Merge Mesa 23.1.9

6 files changed, 76 insertions, 52 deletions

diff --git a/lib/mesa/src/gallium/drivers/vc4/vc4_bufmgr.c b/lib/mesa/src/gallium/drivers/vc4/vc4_bufmgr.c
index 7a366caf6..7c1195e3f 100644
--- a/lib/mesa/src/gallium/drivers/vc4/vc4_bufmgr.c
+++ b/lib/mesa/src/gallium/drivers/vc4/vc4_bufmgr.c
@@ -382,9 +382,11 @@ vc4_bo_open_handle(struct vc4_screen *screen,
 {
         struct vc4_bo *bo;
 
-        assert(size);
+        /* Note: the caller is responsible for locking screen->bo_handles_mutex.
+         * This allows the lock to cover the actual BO import, avoiding a race.
+         */
 
-        mtx_lock(&screen->bo_handles_mutex);
+        assert(size);
 
         bo = util_hash_table_get(screen->bo_handles, (void*)(uintptr_t)handle);
         if (bo) {
@@ -418,10 +420,14 @@ vc4_bo_open_name(struct vc4_screen *screen, uint32_t name)
         struct drm_gem_open o = {
                 .name = name
         };
+
+        mtx_lock(&screen->bo_handles_mutex);
+
         int ret = vc4_ioctl(screen->fd, DRM_IOCTL_GEM_OPEN, &o);
         if (ret) {
                 fprintf(stderr, "Failed to open bo %d: %s\n",
                         name, strerror(errno));
+                mtx_unlock(&screen->bo_handles_mutex);
                 return NULL;
         }
 
@@ -432,10 +438,14 @@ struct vc4_bo *
 vc4_bo_open_dmabuf(struct vc4_screen *screen, int fd)
 {
         uint32_t handle;
+
+        mtx_lock(&screen->bo_handles_mutex);
+
         int ret = drmPrimeFDToHandle(screen->fd, fd, &handle);
         int size;
         if (ret) {
                 fprintf(stderr, "Failed to get vc4 handle for dmabuf %d\n", fd);
+                mtx_unlock(&screen->bo_handles_mutex);
                 return NULL;
         }
 
@@ -443,6 +453,7 @@ vc4_bo_open_dmabuf(struct vc4_screen *screen, int fd)
         size = lseek(fd, 0, SEEK_END);
         if (size == -1) {
                 fprintf(stderr, "Couldn't get size of dmabuf fd %d.\n", fd);
+                mtx_unlock(&screen->bo_handles_mutex);
                 return NULL;
         }
 
diff --git a/lib/mesa/src/gallium/drivers/vc4/vc4_context.c b/lib/mesa/src/gallium/drivers/vc4/vc4_context.c
index f03cf57cc..472a3e1b0 100644
--- a/lib/mesa/src/gallium/drivers/vc4/vc4_context.c
+++ b/lib/mesa/src/gallium/drivers/vc4/vc4_context.c
@@ -116,8 +116,7 @@ vc4_context_destroy(struct pipe_context *pctx)
 
         slab_destroy_child(&vc4->transfer_pool);
 
-        pipe_surface_reference(&vc4->framebuffer.cbufs[0], NULL);
-        pipe_surface_reference(&vc4->framebuffer.zsbuf, NULL);
+        util_unreference_framebuffer_state(&vc4->framebuffer);
 
         if (vc4->yuv_linear_blit_vs)
                 pctx->delete_vs_state(pctx, vc4->yuv_linear_blit_vs);
diff --git a/lib/mesa/src/gallium/drivers/vc4/vc4_program.c b/lib/mesa/src/gallium/drivers/vc4/vc4_program.c
index 426512c31..67a05b994 100644
--- a/lib/mesa/src/gallium/drivers/vc4/vc4_program.c
+++ b/lib/mesa/src/gallium/drivers/vc4/vc4_program.c
@@ -806,14 +806,7 @@ add_output(struct vc4_compile *c,
 static bool
 ntq_src_is_only_ssa_def_user(nir_src *src)
 {
-        if (!src->is_ssa)
-                return false;
-
-        if (!list_is_empty(&src->ssa->if_uses))
-                return false;
-
-        return (src->ssa->uses.next == &src->use_link &&
-                src->ssa->uses.next->next == &src->ssa->uses);
+        return src->is_ssa && list_is_singular(&src->ssa->uses);
 }
 
 /**
@@ -1146,13 +1139,6 @@ ntq_emit_alu(struct vc4_compile *c, nir_alu_instr *instr)
         case nir_op_b2i32:
                 result = qir_AND(c, src[0], qir_uniform_ui(c, 1));
                 break;
-        case nir_op_i2b32:
-        case nir_op_f2b32:
-                qir_SF(c, src[0]);
-                result = qir_MOV(c, qir_SEL(c, QPU_COND_ZC,
-                                            qir_uniform_ui(c, ~0),
-                                            qir_uniform_ui(c, 0)));
-                break;
 
         case nir_op_iadd:
                 result = qir_ADD(c, src[0], src[1]);
@@ -2057,6 +2043,7 @@ static void ntq_emit_cf_list(struct vc4_compile *c, struct exec_list *list);
 static void
 ntq_emit_loop(struct vc4_compile *c, nir_loop *loop)
 {
+        assert(!nir_loop_has_continue_construct(loop));
         if (!c->vc4->screen->has_control_flow) {
                 fprintf(stderr,
                         "loop support requires updated kernel.\n");
diff --git a/lib/mesa/src/gallium/drivers/vc4/vc4_resource.c b/lib/mesa/src/gallium/drivers/vc4/vc4_resource.c
index ab918a05c..65bf7ca53 100644
--- a/lib/mesa/src/gallium/drivers/vc4/vc4_resource.c
+++ b/lib/mesa/src/gallium/drivers/vc4/vc4_resource.c
@@ -95,34 +95,13 @@ vc4_resource_transfer_unmap(struct pipe_context *pctx,
         slab_free(&vc4->transfer_pool, ptrans);
 }
 
-static void *
-vc4_resource_transfer_map(struct pipe_context *pctx,
-                          struct pipe_resource *prsc,
-                          unsigned level, unsigned usage,
-                          const struct pipe_box *box,
-                          struct pipe_transfer **pptrans)
+static void
+vc4_map_usage_prep(struct pipe_context *pctx,
+                   struct pipe_resource *prsc,
+                   unsigned usage)
 {
         struct vc4_context *vc4 = vc4_context(pctx);
         struct vc4_resource *rsc = vc4_resource(prsc);
-        struct vc4_transfer *trans;
-        struct pipe_transfer *ptrans;
-        enum pipe_format format = prsc->format;
-        char *buf;
-
-        /* Upgrade DISCARD_RANGE to WHOLE_RESOURCE if the whole resource is
-         * being mapped.
-         */
-        if ((usage & PIPE_MAP_DISCARD_RANGE) &&
-            !(usage & PIPE_MAP_UNSYNCHRONIZED) &&
-            !(prsc->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT) &&
-            prsc->last_level == 0 &&
-            prsc->width0 == box->width &&
-            prsc->height0 == box->height &&
-            prsc->depth0 == box->depth &&
-            prsc->array_size == 1 &&
-            rsc->bo->private) {
-                usage |= PIPE_MAP_DISCARD_WHOLE_RESOURCE;
-        }
 
         if (usage & PIPE_MAP_DISCARD_WHOLE_RESOURCE) {
                 if (vc4_resource_bo_alloc(rsc)) {
@@ -131,6 +110,8 @@ vc4_resource_transfer_map(struct pipe_context *pctx,
                          */
                         if (prsc->bind & PIPE_BIND_VERTEX_BUFFER)
                                 vc4->dirty |= VC4_DIRTY_VTXBUF;
+                        if (prsc->bind & PIPE_BIND_CONSTANT_BUFFER)
+                                vc4->dirty |= VC4_DIRTY_CONSTBUF;
                 } else {
                         /* If we failed to reallocate, flush users so that we
                          * don't violate any syncing requirements.
@@ -139,7 +120,7 @@ vc4_resource_transfer_map(struct pipe_context *pctx,
                 }
         } else if (!(usage & PIPE_MAP_UNSYNCHRONIZED)) {
                 /* If we're writing and the buffer is being used by the CL, we
-                 * have to flush the CL first.  If we're only reading, we need
+                 * have to flush the CL first. If we're only reading, we need
                  * to flush if the CL has written our buffer.
                  */
                 if (usage & PIPE_MAP_WRITE)
@@ -152,6 +133,38 @@ vc4_resource_transfer_map(struct pipe_context *pctx,
                 rsc->writes++;
                 rsc->initialized_buffers = ~0;
         }
+}
+
+static void *
+vc4_resource_transfer_map(struct pipe_context *pctx,
+                          struct pipe_resource *prsc,
+                          unsigned level, unsigned usage,
+                          const struct pipe_box *box,
+                          struct pipe_transfer **pptrans)
+{
+        struct vc4_context *vc4 = vc4_context(pctx);
+        struct vc4_resource *rsc = vc4_resource(prsc);
+        struct vc4_transfer *trans;
+        struct pipe_transfer *ptrans;
+        enum pipe_format format = prsc->format;
+        char *buf;
+
+        /* Upgrade DISCARD_RANGE to WHOLE_RESOURCE if the whole resource is
+         * being mapped.
+         */
+        if ((usage & PIPE_MAP_DISCARD_RANGE) &&
+            !(usage & PIPE_MAP_UNSYNCHRONIZED) &&
+            !(prsc->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT) &&
+            prsc->last_level == 0 &&
+            prsc->width0 == box->width &&
+            prsc->height0 == box->height &&
+            prsc->depth0 == box->depth &&
+            prsc->array_size == 1 &&
+            rsc->bo->private) {
+                usage |= PIPE_MAP_DISCARD_WHOLE_RESOURCE;
+        }
+
+        vc4_map_usage_prep(pctx, prsc, usage);
 
         trans = slab_zalloc(&vc4->transfer_pool);
         if (!trans)
@@ -240,8 +253,12 @@ vc4_texture_subdata(struct pipe_context *pctx,
         }
 
         /* Otherwise, map and store the texture data directly into the tiled
-         * texture.
+         * texture.  Note that gallium's texture_subdata may be called with
+         * obvious usage flags missing!
          */
+        vc4_map_usage_prep(pctx, prsc, usage | (PIPE_MAP_WRITE |
+                                                PIPE_MAP_DISCARD_RANGE));
+
         void *buf;
         if (usage & PIPE_MAP_UNSYNCHRONIZED)
                 buf = vc4_bo_map_unsynchronized(rsc->bo);
diff --git a/lib/mesa/src/gallium/drivers/vc4/vc4_screen.c b/lib/mesa/src/gallium/drivers/vc4/vc4_screen.c
index aec151168..4a0e34956 100644
--- a/lib/mesa/src/gallium/drivers/vc4/vc4_screen.c
+++ b/lib/mesa/src/gallium/drivers/vc4/vc4_screen.c
@@ -69,7 +69,7 @@ static const struct debug_named_value vc4_debug_options[] = {
         { "dump", VC4_DEBUG_DUMP,
           "Write a GPU command stream trace file" },
 #endif
-        { NULL }
+        DEBUG_NAMED_VALUE_END
 };
 
 DEBUG_GET_ONCE_FLAGS_OPTION(vc4_debug, "VC4_DEBUG", vc4_debug_options, 0)
@@ -294,8 +294,6 @@ vc4_screen_get_shader_param(struct pipe_screen *pscreen,
         case PIPE_SHADER_CAP_INT16:
         case PIPE_SHADER_CAP_GLSL_16BIT_CONSTS:
         case PIPE_SHADER_CAP_DROUND_SUPPORTED:
-        case PIPE_SHADER_CAP_DFRACEXP_DLDEXP_SUPPORTED:
-        case PIPE_SHADER_CAP_LDEXP_SUPPORTED:
         case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
                 return 0;
         case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
@@ -537,8 +535,17 @@ vc4_get_chip_info(struct vc4_screen *screen)
         return true;
 }
 
+static int
+vc4_screen_get_fd(struct pipe_screen *pscreen)
+{
+        struct vc4_screen *screen = vc4_screen(pscreen);
+
+        return screen->fd;
+}
+
 struct pipe_screen *
-vc4_screen_create(int fd, struct renderonly *ro)
+vc4_screen_create(int fd, const struct pipe_screen_config *config,
+                  struct renderonly *ro)
 {
         struct vc4_screen *screen = rzalloc(NULL, struct vc4_screen);
         uint64_t syncobj_cap = 0;
@@ -548,6 +555,7 @@ vc4_screen_create(int fd, struct renderonly *ro)
         pscreen = &screen->base;
 
         pscreen->destroy = vc4_screen_destroy;
+        pscreen->get_screen_fd = vc4_screen_get_fd;
         pscreen->get_param = vc4_screen_get_param;
         pscreen->get_paramf = vc4_screen_get_paramf;
         pscreen->get_shader_param = vc4_screen_get_shader_param;
diff --git a/lib/mesa/src/gallium/drivers/vc4/vc4_screen.h b/lib/mesa/src/gallium/drivers/vc4/vc4_screen.h
index 52e19a32b..aa4506eb5 100644
--- a/lib/mesa/src/gallium/drivers/vc4/vc4_screen.h
+++ b/lib/mesa/src/gallium/drivers/vc4/vc4_screen.h
@@ -26,7 +26,7 @@
 
 #include "pipe/p_screen.h"
 #include "renderonly/renderonly.h"
-#include "os/os_thread.h"
+#include "util/u_thread.h"
 #include "frontend/drm_driver.h"
 #include "util/list.h"
 #include "util/slab.h"
@@ -114,7 +114,9 @@ vc4_screen(struct pipe_screen *screen)
         return (struct vc4_screen *)screen;
 }
 
-struct pipe_screen *vc4_screen_create(int fd, struct renderonly *ro);
+struct pipe_screen *vc4_screen_create(int fd,
+                                      const struct pipe_screen_config *config,
+                                      struct renderonly *ro);
 
 const void *
 vc4_screen_get_compiler_options(struct pipe_screen *pscreen,