drm/amdgpu: add CE preamble flag v3

The CE preamble IB can be dropped for the same context

v2: use the flags directly
v3: remove 'CE' for potential preamble usage by other rings

Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
index 7428c43..cec46eb 100644 (file)
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
@@ -2518,6 +2518,13 @@ static void gfx_v7_0_ring_emit_ib(struct amdgpu_ring *ring,
 {
        u32 header, control = 0;
        u32 next_rptr = ring->wptr + 5;
+
+       /* drop the CE preamble IB for the same context */
+       if ((ring->type == AMDGPU_RING_TYPE_GFX) &&
+           (ib->flags & AMDGPU_IB_FLAG_PREAMBLE) &&
+           !ring->need_ctx_switch)
+               return;
+
        if (ring->type == AMDGPU_RING_TYPE_COMPUTE)
                control |= INDIRECT_BUFFER_VALID;
 

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 48de920..fc8c462 100644 (file)
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -3647,6 +3647,13 @@ static void gfx_v8_0_ring_emit_ib(struct amdgpu_ring *ring,
 {
        u32 header, control = 0;
        u32 next_rptr = ring->wptr + 5;
+
+       /* drop the CE preamble IB for the same context */
+       if ((ring->type == AMDGPU_RING_TYPE_GFX) &&
+           (ib->flags & AMDGPU_IB_FLAG_PREAMBLE) &&
+           !ring->need_ctx_switch)
+               return;
+
        if (ring->type == AMDGPU_RING_TYPE_COMPUTE)
                control |= INDIRECT_BUFFER_VALID;
 

diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
index ca0ea1e..fb428fe 100644 (file)
--- a/include/uapi/drm/amdgpu_drm.h
+++ b/include/uapi/drm/amdgpu_drm.h
@@ -369,6 +369,9 @@ union drm_amdgpu_cs {
 /* GDS is used by this IB */
 #define AMDGPU_IB_FLAG_GDS     (1<<1)
 
+/* CE Preamble */
+#define AMDGPU_IB_FLAG_PREAMBLE (1<<2)
+
 struct drm_amdgpu_cs_chunk_ib {
        /**
         * Handle of GEM object to be used as IB or 0 if it is already in