drivers/panfrost/pan_shader.c

/*
 * Copyright (c) 2022 Amazon.com, Inc. or its affiliates.
 * Copyright (C) 2019-2022 Collabora, Ltd.
 * Copyright (C) 2019 Red Hat Inc.
 * Copyright (C) 2018 Alyssa Rosenzweig
 *
 * 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 (Collabora):
 *   Alyssa Rosenzweig <[email protected]>
 *
 */

#include "pan_shader.h"
#include "nir/tgsi_to_nir.h"
#include "util/u_memory.h"
#include "util/u_prim.h"
#include "nir_builder.h"
#include "nir_serialize.h"
#include "pan_bo.h"
#include "pan_context.h"

static struct panfrost_uncompiled_shader *
panfrost_alloc_shader(const nir_shader *nir)
{
   struct panfrost_uncompiled_shader *so =
      rzalloc(NULL, struct panfrost_uncompiled_shader);

   simple_mtx_init(&so->lock, mtx_plain);
   util_dynarray_init(&so->variants, so);

   so->nir = nir;

   /* Serialize the NIR to a binary blob that we can hash for the disk
    * cache. Drop unnecessary information (like variable names) so the
    * serialized NIR is smaller, and also to let us detect more isomorphic
    * shaders when hashing, increasing cache hits.
    */
   struct blob blob;
   blob_init(&blob);
   nir_serialize(&blob, nir, true);
   _mesa_sha1_compute(blob.data, blob.size, so->nir_sha1);
   blob_finish(&blob);

   return so;
}

static struct panfrost_compiled_shader *
panfrost_alloc_variant(struct panfrost_uncompiled_shader *so)
{
   return util_dynarray_grow(&so->variants, struct panfrost_compiled_shader, 1);
}

static void
lower_load_poly_line_smooth_enabled(nir_shader *nir,
                                    const struct panfrost_shader_key *key)
{
   nir_function_impl *impl = nir_shader_get_entrypoint(nir);
   nir_builder b = nir_builder_create(impl);

   nir_foreach_block_safe(block, impl) {
      nir_foreach_instr_safe(instr, block) {
         if (instr->type != nir_instr_type_intrinsic)
            continue;

         nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
         if (intrin->intrinsic != nir_intrinsic_load_poly_line_smooth_enabled)
            continue;

         b.cursor = nir_before_instr(instr);
         nir_def_replace(&intrin->def, nir_imm_true(&b));
         nir_instr_free(instr);
      }
   }
}

static void
panfrost_shader_compile(struct panfrost_screen *screen, const nir_shader *ir,
                        struct util_debug_callback *dbg,
                        struct panfrost_shader_key *key, unsigned req_local_mem,
                        unsigned fixed_varying_mask,
                        struct panfrost_shader_binary *out)
{
   struct panfrost_device *dev = pan_device(&screen->base);

   nir_shader *s = nir_shader_clone(NULL, ir);

   /* While graphics shaders are preprocessed at CSO create time, compute
    * kernels are not preprocessed until they're cloned since the driver does
    * not get ownership of the NIR from compute CSOs. Do this preprocessing now.
    * Compute CSOs call this function during create time, so preprocessing
    * happens at CSO create time regardless.
    */
   if (gl_shader_stage_is_compute(s->info.stage))
      pan_shader_preprocess(s, panfrost_device_gpu_id(dev));

   struct panfrost_compile_inputs inputs = {
      .debug = dbg,
      .gpu_id = panfrost_device_gpu_id(dev),
   };

   /* Lower this early so the backends don't have to worry about it */
   if (s->info.stage == MESA_SHADER_FRAGMENT) {
      inputs.fixed_varying_mask = key->fs.fixed_varying_mask;
   } else if (s->info.stage == MESA_SHADER_VERTEX) {
      inputs.fixed_varying_mask = fixed_varying_mask;

      /* No IDVS for internal XFB shaders */
      inputs.no_idvs = s->info.has_transform_feedback_varyings;

      if (s->info.has_transform_feedback_varyings) {
         NIR_PASS_V(s, nir_io_add_const_offset_to_base,
                    nir_var_shader_in | nir_var_shader_out);
         NIR_PASS_V(s, nir_io_add_intrinsic_xfb_info);
         NIR_PASS_V(s, pan_lower_xfb);
      }
   }

   util_dynarray_init(&out->binary, NULL);

   if (s->info.stage == MESA_SHADER_FRAGMENT) {
      if (key->fs.nr_cbufs_for_fragcolor) {
         NIR_PASS_V(s, panfrost_nir_remove_fragcolor_stores,
                    key->fs.nr_cbufs_for_fragcolor);
      }

      if (key->fs.sprite_coord_enable) {
         NIR_PASS_V(s, nir_lower_texcoord_replace_late,
                    key->fs.sprite_coord_enable,
                    true /* point coord is sysval */);
      }

      if (key->fs.clip_plane_enable) {
         NIR_PASS_V(s, nir_lower_clip_fs, key->fs.clip_plane_enable, false);
      }

      if (key->fs.line_smooth) {
         NIR_PASS_V(s, nir_lower_poly_line_smooth, 16);
         NIR_PASS_V(s, lower_load_poly_line_smooth_enabled, key);
         NIR_PASS_V(s, nir_lower_alu);
      }
   }

   if (dev->arch <= 5 && s->info.stage == MESA_SHADER_FRAGMENT) {
      NIR_PASS_V(s, pan_lower_framebuffer, key->fs.rt_formats,
                 pan_raw_format_mask_midgard(key->fs.rt_formats), 0,
                 panfrost_device_gpu_id(dev) < 0x700);
   }

   NIR_PASS_V(s, panfrost_nir_lower_sysvals, dev->arch, &out->sysvals);

   /* Lower resource indices */
   NIR_PASS_V(s, panfrost_nir_lower_res_indices, &inputs);

   screen->vtbl.compile_shader(s, &inputs, &out->binary, &out->info);

   assert(req_local_mem >= out->info.wls_size);
   out->info.wls_size = req_local_mem;

   /* In both clone and tgsi_to_nir paths, the shader is ralloc'd against
    * a NULL context
    */
   ralloc_free(s);
}

static void
panfrost_shader_get(struct pipe_screen *pscreen,
                    struct panfrost_pool *shader_pool,
                    struct panfrost_pool *desc_pool,
                    struct panfrost_uncompiled_shader *uncompiled,
                    struct util_debug_callback *dbg,
                    struct panfrost_compiled_shader *state,
                    unsigned req_local_mem)
{
   struct panfrost_screen *screen = pan_screen(pscreen);
   struct panfrost_device *dev = pan_device(pscreen);

   struct panfrost_shader_binary res = {0};

   /* Try to retrieve the variant from the disk cache. If that fails,
    * compile a new variant and store in the disk cache for later reuse.
    */
   if (!panfrost_disk_cache_retrieve(screen->disk_cache, uncompiled,
                                     &state->key, &res)) {
      panfrost_shader_compile(screen, uncompiled->nir, dbg, &state->key,
                              req_local_mem, uncompiled->fixed_varying_mask,
                              &res);

      panfrost_disk_cache_store(screen->disk_cache, uncompiled, &state->key,
                                &res);
   }

   state->info = res.info;
   state->sysvals = res.sysvals;

   if (res.binary.size) {
      state->bin = panfrost_pool_take_ref(
         shader_pool,
         pan_pool_upload_aligned(&shader_pool->base, res.binary.data,
                                 res.binary.size, 128));
   }

   util_dynarray_fini(&res.binary);

   /* Don't upload RSD for fragment shaders since they need draw-time
    * merging for e.g. depth/stencil/alpha. RSDs are replaced by simpler
    * shader program descriptors on Valhall, which can be preuploaded even
    * for fragment shaders. */
   bool upload =
      !(uncompiled->nir->info.stage == MESA_SHADER_FRAGMENT && dev->arch <= 7);
   screen->vtbl.prepare_shader(state, desc_pool, upload);

   panfrost_analyze_sysvals(state);
}

static void
panfrost_build_key(struct panfrost_context *ctx,
                   struct panfrost_shader_key *key,
                   struct panfrost_uncompiled_shader *uncompiled)
{
   const nir_shader *nir = uncompiled->nir;

   /* We don't currently have vertex shader variants */
   if (nir->info.stage != MESA_SHADER_FRAGMENT)
      return;

   struct panfrost_device *dev = pan_device(ctx->base.screen);
   struct pipe_framebuffer_state *fb = &ctx->pipe_framebuffer;
   struct pipe_rasterizer_state *rast = (void *)ctx->rasterizer;
   struct panfrost_uncompiled_shader *vs = ctx->uncompiled[MESA_SHADER_VERTEX];

   /* gl_FragColor lowering needs the number of colour buffers */
   if (uncompiled->fragcolor_lowered) {
      key->fs.nr_cbufs_for_fragcolor = fb->nr_cbufs;
   }

   /* Point sprite lowering needed on Bifrost and newer */
   if (dev->arch >= 6 && rast && ctx->active_prim == MESA_PRIM_POINTS) {
      key->fs.sprite_coord_enable = rast->sprite_coord_enable;
   }

   /* User clip plane lowering needed everywhere */
   if (rast) {
      key->fs.clip_plane_enable = rast->clip_plane_enable;

      if (u_reduced_prim(ctx->active_prim) == MESA_PRIM_LINES)
         key->fs.line_smooth = rast->line_smooth;
   }

   if (dev->arch <= 5) {
      u_foreach_bit(i, (nir->info.outputs_read >> FRAG_RESULT_DATA0)) {
         enum pipe_format fmt = PIPE_FORMAT_R8G8B8A8_UNORM;

         if ((fb->nr_cbufs > i) && fb->cbufs[i])
            fmt = fb->cbufs[i]->format;

         if (panfrost_blendable_formats_v6[fmt].internal)
            fmt = PIPE_FORMAT_NONE;

         key->fs.rt_formats[i] = fmt;
      }
   }

   /* Funny desktop GL varying lowering on Valhall */
   if (dev->arch >= 9) {
      assert(vs != NULL && "too early");
      key->fs.fixed_varying_mask = vs->fixed_varying_mask;
   }
}

static struct panfrost_compiled_shader *
panfrost_new_variant_locked(struct panfrost_context *ctx,
                            struct panfrost_uncompiled_shader *uncompiled,
                            struct panfrost_shader_key *key)
{
   struct panfrost_compiled_shader *prog = panfrost_alloc_variant(uncompiled);

   *prog = (struct panfrost_compiled_shader){
      .key = *key,
      .stream_output = uncompiled->stream_output,
   };

   panfrost_shader_get(ctx->base.screen, &ctx->shaders, &ctx->descs, uncompiled,
                       &ctx->base.debug, prog, 0);

   prog->earlyzs = pan_earlyzs_analyze(&prog->info);

   return prog;
}

static void
panfrost_bind_shader_state(struct pipe_context *pctx, void *hwcso,
                           enum pipe_shader_type type)
{
   struct panfrost_context *ctx = pan_context(pctx);
   ctx->uncompiled[type] = hwcso;
   ctx->prog[type] = NULL;

   ctx->dirty |= PAN_DIRTY_TLS_SIZE;
   ctx->dirty_shader[type] |= PAN_DIRTY_STAGE_SHADER;

   if (hwcso)
      panfrost_update_shader_variant(ctx, type);
}

void
panfrost_update_shader_variant(struct panfrost_context *ctx,
                               enum pipe_shader_type type)
{
   /* No shader variants for compute */
   if (type == PIPE_SHADER_COMPUTE)
      return;

   /* We need linking information, defer this */
   if (type == PIPE_SHADER_FRAGMENT && !ctx->uncompiled[PIPE_SHADER_VERTEX])
      return;

   /* Also defer, happens with GALLIUM_HUD */
   if (!ctx->uncompiled[type])
      return;

   /* Match the appropriate variant */
   struct panfrost_uncompiled_shader *uncompiled = ctx->uncompiled[type];
   struct panfrost_compiled_shader *compiled = NULL;

   simple_mtx_lock(&uncompiled->lock);

   struct panfrost_shader_key key = {0};
   panfrost_build_key(ctx, &key, uncompiled);

   util_dynarray_foreach(&uncompiled->variants, struct panfrost_compiled_shader,
                         so) {
      if (memcmp(&key, &so->key, sizeof(key)) == 0) {
         compiled = so;
         break;
      }
   }

   if (compiled == NULL)
      compiled = panfrost_new_variant_locked(ctx, uncompiled, &key);

   ctx->prog[type] = compiled;

   simple_mtx_unlock(&uncompiled->lock);
}

static void
panfrost_bind_vs_state(struct pipe_context *pctx, void *hwcso)
{
   panfrost_bind_shader_state(pctx, hwcso, PIPE_SHADER_VERTEX);

   /* Fragment shaders are linked with vertex shaders */
   struct panfrost_context *ctx = pan_context(pctx);
   panfrost_update_shader_variant(ctx, PIPE_SHADER_FRAGMENT);
}

static void
panfrost_bind_fs_state(struct pipe_context *pctx, void *hwcso)
{
   panfrost_bind_shader_state(pctx, hwcso, PIPE_SHADER_FRAGMENT);
}

static void *
panfrost_create_shader_state(struct pipe_context *pctx,
                             const struct pipe_shader_state *cso)
{
   nir_shader *nir = (cso->type == PIPE_SHADER_IR_TGSI)
                        ? tgsi_to_nir(cso->tokens, pctx->screen, false)
                        : cso->ir.nir;

   struct panfrost_uncompiled_shader *so = panfrost_alloc_shader(nir);

   /* The driver gets ownership of the nir_shader for graphics. The NIR is
    * ralloc'd. Free the NIR when we free the uncompiled shader.
    */
   ralloc_steal(so, nir);

   so->stream_output = cso->stream_output;
   so->nir = nir;

   /* Fix linkage early */
   if (so->nir->info.stage == MESA_SHADER_VERTEX) {
      so->fixed_varying_mask =
         (so->nir->info.outputs_written & BITFIELD_MASK(VARYING_SLOT_VAR0)) &
         ~VARYING_BIT_POS & ~VARYING_BIT_PSIZ;
   }

   /* gl_FragColor needs to be lowered before lowering I/O, do that now */
   if (nir->info.stage == MESA_SHADER_FRAGMENT &&
       nir->info.outputs_written & BITFIELD_BIT(FRAG_RESULT_COLOR)) {

      NIR_PASS_V(nir, nir_lower_fragcolor, nir->info.fs.color_is_dual_source ? 1 : 8);
      so->fragcolor_lowered = true;
   }

   /* Then run the suite of lowering and optimization, including I/O lowering */
   struct panfrost_device *dev = pan_device(pctx->screen);
   pan_shader_preprocess(nir, panfrost_device_gpu_id(dev));

   /* Vertex shaders get passed images through the vertex attribute descriptor
    * array. We need to add an offset to all image intrinsics so they point
    * to the right attribute.
    */
   if (nir->info.stage == MESA_SHADER_VERTEX && dev->arch <= 7) {
      NIR_PASS_V(nir, pan_lower_image_index,
                 util_bitcount64(nir->info.inputs_read));
   }

   /* If this shader uses transform feedback, compile the transform
    * feedback program. This is a special shader variant.
    */
   struct panfrost_context *ctx = pan_context(pctx);

   if (so->nir->xfb_info) {
      so->xfb = calloc(1, sizeof(struct panfrost_compiled_shader));
      so->xfb->key.vs_is_xfb = true;

      panfrost_shader_get(ctx->base.screen, &ctx->shaders, &ctx->descs, so,
                          &ctx->base.debug, so->xfb, 0);

      /* Since transform feedback is handled via the transform
       * feedback program, the original program no longer uses XFB
       */
      nir->info.has_transform_feedback_varyings = false;
   }

   /* Compile the program. We don't use vertex shader keys, so there will
    * be no further vertex shader variants. We do have fragment shader
    * keys, but we can still compile with a default key that will work most
    * of the time.
    */
   struct panfrost_shader_key key = {0};

   /* gl_FragColor lowering needs the number of colour buffers on desktop
    * GL, where it acts as an implicit broadcast to all colour buffers.
    *
    * However, gl_FragColor is a legacy feature, so assume that if
    * gl_FragColor is used, there is only a single render target. The
    * implicit broadcast is neither especially useful nor required by GLES.
    */
   if (so->fragcolor_lowered)
      key.fs.nr_cbufs_for_fragcolor = 1;

   /* Creating a CSO is single-threaded, so it's ok to use the
    * locked function without explicitly taking the lock. Creating a
    * default variant acts as a precompile.
    */
   panfrost_new_variant_locked(ctx, so, &key);

   return so;
}

static void
panfrost_delete_shader_state(struct pipe_context *pctx, void *so)
{
   struct panfrost_uncompiled_shader *cso =
      (struct panfrost_uncompiled_shader *)so;

   util_dynarray_foreach(&cso->variants, struct panfrost_compiled_shader, so) {
      panfrost_bo_unreference(so->bin.bo);
      panfrost_bo_unreference(so->state.bo);
      panfrost_bo_unreference(so->linkage.bo);
   }

   if (cso->xfb) {
      panfrost_bo_unreference(cso->xfb->bin.bo);
      panfrost_bo_unreference(cso->xfb->state.bo);
      panfrost_bo_unreference(cso->xfb->linkage.bo);
      free(cso->xfb);
   }

   simple_mtx_destroy(&cso->lock);

   ralloc_free(so);
}

/*
 * Create a compute CSO. As compute kernels do not require variants, they are
 * precompiled, creating both the uncompiled and compiled shaders now.
 */
static void *
panfrost_create_compute_state(struct pipe_context *pctx,
                              const struct pipe_compute_state *cso)
{
   struct panfrost_context *ctx = pan_context(pctx);
   struct panfrost_uncompiled_shader *so = panfrost_alloc_shader(cso->prog);
   struct panfrost_compiled_shader *v = panfrost_alloc_variant(so);
   memset(v, 0, sizeof *v);

   assert(cso->ir_type == PIPE_SHADER_IR_NIR && "TGSI kernels unsupported");

   panfrost_shader_get(pctx->screen, &ctx->shaders, &ctx->descs, so,
                       &ctx->base.debug, v, cso->static_shared_mem);

   /* The NIR becomes invalid after this. For compute kernels, we never
    * need to access it again. Don't keep a dangling pointer around.
    */
   ralloc_free((void *)so->nir);
   so->nir = NULL;

   return so;
}

static void
panfrost_bind_compute_state(struct pipe_context *pipe, void *cso)
{
   struct panfrost_context *ctx = pan_context(pipe);
   struct panfrost_uncompiled_shader *uncompiled = cso;

   ctx->uncompiled[PIPE_SHADER_COMPUTE] = uncompiled;

   ctx->prog[PIPE_SHADER_COMPUTE] =
      uncompiled ? util_dynarray_begin(&uncompiled->variants) : NULL;
}

static void
panfrost_get_compute_state_info(struct pipe_context *pipe, void *cso,
                                struct pipe_compute_state_object_info *info)
{
   struct panfrost_device *dev = pan_device(pipe->screen);
   struct panfrost_uncompiled_shader *uncompiled = cso;
   struct panfrost_compiled_shader *cs =
      util_dynarray_begin(&uncompiled->variants);

   info->max_threads = panfrost_compute_max_thread_count(
      &dev->kmod.props, cs->info.work_reg_count);
   info->private_memory = cs->info.tls_size;
   info->simd_sizes = pan_subgroup_size(dev->arch);
   info->preferred_simd_size = info->simd_sizes;
}

void
panfrost_shader_context_init(struct pipe_context *pctx)
{
   pctx->create_vs_state = panfrost_create_shader_state;
   pctx->delete_vs_state = panfrost_delete_shader_state;
   pctx->bind_vs_state = panfrost_bind_vs_state;

   pctx->create_fs_state = panfrost_create_shader_state;
   pctx->delete_fs_state = panfrost_delete_shader_state;
   pctx->bind_fs_state = panfrost_bind_fs_state;

   pctx->create_compute_state = panfrost_create_compute_state;
   pctx->bind_compute_state = panfrost_bind_compute_state;
   pctx->get_compute_state_info = panfrost_get_compute_state_info;
   pctx->delete_compute_state = panfrost_delete_shader_state;
}