winsys/amdgpu: reduce amdgpu_cs size

buffer_indices_hashlist is only used by the current
amdgpu_cs_context (= amdgpu_cs.csc).

So store a single 16k int array instead of 2, and switch
the owner when flushing the cs.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11010>

2 files changed, 17 insertions, 8 deletions

diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
index e2951eccd67..22f5671dfa4 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
@@ -425,7 +425,7 @@ static inline unsigned amdgpu_cs_epilog_dws(struct amdgpu_cs *cs)
 static int amdgpu_lookup_buffer(struct amdgpu_cs_context *cs, struct amdgpu_winsys_bo *bo,
                                 struct amdgpu_cs_buffer *buffers, unsigned num_buffers)
 {
-   unsigned hash = bo->unique_id & (ARRAY_SIZE(cs->buffer_indices_hashlist)-1);
+   unsigned hash = bo->unique_id & (BUFFER_HASHLIST_SIZE-1);
    int i = cs->buffer_indices_hashlist[hash];
 
    /* not found or found */
@@ -522,7 +522,7 @@ amdgpu_lookup_or_add_real_buffer(struct radeon_cmdbuf *rcs, struct amdgpu_cs *ac
 
    idx = amdgpu_do_add_real_buffer(acs->ws, cs, bo);
 
-   hash = bo->unique_id & (ARRAY_SIZE(cs->buffer_indices_hashlist)-1);
+   hash = bo->unique_id & (BUFFER_HASHLIST_SIZE-1);
    cs->buffer_indices_hashlist[hash] = idx;
 
    if (bo->base.placement & RADEON_DOMAIN_VRAM)
@@ -577,7 +577,7 @@ static int amdgpu_lookup_or_add_slab_buffer(struct amdgpu_winsys *ws,
    buffer->u.slab.real_idx = real_idx;
    cs->num_slab_buffers++;
 
-   hash = bo->unique_id & (ARRAY_SIZE(cs->buffer_indices_hashlist)-1);
+   hash = bo->unique_id & (BUFFER_HASHLIST_SIZE-1);
    cs->buffer_indices_hashlist[hash] = idx;
 
    return idx;
@@ -621,7 +621,7 @@ static int amdgpu_lookup_or_add_sparse_buffer(struct amdgpu_winsys *ws,
    amdgpu_winsys_bo_reference(ws, &buffer->bo, bo);
    cs->num_sparse_buffers++;
 
-   hash = bo->unique_id & (ARRAY_SIZE(cs->buffer_indices_hashlist)-1);
+   hash = bo->unique_id & (BUFFER_HASHLIST_SIZE-1);
    cs->buffer_indices_hashlist[hash] = idx;
 
    /* We delay adding the backing buffers until we really have to. However,
@@ -911,7 +911,6 @@ static bool amdgpu_init_cs_context(struct amdgpu_winsys *ws,
    cs->ib[IB_PARALLEL_COMPUTE].ip_type = AMDGPU_HW_IP_COMPUTE;
    cs->ib[IB_PARALLEL_COMPUTE].flags = AMDGPU_IB_FLAG_TC_WB_NOT_INVALIDATE;
 
-   memset(cs->buffer_indices_hashlist, -1, sizeof(cs->buffer_indices_hashlist));
    cs->last_added_bo = NULL;
    return true;
 }
@@ -946,8 +945,6 @@ static void amdgpu_cs_context_cleanup(struct amdgpu_winsys *ws, struct amdgpu_cs
    cs->num_slab_buffers = 0;
    cs->num_sparse_buffers = 0;
    amdgpu_fence_reference(&cs->fence, NULL);
-
-   memset(cs->buffer_indices_hashlist, -1, sizeof(cs->buffer_indices_hashlist));
    cs->last_added_bo = NULL;
 }
 
@@ -1013,10 +1010,16 @@ amdgpu_cs_create(struct radeon_cmdbuf *rcs,
       return false;
    }
 
+   memset(cs->buffer_indices_hashlist, -1, sizeof(cs->buffer_indices_hashlist));
+
    /* Set the first submission context as current. */
    cs->csc = &cs->csc1;
    cs->cst = &cs->csc2;
 
+   /* Assign to both amdgpu_cs_context; only csc will use it. */
+   cs->csc1.buffer_indices_hashlist = cs->buffer_indices_hashlist;
+   cs->csc2.buffer_indices_hashlist = cs->buffer_indices_hashlist;
+
    cs->main.rcs = rcs;
    rcs->priv = cs;
 
@@ -1892,6 +1895,8 @@ static int amdgpu_cs_flush(struct radeon_cmdbuf *rcs,
       amdgpu_cs_context_cleanup(ws, cs->csc);
    }
 
+   memset(cs->csc->buffer_indices_hashlist, -1, sizeof(cs->buffer_indices_hashlist));
+
    amdgpu_get_new_ib(ws, rcs, &cs->main, cs);
    if (cs->compute_ib.ib_mapped)
       amdgpu_get_new_ib(ws, cs->compute_ib.rcs, &cs->compute_ib, cs);
diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h
index 28dbad83b3f..4683ed9b210 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h
@@ -104,7 +104,7 @@ struct amdgpu_cs_context {
    unsigned                    max_sparse_buffers;
    struct amdgpu_cs_buffer     *sparse_buffers;
 
-   int                         buffer_indices_hashlist[4096];
+   int                         *buffer_indices_hashlist;
 
    struct amdgpu_winsys_bo     *last_added_bo;
    unsigned                    last_added_bo_index;
@@ -128,6 +128,8 @@ struct amdgpu_cs_context {
    bool secure;
 };
 
+#define BUFFER_HASHLIST_SIZE 4096
+
 struct amdgpu_cs {
    struct amdgpu_ib main; /* must be first because this is inherited */
    struct amdgpu_ib compute_ib;      /* optional parallel compute IB */
@@ -145,6 +147,8 @@ struct amdgpu_cs {
    struct amdgpu_cs_context *csc;
    /* The CS being currently-owned by the other thread. */
    struct amdgpu_cs_context *cst;
+   /* This is only used by csc, not cst */
+   int buffer_indices_hashlist[BUFFER_HASHLIST_SIZE];
 
    /* Flush CS. */
    void (*flush_cs)(void *ctx, unsigned flags, struct pipe_fence_handle **fence);