nir: Use the flrp lowering pass instead of nir_opt_algebraic
[mesa.git] / src / gallium / drivers / vc4 / vc4_context.c
index c75cea2fcb48ba4ed0a63336b4e2ef7f455ec670..94969dcb133b447e965e5bfa303918115ad1be38 100644 (file)
 
 #include <xf86drm.h>
 #include <err.h>
-#include <stdio.h>
 
 #include "pipe/p_defines.h"
+#include "util/ralloc.h"
 #include "util/u_inlines.h"
 #include "util/u_memory.h"
 #include "util/u_blitter.h"
+#include "util/u_upload_mgr.h"
 #include "indices/u_primconvert.h"
 #include "pipe/p_screen.h"
 
 #include "vc4_context.h"
 #include "vc4_resource.h"
 
-static void
-dump_fbo(struct vc4_context *vc4, struct vc4_bo *fbo)
+void
+vc4_flush(struct pipe_context *pctx)
 {
-#ifndef USE_VC4_SIMULATOR
-        uint32_t *map = vc4_bo_map(fbo);
-        uint32_t width = vc4->framebuffer.width;
-        uint32_t height = vc4->framebuffer.height;
-        uint32_t chunk_w = width / 79;
-        uint32_t chunk_h = height / 40;
-        uint32_t found_colors[10];
-        uint32_t num_found_colors = 0;
-
-        for (int by = 0; by < height; by += chunk_h) {
-                for (int bx = 0; bx < width; bx += chunk_w) {
-                        bool on = false, black = false;
-
-                        for (int y = by; y < MIN2(height, by + chunk_h); y++) {
-                                for (int x = bx; x < MIN2(width, bx + chunk_w); x++) {
-                                        uint32_t pix = map[y * width + x];
-                                        on |= pix != 0;
-                                        black |= pix == 0xff000000;
-
-                                        int i;
-                                        for (i = 0; i < num_found_colors; i++) {
-                                                if (pix == found_colors[i])
-                                                        break;
-                                        }
-                                        if (i == num_found_colors &&
-                                            num_found_colors < Elements(found_colors))
-                                                found_colors[num_found_colors++] = pix;
-                                }
-                        }
-                        if (black)
-                                fprintf(stderr, "O");
-                        else if (on)
-                                fprintf(stderr, "X");
-                        else
-                                fprintf(stderr, ".");
-                }
-                fprintf(stderr, "\n");
-        }
+        struct vc4_context *vc4 = vc4_context(pctx);
 
-        for (int i = 0; i < num_found_colors; i++) {
-                fprintf(stderr, "color %d: 0x%08x\n", i, found_colors[i]);
+        hash_table_foreach(vc4->jobs, entry) {
+                struct vc4_job *job = entry->data;
+                vc4_job_submit(vc4, job);
         }
-#endif
 }
 
-void
-vc4_flush(struct pipe_context *pctx)
+static void
+vc4_pipe_flush(struct pipe_context *pctx, struct pipe_fence_handle **fence,
+               unsigned flags)
 {
         struct vc4_context *vc4 = vc4_context(pctx);
 
-        if (!vc4->needs_flush)
-                return;
+        vc4_flush(pctx);
 
-        cl_u8(&vc4->bcl, VC4_PACKET_FLUSH_ALL);
-        cl_u8(&vc4->bcl, VC4_PACKET_NOP);
-        cl_u8(&vc4->bcl, VC4_PACKET_HALT);
-
-        struct vc4_surface *csurf = vc4_surface(vc4->framebuffer.cbufs[0]);
-        struct vc4_resource *ctex = vc4_resource(csurf->base.texture);
-        struct drm_vc4_submit_cl submit;
-        memset(&submit, 0, sizeof(submit));
-
-        submit.bo_handles = vc4->bo_handles.base;
-        submit.bo_handle_count = (vc4->bo_handles.next -
-                                  vc4->bo_handles.base) / 4;
-        submit.bin_cl = vc4->bcl.base;
-        submit.bin_cl_size = vc4->bcl.next - vc4->bcl.base;
-        submit.render_cl = vc4->rcl.base;
-        submit.render_cl_size = vc4->rcl.next - vc4->rcl.base;
-        submit.shader_rec = vc4->shader_rec.base;
-        submit.shader_rec_size = vc4->shader_rec.next - vc4->shader_rec.base;
-        submit.shader_rec_count = vc4->shader_rec_count;
-        submit.uniforms = vc4->uniforms.base;
-        submit.uniforms_size = vc4->uniforms.next - vc4->uniforms.base;
-
-        if (!(vc4_debug & VC4_DEBUG_NORAST)) {
-                int ret;
-
-#ifndef USE_VC4_SIMULATOR
-                ret = drmIoctl(vc4->fd, DRM_IOCTL_VC4_SUBMIT_CL, &submit);
-#else
-                ret = vc4_simulator_flush(vc4, &submit, csurf);
-#endif
-                if (ret)
-                        errx(1, "VC4 submit failed\n");
+        if (fence) {
+                struct pipe_screen *screen = pctx->screen;
+                int fd = -1;
+
+                if (flags & PIPE_FLUSH_FENCE_FD) {
+                        /* The vc4_fence takes ownership of the returned fd. */
+                        drmSyncobjExportSyncFile(vc4->fd, vc4->job_syncobj,
+                                                 &fd);
+                }
+
+                struct vc4_fence *f = vc4_fence_create(vc4->screen,
+                                                       vc4->last_emit_seqno,
+                                                       fd);
+                screen->fence_reference(screen, fence, NULL);
+                *fence = (struct pipe_fence_handle *)f;
         }
+}
 
-        vc4_reset_cl(&vc4->bcl);
-        vc4_reset_cl(&vc4->rcl);
-        vc4_reset_cl(&vc4->shader_rec);
-        vc4_reset_cl(&vc4->uniforms);
-        vc4_reset_cl(&vc4->bo_handles);
-#ifdef USE_VC4_SIMULATOR
-        vc4_reset_cl(&vc4->bo_pointers);
-#endif
-        vc4->shader_rec_count = 0;
+/* We can't flush the texture cache within rendering a tile, so we have to
+ * flush all rendering to the kernel so that the next job reading from the
+ * tile gets a flushed cache.
+ */
+static void
+vc4_texture_barrier(struct pipe_context *pctx, unsigned flags)
+{
+        vc4_flush(pctx);
+}
 
-        vc4->needs_flush = false;
-        vc4->dirty = ~0;
+static void
+vc4_set_debug_callback(struct pipe_context *pctx,
+                       const struct pipe_debug_callback *cb)
+{
+        struct vc4_context *vc4 = vc4_context(pctx);
 
-        dump_fbo(vc4, ctex->bo);
+        if (cb)
+                vc4->debug = *cb;
+        else
+                memset(&vc4->debug, 0, sizeof(vc4->debug));
 }
 
 static void
-vc4_pipe_flush(struct pipe_context *pctx, struct pipe_fence_handle **fence,
-               unsigned flags)
+vc4_invalidate_resource(struct pipe_context *pctx, struct pipe_resource *prsc)
 {
-        vc4_flush(pctx);
+        struct vc4_context *vc4 = vc4_context(pctx);
+        struct vc4_resource *rsc = vc4_resource(prsc);
+
+        rsc->initialized_buffers = 0;
+
+        struct hash_entry *entry = _mesa_hash_table_search(vc4->write_jobs,
+                                                           prsc);
+        if (!entry)
+                return;
+
+        struct vc4_job *job = entry->data;
+        if (job->key.zsbuf && job->key.zsbuf->texture == prsc)
+                job->resolve &= ~(PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL);
 }
 
 static void
@@ -154,29 +119,54 @@ vc4_context_destroy(struct pipe_context *pctx)
 {
         struct vc4_context *vc4 = vc4_context(pctx);
 
+        vc4_flush(pctx);
+
         if (vc4->blitter)
                 util_blitter_destroy(vc4->blitter);
 
         if (vc4->primconvert)
                 util_primconvert_destroy(vc4->primconvert);
 
-        util_slab_destroy(&vc4->transfer_pool);
+        if (vc4->uploader)
+                u_upload_destroy(vc4->uploader);
+
+        slab_destroy_child(&vc4->transfer_pool);
+
+        pipe_surface_reference(&vc4->framebuffer.cbufs[0], NULL);
+        pipe_surface_reference(&vc4->framebuffer.zsbuf, NULL);
 
-        free(vc4);
+        if (vc4->yuv_linear_blit_vs)
+                pctx->delete_vs_state(pctx, vc4->yuv_linear_blit_vs);
+        if (vc4->yuv_linear_blit_fs_8bit)
+                pctx->delete_fs_state(pctx, vc4->yuv_linear_blit_fs_8bit);
+        if (vc4->yuv_linear_blit_fs_16bit)
+                pctx->delete_fs_state(pctx, vc4->yuv_linear_blit_fs_16bit);
+
+        vc4_program_fini(pctx);
+
+        if (vc4->screen->has_syncobj) {
+                drmSyncobjDestroy(vc4->fd, vc4->job_syncobj);
+                drmSyncobjDestroy(vc4->fd, vc4->in_syncobj);
+        }
+        if (vc4->in_fence_fd >= 0)
+                close(vc4->in_fence_fd);
+
+        ralloc_free(vc4);
 }
 
 struct pipe_context *
-vc4_context_create(struct pipe_screen *pscreen, void *priv)
+vc4_context_create(struct pipe_screen *pscreen, void *priv, unsigned flags)
 {
         struct vc4_screen *screen = vc4_screen(pscreen);
         struct vc4_context *vc4;
+        int err;
 
         /* Prevent dumping of the shaders built during context setup. */
         uint32_t saved_shaderdb_flag = vc4_debug & VC4_DEBUG_SHADERDB;
         vc4_debug &= ~VC4_DEBUG_SHADERDB;
 
-        vc4 = CALLOC_STRUCT(vc4_context);
-        if (vc4 == NULL)
+        vc4 = rzalloc(NULL, struct vc4_context);
+        if (!vc4)
                 return NULL;
         struct pipe_context *pctx = &vc4->base;
 
@@ -186,33 +176,45 @@ vc4_context_create(struct pipe_screen *pscreen, void *priv)
         pctx->priv = priv;
         pctx->destroy = vc4_context_destroy;
         pctx->flush = vc4_pipe_flush;
+        pctx->set_debug_callback = vc4_set_debug_callback;
+        pctx->invalidate_resource = vc4_invalidate_resource;
+        pctx->texture_barrier = vc4_texture_barrier;
 
         vc4_draw_init(pctx);
         vc4_state_init(pctx);
         vc4_program_init(pctx);
+        vc4_query_init(pctx);
         vc4_resource_context_init(pctx);
 
-        vc4_init_cl(vc4, &vc4->bcl);
-        vc4_init_cl(vc4, &vc4->rcl);
-        vc4_init_cl(vc4, &vc4->shader_rec);
-        vc4_init_cl(vc4, &vc4->bo_handles);
-
-        vc4->dirty = ~0;
         vc4->fd = screen->fd;
 
-        util_slab_create(&vc4->transfer_pool, sizeof(struct pipe_transfer),
-                         16, UTIL_SLAB_SINGLETHREADED);
-        vc4->blitter = util_blitter_create(pctx);
+        err = vc4_job_init(vc4);
+        if (err)
+                goto fail;
+
+        err = vc4_fence_context_init(vc4);
+        if (err)
+                goto fail;
+
+        slab_create_child(&vc4->transfer_pool, &screen->transfer_pool);
+
+       vc4->uploader = u_upload_create_default(&vc4->base);
+       vc4->base.stream_uploader = vc4->uploader;
+       vc4->base.const_uploader = vc4->uploader;
+
+       vc4->blitter = util_blitter_create(pctx);
         if (!vc4->blitter)
                 goto fail;
 
         vc4->primconvert = util_primconvert_create(pctx,
-                                                   !((1 << PIPE_PRIM_QUADS) - 1));
+                                                   (1 << PIPE_PRIM_QUADS) - 1);
         if (!vc4->primconvert)
                 goto fail;
 
         vc4_debug |= saved_shaderdb_flag;
 
+        vc4->sample_mask = (1 << VC4_MAX_SAMPLES) - 1;
+
         return &vc4->base;
 
 fail: