nouveau: support for copy-less pushbuf ioctl

diff --git a/libdrm/nouveau/nouveau_private.h b/libdrm/nouveau/nouveau_private.h
index 49dde5e..67144e3 100644 (file)
--- a/libdrm/nouveau/nouveau_private.h
+++ b/libdrm/nouveau/nouveau_private.h
@@ -36,11 +36,17 @@
 #include "nouveau_resource.h"
 #include "nouveau_pushbuf.h"
 
+#define CALPB_BUFFERS 4
+#define CALPB_BUFSZ   16384
 struct nouveau_pushbuf_priv {
        struct nouveau_pushbuf base;
 
        int use_cal;
-       struct nouveau_bo *buffer;
+       uint32_t cal_suffix0;
+       uint32_t cal_suffix1;
+       struct nouveau_bo *buffer[CALPB_BUFFERS];
+       int current;
+       int current_offset;
 
        unsigned *pushbuf;
        unsigned  size;

diff --git a/libdrm/nouveau/nouveau_pushbuf.c b/libdrm/nouveau/nouveau_pushbuf.c
index 480dbd2..86d5a4e 100644 (file)
--- a/libdrm/nouveau/nouveau_pushbuf.c
+++ b/libdrm/nouveau/nouveau_pushbuf.c
@@ -59,7 +59,6 @@ nouveau_pushbuf_emit_reloc(struct nouveau_channel *chan, void *ptr,
                           struct nouveau_bo *bo, uint32_t data, uint32_t data2,
                           uint32_t flags, uint32_t vor, uint32_t tor)
 {
-       struct nouveau_device_priv *nvdev = nouveau_device(chan->device);
        struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(chan->pushbuf);
        struct drm_nouveau_gem_pushbuf_reloc *r;
        struct drm_nouveau_gem_pushbuf_bo *pbbo;
@@ -119,11 +118,50 @@ nouveau_pushbuf_emit_reloc(struct nouveau_channel *chan, void *ptr,
 }
 
 static int
+nouveau_pushbuf_space_call(struct nouveau_channel *chan, unsigned min)
+{
+       struct nouveau_channel_priv *nvchan = nouveau_channel(chan);
+       struct nouveau_pushbuf_priv *nvpb = &nvchan->pb;
+       struct nouveau_bo *bo;
+       int ret;
+
+       if (min < PB_MIN_USER_DWORDS)
+               min = PB_MIN_USER_DWORDS;
+
+       nvpb->current_offset = nvpb->base.cur - nvpb->pushbuf;
+       if (nvpb->current_offset + min + 2 <= nvpb->size)
+               return 0;
+
+       nvpb->current++;
+       if (nvpb->current == CALPB_BUFFERS)
+               nvpb->current = 0;
+       bo = nvpb->buffer[nvpb->current];
+
+       ret = nouveau_bo_map(bo, NOUVEAU_BO_WR);
+       if (ret)
+               return ret;
+
+       nvpb->size = (bo->size - 8) / 4;
+       nvpb->pushbuf = bo->map;
+       nvpb->current_offset = 0;
+
+       nvpb->base.channel = chan;
+       nvpb->base.remaining = nvpb->size;
+       nvpb->base.cur = nvpb->pushbuf;
+
+       nouveau_bo_unmap(bo);
+       return 0;
+}
+
+static int
 nouveau_pushbuf_space(struct nouveau_channel *chan, unsigned min)
 {
        struct nouveau_channel_priv *nvchan = nouveau_channel(chan);
        struct nouveau_pushbuf_priv *nvpb = &nvchan->pb;
 
+       if (nvpb->use_cal)
+               return nouveau_pushbuf_space_call(chan, min);
+
        if (nvpb->pushbuf) {
                free(nvpb->pushbuf);
                nvpb->pushbuf = NULL;
@@ -139,13 +177,69 @@ nouveau_pushbuf_space(struct nouveau_channel *chan, unsigned min)
        return 0;
 }
 
+static void
+nouveau_pushbuf_fini_call(struct nouveau_channel *chan)
+{
+       struct nouveau_channel_priv *nvchan = nouveau_channel(chan);
+       struct nouveau_pushbuf_priv *nvpb = &nvchan->pb;
+       int i;
+
+       for (i = 0; i < CALPB_BUFFERS; i++)
+               nouveau_bo_ref(NULL, &nvpb->buffer[i]);
+       nvpb->use_cal = 0;
+       nvpb->pushbuf = NULL;
+}
+
+static void
+nouveau_pushbuf_init_call(struct nouveau_channel *chan)
+{
+       struct drm_nouveau_gem_pushbuf_call req;
+       struct nouveau_channel_priv *nvchan = nouveau_channel(chan);
+       struct nouveau_pushbuf_priv *nvpb = &nvchan->pb;
+       struct nouveau_device *dev = chan->device;
+       int i, ret;
+
+       req.channel = chan->id;
+       req.handle = 0;
+       ret = drmCommandWriteRead(nouveau_device(dev)->fd,
+                                 DRM_NOUVEAU_GEM_PUSHBUF_CALL,
+                                 &req, sizeof(req));
+       if (ret)
+               return;
+
+       for (i = 0; i < CALPB_BUFFERS; i++) {
+               ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP,
+                                    0, CALPB_BUFSZ, &nvpb->buffer[i]);
+               if (ret) {
+                       nouveau_pushbuf_fini_call(chan);
+                       return;
+               }
+       }
+
+       nvpb->use_cal = 1;
+       nvpb->cal_suffix0 = req.suffix0;
+       nvpb->cal_suffix1 = req.suffix1;
+}
+
 int
 nouveau_pushbuf_init(struct nouveau_channel *chan)
 {
        struct nouveau_channel_priv *nvchan = nouveau_channel(chan);
        struct nouveau_pushbuf_priv *nvpb = &nvchan->pb;
+       int ret;
 
-       nouveau_pushbuf_space(chan, 0);
+       nouveau_pushbuf_init_call(chan);
+
+       ret = nouveau_pushbuf_space(chan, 0);
+       if (ret) {
+               if (nvpb->use_cal) {
+                       nouveau_pushbuf_fini_call(chan);
+                       ret = nouveau_pushbuf_space(chan, 0);
+               }
+
+               if (ret)
+                       return ret;
+       }
 
        nvpb->buffers = calloc(NOUVEAU_GEM_MAX_BUFFERS,
                               sizeof(struct drm_nouveau_gem_pushbuf_bo));
@@ -162,24 +256,49 @@ nouveau_pushbuf_flush(struct nouveau_channel *chan, unsigned min)
        struct nouveau_device_priv *nvdev = nouveau_device(chan->device);
        struct nouveau_channel_priv *nvchan = nouveau_channel(chan);
        struct nouveau_pushbuf_priv *nvpb = &nvchan->pb;
-       struct drm_nouveau_gem_pushbuf req;
        unsigned i;
        int ret;
 
        if (nvpb->base.remaining == nvpb->size)
                return 0;
-       nvpb->size -= nvpb->base.remaining;
 
-       req.channel = chan->id;
-       req.nr_dwords = nvpb->size;
-       req.dwords = (uint64_t)(unsigned long)nvpb->pushbuf;
-       req.nr_buffers = nvpb->nr_buffers;
-       req.buffers = (uint64_t)(unsigned long)nvpb->buffers;
-       req.nr_relocs = nvpb->nr_relocs;
-       req.relocs = (uint64_t)(unsigned long)nvpb->relocs;
-       ret = drmCommandWrite(nvdev->fd, DRM_NOUVEAU_GEM_PUSHBUF,
-                             &req, sizeof(req));
-       assert(ret == 0);
+       if (nvpb->use_cal) {
+               struct drm_nouveau_gem_pushbuf_call req;
+
+               *(nvpb->base.cur++) = nvpb->cal_suffix0;
+               *(nvpb->base.cur++) = nvpb->cal_suffix1;
+
+               req.channel = chan->id;
+               req.handle = nvpb->buffer[nvpb->current]->handle;
+               req.offset = nvpb->current_offset * 4;
+               req.nr_buffers = nvpb->nr_buffers;
+               req.buffers = (uint64_t)(unsigned long)nvpb->buffers;
+               req.nr_relocs = nvpb->nr_relocs;
+               req.relocs = (uint64_t)(unsigned long)nvpb->relocs;
+               req.nr_dwords = (nvpb->base.cur - nvpb->pushbuf) -
+                               nvpb->current_offset;
+               req.suffix0 = nvpb->cal_suffix0;
+               req.suffix1 = nvpb->cal_suffix1;
+               ret = drmCommandWriteRead(nvdev->fd,
+                                         DRM_NOUVEAU_GEM_PUSHBUF_CALL,
+                                         &req, sizeof(req));
+               nvpb->cal_suffix0 = req.suffix0;
+               nvpb->cal_suffix1 = req.suffix1;
+               assert(ret == 0);
+       } else {
+               struct drm_nouveau_gem_pushbuf req;
+
+               req.channel = chan->id;
+               req.nr_dwords = nvpb->size - nvpb->base.remaining;
+               req.dwords = (uint64_t)(unsigned long)nvpb->pushbuf;
+               req.nr_buffers = nvpb->nr_buffers;
+               req.buffers = (uint64_t)(unsigned long)nvpb->buffers;
+               req.nr_relocs = nvpb->nr_relocs;
+               req.relocs = (uint64_t)(unsigned long)nvpb->relocs;
+               ret = drmCommandWrite(nvdev->fd, DRM_NOUVEAU_GEM_PUSHBUF,
+                                     &req, sizeof(req));
+               assert(ret == 0);
+       }
 
 
        /* Update presumed offset/domain for any buffers that moved.

diff --git a/shared-core/nouveau_drm.h b/shared-core/nouveau_drm.h
index 3e52b24..2050357 100644 (file)
--- a/shared-core/nouveau_drm.h
+++ b/shared-core/nouveau_drm.h
@@ -150,9 +150,11 @@ struct drm_nouveau_gem_pushbuf_call {
        uint32_t offset;
        uint32_t nr_buffers;
        uint32_t nr_relocs;
-       uint32_t pad0;
+       uint32_t nr_dwords;
        uint64_t buffers;
        uint64_t relocs;
+       uint32_t suffix0;
+       uint32_t suffix1;
 };
 
 struct drm_nouveau_gem_pin {