Mesa (staging/20.2): radeonsi: remove the NGG hack decreasing LDS usage to deal with overflows

[GitLab Mirror]

Module: Mesa
Branch: staging/20.2
Commit: ee4ccf003114e7200a693512c900aefc9bc4a2cf
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=ee4ccf003114e7200a693512c900aefc9bc4a2cf

Author: Marek Olšák <marek.olsak at amd.com>
Date:   Thu Jul 30 08:32:15 2020 -0400

radeonsi: remove the NGG hack decreasing LDS usage to deal with overflows

The LDS size can't overflow anymore, so we can use the correct max LDS size.

Fixes: a23802bcb9a - ac,radeonsi: start adding support for gfx10.3

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer at amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6137>
(cherry picked from commit e2e700f6053d0b16ba46e4d5c5b20e965fb2224e)

---

 .pick_status.json                               | 2 +-
 src/gallium/drivers/radeonsi/gfx10_shader_ngg.c | 9 ++-------
 2 files changed, 3 insertions(+), 8 deletions(-)

diff --git a/.pick_status.json b/.pick_status.json
index 601b5bd96bd..33b2d315036 100644
--- a/.pick_status.json
+++ b/.pick_status.json
@@ -22,7 +22,7 @@
         "description": "radeonsi: remove the NGG hack decreasing LDS usage to deal with overflows",
         "nominated": true,
         "nomination_type": 1,
-        "resolution": 0,
+        "resolution": 1,
         "master_sha": null,
         "because_sha": "a23802bcb9a42a02d34a5a36d6e66d6532813a0d"
     },
diff --git a/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c b/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c
index b484ebae522..0654ab32855 100644
--- a/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c
+++ b/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c
@@ -1917,14 +1917,9 @@ bool gfx10_ngg_calculate_subgroup_info(struct si_shader *shader)
    const unsigned min_verts_per_prim = gs_type == PIPE_SHADER_GEOMETRY ? max_verts_per_prim : 1;
 
    /* All these are in dwords: */
-   /* We can't allow using the whole LDS, because GS waves compete with
-    * other shader stages for LDS space.
-    *
-    * TODO: We should really take the shader's internal LDS use into
-    *       account. The linker will fail if the size is greater than
-    *       8K dwords.
+   /* GE can only use 8K dwords (32KB) of LDS per workgroup.
     */
-   const unsigned max_lds_size = 8 * 1024 - 768;
+   const unsigned max_lds_size = 8 * 1024 - gfx10_ngg_get_scratch_dw_size(shader);
    const unsigned target_lds_size = max_lds_size;
    unsigned esvert_lds_size = 0;
    unsigned gsprim_lds_size = 0;