[0] = ir3_NOT_B(b, dst[0], 0);
                break;
        case nir_intrinsic_load_local_invocation_id:
                if (!ctx->local_invocation_id) {
                        ctx->local_invocation_id = create_input_compmask(ctx, 0, 0x7);
                        add_sysval_input_compmask(ctx, SYSTEM_VALUE_LOCAL_INVOCATION_ID,
                                        0x7, ctx->local_invocation_id);
                }
                split_dest(b, dst, ctx->local_invocation_id, 0, 3);
                break;
        case nir_intrinsic_load_work_group_id:
                if (!ctx->work_group_id) {
                        ctx->work_group_id = create_input_compmask(ctx, 0, 0x7);
                        add_sysval_input_compmask(ctx, SYSTEM_VALUE_WORK_GROUP_ID,
                                        0x7, ctx->work_group_id);
                        ctx->work_group_id->regs[0]->flags |= IR3_REG_HIGH;
                }
                split_dest(b, dst, ctx->work_group_id, 0, 3);
                break;
        case nir_intrinsic_load_num_work_groups:
                for (int i = 0; i < intr->num_components; i++) {
                        dst[i] = create_driver_param(ctx, IR3_DP_NUM_WORK_GROUPS_X + i);
                }
                break;
        case nir_intrinsic_load_local_group_size:
                for (int i = 0; i < intr->num_components; i++) {
                        dst[i] = create_driver_param(ctx, IR3_DP_LOCAL_GROUP_SIZE_X + i);
                }
                break;
        case nir_intrinsic_discard_if:
        case nir_intrinsic_discard: {
                struct ir3_instruction *cond, *kill;

                if (intr->intrinsic == nir_intrinsic_discard_if) {
                        /* conditional discard: */
                        src = get_src(ctx, &intr->src[0]);
                        cond = ir3_b2n(b, src[0]);
                } else {
                        /* unconditional discard: */
                        cond = create_immed(b, 1);
                }

                /* NOTE: only cmps.*.* can write p0.x: */
                cond = ir3_CMPS_S(b, cond, 0, create_immed(b, 0), 0);
                cond->cat2.condition = IR3_COND_NE;

                /* condition always goes in predicate register: */
                cond->regs[0]->num = regid(REG_P0, 0);

                kill = ir3_KILL(b, cond, 0);
                array_insert(ctx->ir, ctx->ir->predicates, kill);

                array_insert(b, b->keeps, kill);
                ctx->so->has_kill = true;

                break;
        }
        default:
                compile_error(ctx, "Unhandled intrinsic type: %s\n",
                                nir_intrinsic_infos[intr->intrinsic].name);
                break;
        }

        if (info->has_dest)
                put_dst(ctx, &intr->dest);
}

static void
emit_load_const(struct ir3_context *ctx, nir_load_const_instr *instr)
{
        struct ir3_instruction **dst = get_dst_ssa(ctx, &instr->def,
                        instr->def.num_components);
        type_t type = (instr->def.bit_size < 32) ? TYPE_U16 : TYPE_U32;

        for (int i = 0; i < instr->def.num_components; i++)
                dst[i] = create_immed_typed(ctx->block, instr->value.u32[i], type);
}

static void
emit_undef(struct ir3_context *ctx, nir_ssa_undef_instr *undef)
{
        struct ir3_instruction **dst = get_dst_ssa(ctx, &undef->def,
                        undef->def.num_components);
        type_t type = (undef->def.bit_size < 32) ? TYPE_U16 : TYPE_U32;

        /* backend doesn't want undefined instructions, so just plug
         * in 0.0..
         */
        for (int i = 0; i < undef->def.num_components; i++)
                dst[i] = create_immed_typed(ctx->block, fui(0.0), type);
}

/*
 * texture fetch/sample instructions:
 */

static void
tex_info(nir_tex_instr *tex, unsigned *flagsp, unsigned *coordsp)
{
        unsigned coords, flags = 0;

        /* note: would use tex->coord_components.. except txs.. also,
         * since array index goes after shadow ref, we don't want to
         * count it:
         */
        switch (tex->sampler_dim) {
        case GLSL_SAMPLER_DIM_1D:
        case GLSL_SAMPLER_DIM_BUF:
                coords = 1;
                break;
        case GLSL_SAMPLER_DIM_2D:
        case GLSL_SAMPLER_DIM_RECT:
        case GLSL_SAMPLER_DIM_EXTERNAL:
        case GLSL_SAMPLER_DIM_MS:
                coords = 2;
                break;
        case GLSL_SAMPLER_DIM_3D:
        case GLSL_SAMPLER_DIM_CUBE:
                coords = 3;
                flags |= IR3_INSTR_3D;
                break;
        default:
                unreachable("bad sampler_dim");
        }

        if (tex->is_shadow && tex->op != nir_texop_lod)
                flags |= IR3_INSTR_S;

        if (tex->is_array && tex->op != nir_texop_lod)
                flags |= IR3_INSTR_A;

        *flagsp = flags;
        *coordsp = coords;
}

static void
emit_tex(struct ir3_context *ctx, nir_tex_instr *tex)
{
        struct ir3_block *b = ctx->block;
        struct ir3_instruction **dst, *sam, *src0[12], *src1[4];
        struct ir3_instruction * const *coord, * const *off, * const *ddx, * const *ddy;
        struct ir3_instruction *lod, *compare, *proj, *sample_index;
        bool has_bias = false, has_lod = false, has_proj = false, has_off = false;
        unsigned i, coords, flags;
        unsigned nsrc0 = 0, nsrc1 = 0;
        type_t type;
        opc_t opc = 0;

        coord = off = ddx = ddy = NULL;
        lod = proj = compare = sample_index = NULL;

        /* TODO: might just be one component for gathers? */
        dst = get_dst(ctx, &tex->dest, 4);

        for (unsigned i = 0; i < tex->num_srcs; i++) {
                switch (tex->src[i].src_type) {
                case nir_tex_src_coord:
                        coord = get_src(ctx, &tex->src[i].src);
                        break;
                case nir_tex_src_bias:
                        lod = get_src(ctx, &tex->src[i].src)[0];
                        has_bias = true;
                        break;
                case nir_tex_src_lod:
                        lod = get_src(ctx, &tex->src[i].src)[0];
                        has_lod = true;
                        break;
                case nir_tex_src_comparator: /* shadow comparator */
                        compare = get_src(ctx, &tex->src[i].src)[0];
                        break;
                case nir_tex_src_projector:
                        proj = get_src(ctx, &tex->src[i].src)[0];
                        has_proj = true;
                        break;
                case nir_tex_src_offset:
                        off = get_src(ctx, &tex->src[i].src);
                        has_off = true;
                        break;
                case nir_tex_src_ddx:
                        ddx = get_src(ctx, &tex->src[i].src);
                        break;
                case nir_tex_src_ddy:
                        ddy = get_src(ctx, &tex->src[i].src);
                        break;
                case nir_tex_src_ms_index:
                        sample_index = get_src(ctx, &tex->src[i].src)[0];
                        break;
                default:
                        compile_error(ctx, "Unhandled NIR tex src type: %d\n",
                                        tex->src[i].src_type);
                        return;
                }
        }

        switch (tex->op) {
        case nir_texop_tex:      opc = has_lod ? OPC_SAML : OPC_SAM; break;
        case nir_texop_txb:      opc = OPC_SAMB;     break;
        case nir_texop_txl:      opc = OPC_SAML;     break;
        case nir_texop_txd:      opc = OPC_SAMGQ;    break;
        case nir_texop_txf:      opc = OPC_ISAML;    break;
        case nir_texop_lod:      opc = OPC_GETLOD;   break;
        case nir_texop_tg4:
                /* NOTE: a4xx might need to emulate gather w/ txf (this is
                 * what blob does, seems gather  is broken?), and a3xx did
                 * not support it (but probably could also emulate).
                 */
                switch (tex->component) {
                case 0:              opc = OPC_GATHER4R; break;
                case 1:              opc = OPC_GATHER4G; break;
                case 2:              opc = OPC_GATHER4B; break;
                case 3:              opc = OPC_GATHER4A; break;
                }
                break;
        case nir_texop_txf_ms:   opc = OPC_ISAMM;    break;
        case nir_texop_txs:
        case nir_texop_query_levels:
        case nir_texop_texture_samples:
        case nir_texop_samples_identical:
        case nir_texop_txf_ms_mcs:
                compile_error(ctx, "Unhandled NIR tex type: %d\n", tex->op);
                return;
        }

        tex_info(tex, &flags, &coords);

        /*
         * lay out the first argument in the proper order:
         *  - actual coordinates first
         *  - shadow reference
         *  - array index
         *  - projection w
         *  - starting at offset 4, dpdx.xy, dpdy.xy
         *
         * bias/lod go into the second arg
         */

        /* insert tex coords: */
        for (i = 0; i < coords; i++)
                src0[i] = coord[i];

        nsrc0 = i;

        /* NOTE a3xx (and possibly a4xx?) might be different, using isaml
         * with scaled x coord according to requested sample:
         */
        if (tex->op == nir_texop_txf_ms) {
                if (ctx->compiler->txf_ms_with_isaml) {
                        /* the samples are laid out in x dimension as
                         *     0 1 2 3
                         * x_ms = (x << ms) + sample_index;
                         */
                        struct ir3_instruction *ms;
                        ms = create_immed(b, (ctx->samples >> (2 * tex->texture_index)) & 3);

                        src0[0] = ir3_SHL_B(b, src0[0], 0, ms, 0);
                        src0[0] = ir3_ADD_U(b, src0[0], 0, sample_index, 0);

                        opc = OPC_ISAML;
                } else {
                        src0[nsrc0++] = sample_index;
                }
        }

        /* scale up integer coords for TXF based on the LOD */
        if (ctx->compiler->unminify_coords && (opc == OPC_ISAML)) {
                assert(has_lod);
                for (i = 0; i < coords; i++)
                        src0[i] = ir3_SHL_B(b, src0[i], 0, lod, 0);
        }

        if (coords == 1) {
                /* hw doesn't do 1d, so we treat it as 2d with
                 * height of 1, and patch up the y coord.
                 * TODO: y coord should be (int)0 in some cases..
                 */
                src0[nsrc0++] = create_immed(b, fui(0.5));
        }

        if (tex->is_shadow && tex->op != nir_texop_lod)
                src0[nsrc0++] = compare;

        if (tex->is_array && tex->op != nir_texop_lod) {
                struct ir3_instruction *idx = coord[coords];

                /* the array coord for cube arrays needs 0.5 added to it */
                if (ctx->compiler->array_index_add_half && (opc != OPC_ISAML))
                        idx = ir3_ADD_F(b, idx, 0, create_immed(b, fui(0.5)), 0);

                src0[nsrc0++] = idx;
        }

        if (has_proj) {
                src0[nsrc0++] = proj;
                flags |= IR3_INSTR_P;
        }

        /* pad to 4, then ddx/ddy: */
        if (tex->op == nir_texop_txd) {
                while (nsrc0 < 4)
                        src0[nsrc0++] = create_immed(b, fui(0.0));
                for (i = 0; i < coords; i++)
                        src0[nsrc0++] = ddx[i];
                if (coords < 2)
                        src0[nsrc0++] = create_immed(b, fui(0.0));
                for (i = 0; i < coords; i++)
                        src0[nsrc0++] = ddy[i];
                if (coords < 2)
                        src0[nsrc0++] = create_immed(b, fui(0.0));
        }

        /*
         * second argument (if applicable):
         *  - offsets
         *  - lod
         *  - bias
         */
        if (has_off | has_lod | has_bias) {
                if (has_off) {
                        unsigned off_coords = coords;
                        if (tex->sampler_dim == GLSL_SAMPLER_DIM_CUBE)
                                off_coords--;
                        for (i = 0; i < off_coords; i++)
                                src1[nsrc1++] = off[i];
                        if (off_coords < 2)
                                src1[nsrc1++] = create_immed(b, fui(0.0));
                        flags |= IR3_INSTR_O;
                }

                if (has_lod | has_bias)
                        src1[nsrc1++] = lod;
        }

        switch (tex->dest_type) {
        case nir_type_invalid:
        case nir_type_float:
                type = TYPE_F32;
                break;
        case nir_type_int:
                type = TYPE_S32;
                break;
        case nir_type_uint:
        case nir_type_bool:
                type = TYPE_U32;
                break;
        default:
                unreachable("bad dest_type");
        }

        if (opc == OPC_GETLOD)
                type = TYPE_U32;

        unsigned tex_idx = tex->texture_index;

        ctx->max_texture_index = MAX2(ctx->max_texture_index, tex_idx);

        struct ir3_instruction *col0 = create_collect(ctx, src0, nsrc0);
        struct ir3_instruction *col1 = create_collect(ctx, src1, nsrc1);

        sam = ir3_SAM(b, opc, type, 0b1111, flags,
                        tex_idx, tex_idx, col0, col1);

        if ((ctx->astc_srgb & (1 << tex_idx)) && !nir_tex_instr_is_query(tex)) {
                /* only need first 3 components: */
                sam->regs[0]->wrmask = 0x7;
                split_dest(b, dst, sam, 0, 3);

                /* we need to sample the alpha separately with a non-ASTC
                 * texture state:
                 */
                sam = ir3_SAM(b, opc, type, 0b1000, flags,
                                tex_idx, tex_idx, col0, col1);

                array_insert(ctx->ir, ctx->ir->astc_srgb, sam);

                /* fixup .w component: */
                split_dest(b, &dst[3], sam, 3, 1);
        } else {
                /* normal (non-workaround) case: */
                split_dest(b, dst, sam, 0, 4);
        }

        /* GETLOD returns results in 4.8 fixed point */
        if (opc == OPC_GETLOD) {
                struct ir3_instruction *factor = create_immed(b, fui(1.0 / 256));

                compile_assert(ctx, tex->dest_type == nir_type_float);
                for (i = 0; i < 2; i++) {
                        dst[i] = ir3_MUL_F(b, ir3_COV(b, dst[i], TYPE_U32, TYPE_F32), 0,
                                                           factor, 0);
                }
        }

        put_dst(ctx, &tex->dest);
}

static void
emit_tex_query_levels(struct ir3_context *ctx, nir_tex_instr *tex)
{
        struct ir3_block *b = ctx->block;
        struct ir3_instruction **dst, *sam;

        dst = get_dst(ctx, &tex->dest, 1);

        sam = ir3_SAM(b, OPC_GETINFO, TYPE_U32, 0b0100, 0,
                        tex->texture_index, tex->texture_index, NULL, NULL);

        /* even though there is only one component, since it ends
         * up in .z rather than .x, we need a split_dest()
         */
        split_dest(b, dst, sam, 0, 3);

        /* The # of levels comes from getinfo.z. We need to add 1 to it, since
         * the value in TEX_CONST_0 is zero-based.
         */
        if (ctx->compiler->levels_add_one)
                dst[0] = ir3_ADD_U(b, dst[0], 0, create_immed(b, 1), 0);

        put_dst(ctx, &tex->dest);
}

static void
emit_tex_txs(struct ir3_context *ctx, nir_tex_instr *tex)
{
        struct ir3_block *b = ctx->block;
        struct ir3_instruction **dst, *sam;
        struct ir3_instruction *lod;
        unsigned flags, coords;

        tex_info(tex, &flags, &coords);

        /* Actually we want the number of dimensions, not coordinates. This
         * distinction only matters for cubes.
         */
        if (tex->sampler_dim == GLSL_SAMPLER_DIM_CUBE)
                coords = 2;

        dst = get_dst(ctx, &tex->dest, 4);

        compile_assert(ctx, tex->num_srcs == 1);
        compile_assert(ctx, tex->src[0].src_type == nir_tex_src_lod);

        lod = get_src(ctx, &tex->src[0].src)[0];

        sam = ir3_SAM(b, OPC_GETSIZE, TYPE_U32, 0b1111, flags,
                        tex->texture_index, tex->texture_index, lod, NULL);

        split_dest(b, dst, sam, 0, 4);

        /* Array size actually ends up in .w rather than .z. This doesn't
         * matter for miplevel 0, but for higher mips the value in z is
         * minified whereas w stays. Also, the value in TEX_CONST_3_DEPTH is
         * returned, which means that we have to add 1 to it for arrays.
         */
        if (tex->is_array) {
                if (ctx->compiler->levels_add_one) {
                        dst[coords] = ir3_ADD_U(b, dst[3], 0, create_immed(b, 1), 0);
                } else {
                        dst[coords] = ir3_MOV(b, dst[3], TYPE_U32);
                }
        }

        put_dst(ctx, &tex->dest);
}

static void
emit_jump(struct ir3_context *ctx, nir_jump_instr *jump)
{
        switch (jump->type) {
        case nir_jump_break:
        case nir_jump_continue:
        case nir_jump_return:
                /* I *think* we can simply just ignore this, and use the
                 * successor block link to figure out where we need to
                 * jump to for break/continue
                 */
                break;
        default:
                compile_error(ctx, "Unhandled NIR jump type: %d\n", jump->type);
                break;
        }
}

static void
emit_instr(struct ir3_context *ctx, nir_instr *instr)
{
        switch (instr->type) {
        case nir_instr_type_alu:
                emit_alu(ctx, nir_instr_as_alu(instr));
                break;
        case nir_instr_type_deref:
                /* ignored, handled as part of the intrinsic they are src to */
                break;
        case nir_instr_type_intrinsic:
                emit_intrinsic(ctx, nir_instr_as_intrinsic(instr));
                break;
        case nir_instr_type_load_const:
                emit_load_const(ctx, nir_instr_as_load_const(instr));
                break;
        case nir_instr_type_ssa_undef:
                emit_undef(ctx, nir_instr_as_ssa_undef(instr));
                break;
        case nir_instr_type_tex: {
                nir_tex_instr *tex = nir_instr_as_tex(instr);
                /* couple tex instructions get special-cased:
                 */
                switch (tex->op) {
                case nir_texop_txs:
                        emit_tex_txs(ctx, tex);
                        break;
                case nir_texop_query_levels:
                        emit_tex_query_levels(ctx, tex);
                        break;
                default:
                        emit_tex(ctx, tex);
                        break;
                }
                break;
        }
        case nir_instr_type_jump:
                emit_jump(ctx, nir_instr_as_jump(instr));
                break;
        case nir_instr_type_phi:
                /* we have converted phi webs to regs in NIR by now */
                compile_error(ctx, "Unexpected NIR instruction type: %d\n", instr->type);
                break;
        case nir_instr_type_call:
        case nir_instr_type_parallel_copy:
                compile_error(ctx, "Unhandled NIR instruction type: %d\n", instr->type);
                break;
        }
}

static struct ir3_block *
get_block(struct ir3_context *ctx, const nir_block *nblock)
{
        struct ir3_block *block;
        struct hash_entry *hentry;
        unsigned i;

        hentry = _mesa_hash_table_search(ctx->block_ht, nblock);
        if (hentry)
                return hentry->data;

        block = ir3_block_create(ctx->ir);
        block->nblock = nblock;
        _mesa_hash_table_insert(ctx->block_ht, nblock, block);

        block->predecessors_count = nblock->predecessors->entries;
        block->predecessors = ralloc_array_size(block,
                sizeof(block->predecessors[0]), block->predecessors_count);
        i = 0;
        set_foreach(nblock->predecessors, sentry) {
                block->predecessors[i++] = get_block(ctx, sentry->key);
        }

        return block;
}

static void
emit_block(struct ir3_context *ctx, nir_block *nblock)
{
        struct ir3_block *block = get_block(ctx, nblock);

        for (int i = 0; i < ARRAY_SIZE(block->successors); i++) {
                if (nblock->successors[i]) {
                        block->successors[i] =
                                get_block(ctx, nblock->successors[i]);
                }
        }

        ctx->block = block;
        list_addtail(&block->node, &ctx->ir->block_list);

        /* re-emit addr register in each block if needed: */
        for (int i = 0; i < ARRAY_SIZE(ctx->addr_ht); i++) {
                _mesa_hash_table_destroy(ctx->addr_ht[i], NULL);
                ctx->addr_ht[i] = NULL;
        }

        nir_foreach_instr(instr, nblock) {
                ctx->cur_instr = instr;
                emit_instr(ctx, instr);
                ctx->cur_instr = NULL;
                if (ctx->error)
                        return;
        }
}

static void emit_cf_list(struct ir3_context *ctx, struct exec_list *list);

static void
emit_if(struct ir3_context *ctx, nir_if *nif)
{
        struct ir3_instruction *condition = get_src(ctx, &nif->condition)[0];

        ctx->block->condition =
                get_predicate(ctx, ir3_b2n(condition->block, condition));

        emit_cf_list(ctx, &nif->then_list);
        emit_cf_list(ctx, &nif->else_list);
}

static void
emit_loop(struct ir3_context *ctx, nir_loop *nloop)
{
        emit_cf_list(ctx, &nloop->body);
}

static void
emit_cf_list(struct ir3_context *ctx, struct exec_list *list)
{
        foreach_list_typed(nir_cf_node, node, node, list) {
                switch (node->type) {
                case nir_cf_node_block:
                        emit_block(ctx, nir_cf_node_as_block(node));
                        break;
                case nir_cf_node_if:
                        emit_if(ctx, nir_cf_node_as_if(node));
                        break;
                case nir_cf_node_loop:
                        emit_loop(ctx, nir_cf_node_as_loop(node));
                        break;
                case nir_cf_node_function:
                        compile_error(ctx, "TODO\n");
                        break;
                }
        }
}

/* emit stream-out code.  At this point, the current block is the original
 * (nir) end block, and nir ensures that all flow control paths terminate
 * into the end block.  We re-purpose the original end block to generate
 * the 'if (vtxcnt < maxvtxcnt)' condition, then append the conditional
 * block holding stream-out write instructions, followed by the new end
 * block:
 *
 *   blockOrigEnd {
 *      p0.x = (vtxcnt < maxvtxcnt)
 *      // succs: blockStreamOut, blockNewEnd
 *   }
 *   blockStreamOut {
 *      ... stream-out instructions ...
 *      // succs: blockNewEnd
 *   }
 *   blockNewEnd {
 *   }
 */
static void
emit_stream_out(struct ir3_context *ctx)
{
        struct ir3_shader_variant *v = ctx->so;
        struct ir3 *ir = ctx->ir;
        struct ir3_stream_output_info *strmout =
                        &ctx->so->shader->stream_output;
        struct ir3_block *orig_end_block, *stream_out_block, *new_end_block;
        struct ir3_instruction *vtxcnt, *maxvtxcnt, *cond;
        struct ir3_instruction *bases[IR3_MAX_SO_BUFFERS];

        /* create vtxcnt input in input block at top of shader,
         * so that it is seen as live over the entire duration
         * of the shader:
         */
        vtxcnt = create_input(ctx, 0);
        add_sysval_input(ctx, SYSTEM_VALUE_VERTEX_CNT, vtxcnt);

        maxvtxcnt = create_driver_param(ctx, IR3_DP_VTXCNT_MAX);

        /* at this point, we are at the original 'end' block,
         * re-purpose this block to stream-out condition, then
         * append stream-out block and new-end block
         */
        orig_end_block = ctx->block;

// TODO these blocks need to update predecessors..
// maybe w/ store_global intrinsic, we could do this
// stuff in nir->nir pass

        stream_out_block = ir3_block_create(ir);
        list_addtail(&stream_out_block->node, &ir->block_list);

        new_end_block = ir3_block_create(ir);
        list_addtail(&new_end_block->node, &ir->block_list);

        orig_end_block->successors[0] = stream_out_block;
        orig_end_block->successors[1] = new_end_block;
        stream_out_block->successors[0] = new_end_block;

        /* setup 'if (vtxcnt < maxvtxcnt)' condition: */
        cond = ir3_CMPS_S(ctx->block, vtxcnt, 0, maxvtxcnt, 0);
        cond->regs[0]->num = regid(REG_P0, 0);
        cond->cat2.condition = IR3_COND_LT;

        /* condition goes on previous block to the conditional,
         * since it is used to pick which of the two successor
         * paths to take:
         */
        orig_end_block->condition = cond;

        /* switch to stream_out_block to generate the stream-out
         * instructions:
         */
        ctx->block = stream_out_block;

        /* Calculate base addresses based on vtxcnt.  Instructions
         * generated for bases not used in following loop will be
         * stripped out in the backend.
         */
        for (unsigned i = 0; i < IR3_MAX_SO_BUFFERS; i++) {
                unsigned stride = strmout->stride[i];
                struct ir3_instruction *base, *off;

                base = create_uniform(ctx, regid(v->constbase.tfbo, i));

                /* 24-bit should be enough: */
                off = ir3_MUL_U(ctx->block, vtxcnt, 0,
                                create_immed(ctx->block, stride * 4), 0);

                bases[i] = ir3_ADD_S(ctx->block, off, 0, base, 0);
        }

        /* Generate the per-output store instructions: */
        for (unsigned i = 0; i < strmout->num_outputs; i++) {
                for (unsigned j = 0; j < strmout->output[i].num_components; j++) {
                        unsigned c = j + strmout->output[i].start_component;
                        struct ir3_instruction *base, *out, *stg;

                        base = bases[strmout->output[i].output_buffer];
                        out = ctx->ir->outputs[regid(strmout->output[i].register_index, c)];

                        stg = ir3_STG(ctx->block, base, 0, out, 0,
                                        create_immed(ctx->block, 1), 0);
                        stg->cat6.type = TYPE_U32;
                        stg->cat6.dst_offset = (strmout->output[i].dst_offset + j) * 4;

                        array_insert(ctx->block, ctx->block->keeps, stg);
                }
        }

        /* and finally switch to the new_end_block: */
        ctx->block = new_end_block;
}

static void
emit_function(struct ir3_context *ctx, nir_function_impl *impl)
{
        nir_metadata_require(impl, nir_metadata_block_index);

        emit_cf_list(ctx, &impl->body);
        emit_block(ctx, impl->end_block);

        /* at this point, we should have a single empty block,
         * into which we emit the 'end' instruction.
         */
        compile_assert(ctx, list_empty(&ctx->block->instr_list));

        /* If stream-out (aka transform-feedback) enabled, emit the
         * stream-out instructions, followed by a new empty block (into
         * which the 'end' instruction lands).
         *
         * NOTE: it is done in this order, rather than inserting before
         * we emit end_block, because NIR guarantees that all blocks
         * flow into end_block, and that end_block has no successors.
         * So by re-purposing end_block as the first block of stream-
         * out, we guarantee that all exit paths flow into the stream-
         * out instructions.
         */
        if ((ctx->compiler->gpu_id < 500) &&
                        (ctx->so->shader->stream_output.num_outputs > 0) &&
                        !ctx->so->binning_pass) {
                debug_assert(ctx->so->type == MESA_SHADER_VERTEX);
                emit_stream_out(ctx);
        }

        ir3_END(ctx->block);
}

static struct ir3_instruction *
create_frag_coord(struct ir3_context *ctx, unsigned comp)
{
        struct ir3_block *block = ctx->block;
        struct ir3_instruction *instr;

        if (!ctx->frag_coord) {
                ctx->frag_coord = create_input_compmask(ctx, 0, 0xf);
                /* defer add_sysval_input() until after all inputs created */
        }

        split_dest(block, &instr, ctx->frag_coord, comp, 1);

        switch (comp) {
        case 0: /* .x */
        case 1: /* .y */
                /* for frag_coord, we get unsigned values.. we need
                 * to subtract (integer) 8 and divide by 16 (right-
                 * shift by 4) then convert to float:
                 *
                 *    sub.s tmp, src, 8
                 *    shr.b tmp, tmp, 4
                 *    mov.u32f32 dst, tmp
                 *
                 */
                instr = ir3_SUB_S(block, instr, 0,
                                create_immed(block, 8), 0);
                instr = ir3_SHR_B(block, instr, 0,
                                create_immed(block, 4), 0);
                instr = ir3_COV(block, instr, TYPE_U32, TYPE_F32);

                return instr;
        case 2: /* .z */
        case 3: /* .w */
        default:
                /* seems that we can use these as-is: */
                return instr;
        }
}

static void
setup_input(struct ir3_context *ctx, nir_variable *in)
{
        struct ir3_shader_variant *so = ctx->so;
        unsigned ncomp = glsl_get_components(in->type);
        unsigned n = in->data.driver_location;
        unsigned slot = in->data.location;

        /* let's pretend things other than vec4 don't exist: */
        ncomp = MAX2(ncomp, 4);

        /* skip unread inputs, we could end up with (for example), unsplit
         * matrix/etc inputs in the case they are not read, so just silently
         * skip these.
         */
        if (ncomp > 4)
                return;

        compile_assert(ctx, ncomp == 4);

        so->inputs[n].slot = slot;
        so->inputs[n].compmask = (1 << ncomp) - 1;
        so->inputs_count = MAX2(so->inputs_count, n + 1);
        so->inputs[n].interpolate = in->data.interpolation;

        if (ctx->so->type == MESA_SHADER_FRAGMENT) {
                for (int i = 0; i < ncomp; i++) {
                        struct ir3_instruction *instr = NULL;
                        unsigned idx = (n * 4) + i;

                        if (slot == VARYING_SLOT_POS) {
                                so->inputs[n].bary = false;
                                so->frag_coord = true;
                                instr = create_frag_coord(ctx, i);
                        } else if (slot == VARYING_SLOT_PNTC) {
                                /* see for example st_nir_fixup_varying_slots().. this is
                                 * maybe a bit mesa/st specific.  But we need things to line
                                 * up for this in fdN_program:
                                 *    unsigned texmask = 1 << (slot - VARYING_SLOT_VAR0);
                                 *    if (emit->sprite_coord_enable & texmask) {
                                 *       ...
                                 *    }
                                 */
                                so->inputs[n].slot = VARYING_SLOT_VAR8;
                                so->inputs[n].bary = true;
                                instr = create_frag_input(ctx, false);
                        } else {
                                bool use_ldlv = false;

                                /* detect the special case for front/back colors where
                                 * we need to do flat vs smooth shading depending on
                                 * rast state:
                                 */
                                if (in->data.interpolation == INTERP_MODE_NONE) {
                                        switch (slot) {
                                        case VARYING_SLOT_COL0:
                                        case VARYING_SLOT_COL1:
                                        case VARYING_SLOT_BFC0:
                                        case VARYING_SLOT_BFC1:
                                                so->inputs[n].rasterflat = true;
                                                break;
                                        default:
                                                break;
                                        }
                                }

                                if (ctx->compiler->flat_bypass) {
                                        if ((so->inputs[n].interpolate == INTERP_MODE_FLAT) ||
                                                        (so->inputs[n].rasterflat && ctx->so->key.rasterflat))
                                                use_ldlv = true;
                                }

                                so->inputs[n].bary = true;

                                instr = create_frag_input(ctx, use_ldlv);
                        }

                        compile_assert(ctx, idx < ctx->ir->ninputs);

                        ctx->ir->inputs[idx] = instr;
                }
        } else if (ctx->so->type == MESA_SHADER_VERTEX) {
                for (int i = 0; i < ncomp; i++) {
                        unsigned idx = (n * 4) + i;
                        compile_assert(ctx, idx < ctx->ir->ninputs);
                        ctx->ir->inputs[idx] = create_input(ctx, idx);
                }
        } else {
                compile_error(ctx, "unknown shader type: %d\n", ctx->so->type);
        }

        if (so->inputs[n].bary || (ctx->so->type == MESA_SHADER_VERTEX)) {
                so->total_in += ncomp;
        }
}

static void
setup_output(struct ir3_context *ctx, nir_variable *out)
{
        struct ir3_shader_variant *so = ctx->so;
        unsigned ncomp = glsl_get_components(out->type);
        unsigned n = out->data.driver_location;
        unsigned slot = out->data.location;
        unsigned comp = 0;

        /* let's pretend things other than vec4 don't exist: */
        ncomp = MAX2(ncomp, 4);
        compile_assert(ctx, ncomp == 4);

        if (ctx->so->type == MESA_SHADER_FRAGMENT) {
                switch (slot) {
                case FRAG_RESULT_DEPTH:
                        comp = 2;  /* tgsi will write to .z component */
                        so->writes_pos = true;
                        break;
                case FRAG_RESULT_COLOR:
                        so->color0_mrt = 1;
                        break;
                default:
                        if (slot >= FRAG_RESULT_DATA0)
                                break;
                        compile_error(ctx, "unknown FS output name: %s\n",
                                        gl_frag_result_name(slot));
                }
        } else if (ctx->so->type == MESA_SHADER_VERTEX) {
                switch (slot) {
                case VARYING_SLOT_POS:
                        so->writes_pos = true;
                        break;
                case VARYING_SLOT_PSIZ:
                        so->writes_psize = true;
                        break;
                case VARYING_SLOT_COL0:
                case VARYING_SLOT_COL1:
                case VARYING_SLOT_BFC0:
                case VARYING_SLOT_BFC1:
                case VARYING_SLOT_FOGC:
                case VARYING_SLOT_CLIP_DIST0:
                case VARYING_SLOT_CLIP_DIST1:
                case VARYING_SLOT_CLIP_VERTEX:
                        break;
                default:
                        if (slot >= VARYING_SLOT_VAR0)
                                break;
                        if ((VARYING_SLOT_TEX0 <= slot) && (slot <= VARYING_SLOT_TEX7))
                                break;
                        compile_error(ctx, "unknown VS output name: %s\n",
                                        gl_varying_slot_name(slot));
                }
        } else {
                compile_error(ctx, "unknown shader type: %d\n", ctx->so->type);
        }

        compile_assert(ctx, n < ARRAY_SIZE(so->outputs));

        so->outputs[n].slot = slot;
        so->outputs[n].regid = regid(n, comp);
        so->outputs_count = MAX2(so->outputs_count, n + 1);

        for (int i = 0; i < ncomp; i++) {
                unsigned idx = (n * 4) + i;
                compile_assert(ctx, idx < ctx->ir->noutputs);
                ctx->ir->outputs[idx] = create_immed(ctx->block, fui(0.0));
        }
}

static int
max_drvloc(struct exec_list *vars)
{
        int drvloc = -1;
        nir_foreach_variable(var, vars) {
                drvloc = MAX2(drvloc, (int)var->data.driver_location);
        }
        return drvloc;
}

static const unsigned max_sysvals[] = {
        [MESA_SHADER_FRAGMENT] = 24,  // TODO
        [MESA_SHADER_VERTEX]  = 16,
        [MESA_SHADER_COMPUTE] = 16, // TODO how many do we actually need?
};

static void
emit_instructions(struct ir3_context *ctx)
{
        unsigned ninputs, noutputs;
        nir_function_impl *fxn = nir_shader_get_entrypoint(ctx->s);

        ninputs  = (max_drvloc(&ctx->s->inputs) + 1) * 4;
        noutputs = (max_drvloc(&ctx->s->outputs) + 1) * 4;

        /* we need to leave room for sysvals:
         */
        ninputs += max_sysvals[ctx->so->type];

        ctx->ir = ir3_create(ctx->compiler, ninputs, noutputs);

        /* Create inputs in first block: */
        ctx->block = get_block(ctx, nir_start_block(fxn));
        ctx->in_block = ctx->block;
        list_addtail(&ctx->block->node, &ctx->ir->block_list);

        ninputs -= max_sysvals[ctx->so->type];

        /* for fragment shader, the vcoord input register is used as the
         * base for bary.f varying fetch instrs:
         */
        struct ir3_instruction *vcoord = NULL;
        if (ctx->so->type == MESA_SHADER_FRAGMENT) {
                struct ir3_instruction *xy[2];

                vcoord = create_input_compmask(ctx, 0, 0x3);
                split_dest(ctx->block, xy, vcoord, 0, 2);

                ctx->frag_vcoord = create_collect(ctx, xy, 2);
        }

        /* Setup inputs: */
        nir_foreach_variable(var, &ctx->s->inputs) {
                setup_input(ctx, var);
        }

        /* Defer add_sysval_input() stuff until after setup_inputs(),
         * because sysvals need to be appended after varyings:
         */
        if (vcoord) {
                add_sysval_input_compmask(ctx, SYSTEM_VALUE_VARYING_COORD,
                                0x3, vcoord);
        }

        if (ctx->frag_coord) {
                add_sysval_input_compmask(ctx, SYSTEM_VALUE_FRAG_COORD,
                                0xf, ctx->frag_coord);
        }

        /* Setup outputs: */
        nir_foreach_variable(var, &ctx->s->outputs) {
                setup_output(ctx, var);
        }

        /* Setup registers (which should only be arrays): */
        nir_foreach_register(reg, &ctx->s->registers) {
                declare_array(ctx, reg);
        }

        /* NOTE: need to do something more clever when we support >1 fxn */
        nir_foreach_register(reg, &fxn->registers) {
                declare_array(ctx, reg);
        }
        /* And emit the body: */
        ctx->impl = fxn;
        emit_function(ctx, fxn);
}

/* from NIR perspective, we actually have varying inputs.  But the varying
 * inputs, from an IR standpoint, are just bary.f/ldlv instructions.  The
 * only actual inputs are the sysvals.
 */
static void
fixup_frag_inputs(struct ir3_context *ctx)
{
        struct ir3_shader_variant *so = ctx->so;
        struct ir3 *ir = ctx->ir;
        unsigned i = 0;

        /* sysvals should appear at the end of the inputs, drop everything else: */
        while ((i < so->inputs_count) && !so->inputs[i].sysval)
                i++;

        /* at IR level, inputs are always blocks of 4 scalars: */
        i *= 4;

        ir->inputs = &ir->inputs[i];
        ir->ninputs -= i;
}

/* Fixup tex sampler state for astc/srgb workaround instructions.  We
 * need to assign the tex state indexes for these after we know the
 * max tex index.
 */
static void
fixup_astc_srgb(struct ir3_context *ctx)
{
        struct ir3_shader_variant *so = ctx->so;
        /* indexed by original tex idx, value is newly assigned alpha sampler
         * state tex idx.  Zero is invalid since there is at least one sampler
         * if we get here.
         */
        unsigned alt_tex_state[16] = {0};
        unsigned tex_idx = ctx->max_texture_index + 1;
        unsigned idx = 0;

        so->astc_srgb.base = tex_idx;

        for (unsigned i = 0; i < ctx->ir->astc_srgb_count; i++) {
                struct ir3_instruction *sam = ctx->ir->astc_srgb[i];

                compile_assert(ctx, sam->cat5.tex < ARRAY_SIZE(alt_tex_state));

                if (alt_tex_state[sam->cat5.tex] == 0) {
                        /* assign new alternate/alpha tex state slot: */
                        alt_tex_state[sam->cat5.tex] = tex_idx++;
                        so->astc_srgb.orig_idx[idx++] = sam->cat5.tex;
                        so->astc_srgb.count++;
                }

                sam->cat5.tex = alt_tex_state[sam->cat5.tex];
        }
}

static void
fixup_binning_pass(struct ir3_context *ctx)
{
        struct ir3_shader_variant *so = ctx->so;
        struct ir3 *ir = ctx->ir;
        unsigned i, j;

        for (i = 0, j = 0; i < so->outputs_count; i++) {
                unsigned slot = so->outputs[i].slot;

                /* throw away everything but first position/psize */
                if ((slot == VARYING_SLOT_POS) || (slot == VARYING_SLOT_PSIZ)) {
                        if (i != j) {
                                so->outputs[j] = so->outputs[i];
                                ir->outputs[(j*4)+0] = ir->outputs[(i*4)+0];
                                ir->outputs[(j*4)+1] = ir->outputs[(i*4)+1];
                                ir->outputs[(j*4)+2] = ir->outputs[(i*4)+2];
                                ir->outputs[(j*4)+3] = ir->outputs[(i*4)+3];
                        }
                        j++;
                }
        }
        so->outputs_count = j;
        ir->noutputs = j * 4;
}

int
ir3_compile_shader_nir(struct ir3_compiler *compiler,
                struct ir3_shader_variant *so)
{
        struct ir3_context *ctx;
        struct ir3 *ir;
        struct ir3_instruction **inputs;
        unsigned i, actual_in, inloc;
        int ret = 0, max_bary;

        assert(!so->ir);

        ctx = compile_init(compiler, so);
        if (!ctx) {
                DBG("INIT failed!");
                ret = -1;
                goto out;
        }

        emit_instructions(ctx);

        if (ctx->error) {
                DBG("EMIT failed!");
                ret = -1;
                goto out;
        }

        ir = so->ir = ctx->ir;

        /* keep track of the inputs from TGSI perspective.. */
        inputs = ir->inputs;

        /* but fixup actual inputs for frag shader: */
        if (so->type == MESA_SHADER_FRAGMENT)
                fixup_frag_inputs(ctx);

        /* at this point, for binning pass, throw away unneeded outputs: */
        if (so->binning_pass && (ctx->compiler->gpu_id < 600))
                fixup_binning_pass(ctx);

        /* if we want half-precision outputs, mark the output registers
         * as half:
         */
        if (so->key.half_precision) {
                for (i = 0; i < ir->noutputs; i++) {
                        struct ir3_instruction *out = ir->outputs[i];

                        if (!out)
                                continue;

                        /* if frag shader writes z, that needs to be full precision: */
                        if (so->outputs[i/4].slot == FRAG_RESULT_DEPTH)
                                continue;

                        out->regs[0]->flags |= IR3_REG_HALF;
                        /* output could be a fanout (ie. texture fetch output)
                         * in which case we need to propagate the half-reg flag
                         * up to the definer so that RA sees it:
                         */
                        if (out->opc == OPC_META_FO) {
                                out = out->regs[1]->instr;
                                out->regs[0]->flags |= IR3_REG_HALF;
                        }

                        if (out->opc == OPC_MOV) {
                                out->cat1.dst_type = half_type(out->cat1.dst_type);
                        }
                }
        }

        if (ir3_shader_debug & IR3_DBG_OPTMSGS) {
                printf("BEFORE CP:\n");
                ir3_print(ir);
        }

        ir3_cp(ir, so);

        /* at this point, for binning pass, throw away unneeded outputs:
         * Note that for a6xx and later, we do this after ir3_cp to ensure
         * that the uniform/constant layout for BS and VS matches, so that
         * we can re-use same VS_CONST state group.
         */
        if (so->binning_pass && (ctx->compiler->gpu_id >= 600))
                fixup_binning_pass(ctx);

        /* Insert mov if there's same instruction for each output.
         * eg. dEQP-GLES31.functional.shaders.opaque_type_indexing.sampler.const_expression.vertex.sampler2dshadow
         */
        for (int i = ir->noutputs - 1; i >= 0; i--) {
                if (!ir->outputs[i])
                        continue;
                for (unsigned j = 0; j < i; j++) {
                        if (ir->outputs[i] == ir->outputs[j]) {
                                ir->outputs[i] =
                                        ir3_MOV(ir->outputs[i]->block, ir->outputs[i], TYPE_F32);
                        }
                }
        }

        if (ir3_shader_debug & IR3_DBG_OPTMSGS) {
                printf("BEFORE GROUPING:\n");
                ir3_print(ir);
        }

        ir3_sched_add_deps(ir);

        /* Group left/right neighbors, inserting mov's where needed to
         * solve conflicts:
         */
        ir3_group(ir);

        if (ir3_shader_debug & IR3_DBG_OPTMSGS) {
                printf("AFTER GROUPING:\n");
                ir3_print(ir);
        }

        ir3_depth(ir);

        if (ir3_shader_debug & IR3_DBG_OPTMSGS) {
                printf("AFTER DEPTH:\n");
                ir3_print(ir);
        }

        ret = ir3_sched(ir);
        if (ret) {
                DBG("SCHED failed!");
                goto out;
        }

        if (ir3_shader_debug & IR3_DBG_OPTMSGS) {
                printf("AFTER SCHED:\n");
                ir3_print(ir);
        }

        ret = ir3_ra(ir, so->type, so->frag_coord, so->frag_face);
        if (ret) {
                DBG("RA failed!");
                goto out;
        }

        if (ir3_shader_debug & IR3_DBG_OPTMSGS) {
                printf("AFTER RA:\n");
                ir3_print(ir);
        }

        /* fixup input/outputs: */
        for (i = 0; i < so->outputs_count; i++) {
                so->outputs[i].regid = ir->outputs[i*4]->regs[0]->num;
        }

        /* Note that some or all channels of an input may be unused: */
        actual_in = 0;
        inloc = 0;
        for (i = 0; i < so->inputs_count; i++) {
                unsigned j, reg = regid(63,0), compmask = 0, maxcomp = 0;
                so->inputs[i].ncomp = 0;
                so->inputs[i].inloc = inloc;
                for (j = 0; j < 4; j++) {
                        struct ir3_instruction *in = inputs[(i*4) + j];
                        if (in && !(in->flags & IR3_INSTR_UNUSED)) {
                                compmask |= (1 << j);
                                reg = in->regs[0]->num - j;
                                actual_in++;
                                so->inputs[i].ncomp++;
                                if ((so->type == MESA_SHADER_FRAGMENT) && so->inputs[i].bary) {
                                        /* assign inloc: */
                                        assert(in->regs[1]->flags & IR3_REG_IMMED);
                                        in->regs[1]->iim_val = inloc + j;
                                        maxcomp = j + 1;
                                }
                        }
                }
                if ((so->type == MESA_SHADER_FRAGMENT) && compmask && so->inputs[i].bary) {
                        so->varying_in++;
                        so->inputs[i].compmask = (1 << maxcomp) - 1;
                        inloc += maxcomp;
                } else if (!so->inputs[i].sysval) {
                        so->inputs[i].compmask = compmask;
                }
                so->inputs[i].regid = reg;
        }

        if (ctx->astc_srgb)
                fixup_astc_srgb(ctx);

        /* We need to do legalize after (for frag shader's) the "bary.f"
         * offsets (inloc) have been assigned.
         */
        ir3_legalize(ir, &so->num_samp, &so->has_ssbo, &max_bary);

        if (ir3_shader_debug & IR3_DBG_OPTMSGS) {
                printf("AFTER LEGALIZE:\n");
                ir3_print(ir);
        }

        /* Note that actual_in counts inputs that are not bary.f'd for FS: */
        if (so->type == MESA_SHADER_VERTEX)
                so->total_in = actual_in;
        else
                so->total_in = max_bary + 1;

out:
        if (ret) {
                if (so->ir)
                        ir3_destroy(so->ir);
                so->ir = NULL;
        }
        compile_free(ctx);

        return ret;
}