[Mesa-dev] [PATCH v3 04/12] winsys/amdgpu: Enlarge const IB size.

[Bas Nieuwenhuizen]

Necessary to prevent performance regressions due to extra flushing.

Probably should enlarge it even further when also updating
uniforms through the CE, but this seems large enough for now.

v2: Add preamble IB.

Signed-off-by: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak at amd.com>
---
 src/gallium/winsys/amdgpu/drm/amdgpu_cs.c | 28 ++++++++++++++++++++--------
 1 file changed, 20 insertions(+), 8 deletions(-)

diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
index 0182660..69902c4 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
@@ -199,14 +199,26 @@ amdgpu_ctx_query_reset_status(struct radeon_winsys_ctx *rwctx)
 /* COMMAND SUBMISSION */
 
 static bool amdgpu_get_new_ib(struct radeon_winsys *ws, struct amdgpu_ib *ib,
-                              struct amdgpu_cs_ib_info *info)
+                              struct amdgpu_cs_ib_info *info, unsigned ib_type)
 {
    /* Small IBs are better than big IBs, because the GPU goes idle quicker
     * and there is less waiting for buffers and fences. Proof:
     *   http://www.phoronix.com/scan.php?page=article&item=mesa-111-si&num=1
     */
-   const unsigned buffer_size = 128 * 1024 * 4;
-   const unsigned ib_size = 20 * 1024 * 4;
+   unsigned buffer_size, ib_size;
+
+   switch (ib_type) {
+   case IB_CONST_PREAMBLE:
+      buffer_size = 4 * 1024 * 4;
+      ib_size = 1024 * 4;
+   case IB_CONST:
+      buffer_size = 512 * 1024 * 4;
+      ib_size = 128 * 1024 * 4;
+      break;
+   case IB_MAIN:
+      buffer_size = 128 * 1024 * 4;
+      ib_size = 20 * 1024 * 4;
+   }
 
    ib->base.cdw = 0;
    ib->base.buf = NULL;
@@ -350,7 +362,7 @@ amdgpu_cs_create(struct radeon_winsys_ctx *rwctx,
       return NULL;
    }
 
-   if (!amdgpu_get_new_ib(&ctx->ws->base, &cs->main, &cs->ib[IB_MAIN])) {
+   if (!amdgpu_get_new_ib(&ctx->ws->base, &cs->main, &cs->ib[IB_MAIN], IB_MAIN)) {
       amdgpu_destroy_cs_context(cs);
       FREE(cs);
       return NULL;
@@ -373,7 +385,7 @@ amdgpu_cs_add_const_ib(struct radeon_winsys_cs *rcs)
    if (cs->ring_type != RING_GFX || cs->const_ib.ib_mapped)
       return NULL;
 
-   if (!amdgpu_get_new_ib(&ws->base, &cs->const_ib, &cs->ib[IB_CONST]))
+   if (!amdgpu_get_new_ib(&ws->base, &cs->const_ib, &cs->ib[IB_CONST], IB_CONST))
       return NULL;
 
    cs->request.number_of_ibs = 2;
@@ -760,12 +772,12 @@ static void amdgpu_cs_flush(struct radeon_winsys_cs *rcs,
 cleanup:
    amdgpu_cs_context_cleanup(cs);
 
-   amdgpu_get_new_ib(&ws->base, &cs->main, &cs->ib[IB_MAIN]);
+   amdgpu_get_new_ib(&ws->base, &cs->main, &cs->ib[IB_MAIN], IB_MAIN);
    if (cs->const_ib.ib_mapped)
-      amdgpu_get_new_ib(&ws->base, &cs->const_ib, &cs->ib[IB_CONST]);
+      amdgpu_get_new_ib(&ws->base, &cs->const_ib, &cs->ib[IB_CONST], IB_CONST);
    if (cs->const_preamble_ib.ib_mapped)
       amdgpu_get_new_ib(&ws->base, &cs->const_preamble_ib,
-                                 &cs->ib[IB_CONST_PREAMBLE]);
+                                 &cs->ib[IB_CONST_PREAMBLE], IB_CONST_PREAMBLE);
 
    ws->num_cs_flushes++;
 }
-- 
2.8.0