Mesa (master): winsys/amdgpu: don' t do read-modify-write on command buffers

[Marek Olšák]

Module: Mesa
Branch: master
Commit: 49f5ce39c10a289166df9d59c957e347925ca951
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=49f5ce39c10a289166df9d59c957e347925ca951

Author: Marek Olšák <marek.olsak at amd.com>
Date:   Sun Oct  8 23:47:30 2017 +0200

winsys/amdgpu: don't do read-modify-write on command buffers

i.e. don't use |=

Reviewed-by: Nicolai Hähnle <nicolai.haehnle at amd.com>

---

 src/gallium/winsys/amdgpu/drm/amdgpu_cs.c | 19 +++++++++++++++----
 src/gallium/winsys/amdgpu/drm/amdgpu_cs.h |  1 +
 2 files changed, 16 insertions(+), 4 deletions(-)

diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
index d9d2a8b923..0a657f7505 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
@@ -725,6 +725,7 @@ static bool amdgpu_get_new_ib(struct radeon_winsys *ws, struct amdgpu_cs *cs,
    /* ib_bytes is in dwords and the conversion to bytes will be done before
     * the CS ioctl. */
    ib->ptr_ib_size = &info->ib_bytes;
+   ib->ptr_ib_size_inside_ib = false;
 
    amdgpu_cs_add_buffer(&cs->main.base, ib->big_ib_buffer,
                         RADEON_USAGE_READ, 0, RADEON_PRIO_IB1);
@@ -736,9 +737,19 @@ static bool amdgpu_get_new_ib(struct radeon_winsys *ws, struct amdgpu_cs *cs,
    return true;
 }
 
+static void amdgpu_set_ib_size(struct amdgpu_ib *ib)
+{
+   if (ib->ptr_ib_size_inside_ib) {
+      *ib->ptr_ib_size = ib->base.current.cdw |
+                         S_3F2_CHAIN(1) | S_3F2_VALID(1);
+   } else {
+      *ib->ptr_ib_size = ib->base.current.cdw;
+   }
+}
+
 static void amdgpu_ib_finalize(struct amdgpu_ib *ib)
 {
-   *ib->ptr_ib_size |= ib->base.current.cdw;
+   amdgpu_set_ib_size(ib);
    ib->used_ib_space += ib->base.current.cdw * 4;
    ib->max_ib_size = MAX2(ib->max_ib_size, ib->base.prev_dw + ib->base.current.cdw);
 }
@@ -933,14 +944,14 @@ static bool amdgpu_cs_check_space(struct radeon_winsys_cs *rcs, unsigned dw)
                                            : PKT3_INDIRECT_BUFFER_CONST, 2, 0));
    radeon_emit(rcs, va);
    radeon_emit(rcs, va >> 32);
-   new_ptr_ib_size = &rcs->current.buf[rcs->current.cdw];
-   radeon_emit(rcs, S_3F2_CHAIN(1) | S_3F2_VALID(1));
+   new_ptr_ib_size = &rcs->current.buf[rcs->current.cdw++];
 
    assert((rcs->current.cdw & 7) == 0);
    assert(rcs->current.cdw <= rcs->current.max_dw);
 
-   *ib->ptr_ib_size |= rcs->current.cdw;
+   amdgpu_set_ib_size(ib);
    ib->ptr_ib_size = new_ptr_ib_size;
+   ib->ptr_ib_size_inside_ib = true;
 
    /* Hook up the new chunk */
    rcs->prev[rcs->num_prev].buf = rcs->current.buf;
diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h
index 21e13547fc..1c3d0f0be1 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h
@@ -73,6 +73,7 @@ struct amdgpu_ib {
    unsigned                used_ib_space;
    unsigned                max_ib_size;
    uint32_t                *ptr_ib_size;
+   bool                    ptr_ib_size_inside_ib;
    enum ib_type            ib_type;
 };