freedreno: Move/add some attach_bo()

In some cases we were missing this, in others we just needed to move it
before the OUT_RELOC().

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25465>

diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_emit.cc b/src/gallium/drivers/freedreno/a6xx/fd6_emit.cc
index bd639bd4042c7d5609d2e91926f37b81a7060625..83b2c271d207a7d5c73daea8f44e16a0dd703ac2 100644 (file)
--- a/src/gallium/drivers/freedreno/a6xx/fd6_emit.cc
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_emit.cc
@@ -976,6 +976,9 @@ fd6_mem_to_mem(struct fd_ringbuffer *ring, struct pipe_resource *dst,
    struct fd_bo *dst_bo = fd_resource(dst)->bo;
    unsigned i;
 
+   fd_ringbuffer_attach_bo(ring, dst_bo);
+   fd_ringbuffer_attach_bo(ring, src_bo);
+
    for (i = 0; i < sizedwords; i++) {
       OUT_PKT7(ring, CP_MEM_TO_MEM, 5);
       OUT_RING(ring, 0x00000000);
@@ -985,8 +988,6 @@ fd6_mem_to_mem(struct fd_ringbuffer *ring, struct pipe_resource *dst,
       dst_off += 4;
       src_off += 4;
    }
-   fd_ringbuffer_attach_bo(ring, dst_bo);
-   fd_ringbuffer_attach_bo(ring, src_bo);
 }
 
 void

diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_gmem.cc b/src/gallium/drivers/freedreno/a6xx/fd6_gmem.cc
index e6a7463f4e7f8467682e4987329d28b9f56356c5..3ab87635ef98aed889386247bed79729bd0da9f7 100644 (file)
--- a/src/gallium/drivers/freedreno/a6xx/fd6_gmem.cc
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_gmem.cc
@@ -126,6 +126,9 @@ emit_mrt(struct fd_ringbuffer *ring, struct pipe_framebuffer_state *pfb,
 
       assert((offset + slice->size0) <= fd_bo_size(rsc->bo));
 
+      /* Batch with no draws? */
+      fd_ringbuffer_attach_bo(ring, rsc->bo);
+
       OUT_REG(
          ring,
          RB_MRT_BUF_INFO(CHIP, i, .color_format = format,
@@ -170,6 +173,12 @@ emit_zs(struct fd_ringbuffer *ring, struct pipe_surface *zsbuf,
       uint32_t offset =
          fd_resource_offset(rsc, zsbuf->u.tex.level, zsbuf->u.tex.first_layer);
 
+      /* We could have a depth buffer, but no draws with depth write/test
+       * enabled, in which case it wouldn't have been part of the batch
+       * resource tracking
+       */
+      fd_ringbuffer_attach_bo(ring, rsc->bo);
+
       OUT_REG(
          ring, RB_DEPTH_BUFFER_INFO(CHIP, .depth_format = fmt),
          A6XX_RB_DEPTH_BUFFER_PITCH(stride),
@@ -196,6 +205,8 @@ emit_zs(struct fd_ringbuffer *ring, struct pipe_surface *zsbuf,
          uint32_t offset =
             fd_resource_offset(rsc->stencil, zsbuf->u.tex.level, zsbuf->u.tex.first_layer);
 
+         fd_ringbuffer_attach_bo(ring, rsc->stencil->bo);
+
          OUT_REG(ring, RB_STENCIL_INFO(CHIP, .separate_stencil = true),
                  A6XX_RB_STENCIL_BUFFER_PITCH(stride),
                  A6XX_RB_STENCIL_BUFFER_ARRAY_PITCH(array_stride),
@@ -274,6 +285,13 @@ emit_lrz_clears(struct fd_batch *batch)
    struct fd_resource *zsbuf = fd_resource(pfb->zsbuf->texture);
 
    foreach_subpass (subpass, batch) {
+      /* The lrz buffer isn't explicitly tracked by the batch resource
+       * tracking (tracking the zsbuf is sufficient), but it still needs
+       * to be attached to the ring
+       */
+      if (subpass->lrz)
+         fd_ringbuffer_attach_bo(batch->gmem, subpass->lrz);
+
       if (!(subpass->fast_cleared & FD_BUFFER_LRZ))
          continue;
 
@@ -545,6 +563,9 @@ update_vsc_pipe(struct fd_batch *batch)
          fd_bo_new(ctx->screen->dev, sz, FD_BO_NOMAP, "vsc_prim_strm");
    }
 
+   fd_ringbuffer_attach_bo(ring, fd6_ctx->vsc_draw_strm);
+   fd_ringbuffer_attach_bo(ring, fd6_ctx->vsc_prim_strm);
+
    OUT_REG(ring, A6XX_VSC_BIN_SIZE(.width = gmem->bin_w, .height = gmem->bin_h),
            A6XX_VSC_DRAW_STRM_SIZE_ADDRESS(.bo = fd6_ctx->vsc_draw_strm,
                                            .bo_offset = max_vsc_pipes *
@@ -570,9 +591,6 @@ update_vsc_pipe(struct fd_batch *batch)
       ring, A6XX_VSC_DRAW_STRM_ADDRESS(.bo = fd6_ctx->vsc_draw_strm),
       A6XX_VSC_DRAW_STRM_PITCH(.dword = fd6_ctx->vsc_draw_strm_pitch),
       A6XX_VSC_DRAW_STRM_LIMIT(.dword = fd6_ctx->vsc_draw_strm_pitch - 64));
-
-   fd_ringbuffer_attach_bo(ring, fd6_ctx->vsc_draw_strm);
-   fd_ringbuffer_attach_bo(ring, fd6_ctx->vsc_prim_strm);
 }
 
 /*
@@ -697,6 +715,8 @@ emit_common_init(struct fd_batch *batch)
    if (!result)
       return;
 
+   fd_ringbuffer_attach_bo(ring, at->results_mem);
+
    OUT_PKT4(ring, REG_A6XX_RB_SAMPLE_COUNT_CONTROL, 1);
    OUT_RING(ring, A6XX_RB_SAMPLE_COUNT_CONTROL_COPY);
 

diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_program.cc b/src/gallium/drivers/freedreno/a6xx/fd6_program.cc
index d02b350cadeeb04333454ff633bedb08e7e82e20..53aee8b7a7ef125352e46c01f13ef013c446f354 100644 (file)
--- a/src/gallium/drivers/freedreno/a6xx/fd6_program.cc
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_program.cc
@@ -195,13 +195,15 @@ fd6_emit_shader(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
    uint32_t per_sp_size = ctx->pvtmem[so->pvtmem_per_wave].per_sp_size;
 
+   fd_ringbuffer_attach_bo(ring, so->bo);
+
    OUT_PKT4(ring, cfg->reg_sp_xs_first_exec_offset, 7);
    OUT_RING(ring, 0);                /* SP_xS_OBJ_FIRST_EXEC_OFFSET */
    OUT_RELOC(ring, so->bo, 0, 0, 0); /* SP_xS_OBJ_START_LO */
    OUT_RING(ring, A6XX_SP_VS_PVT_MEM_PARAM_MEMSIZEPERITEM(ctx->pvtmem[so->pvtmem_per_wave].per_fiber_size));
    if (so->pvtmem_size > 0) { /* SP_xS_PVT_MEM_ADDR */
-      OUT_RELOC(ring, ctx->pvtmem[so->pvtmem_per_wave].bo, 0, 0, 0);
       fd_ringbuffer_attach_bo(ring, ctx->pvtmem[so->pvtmem_per_wave].bo);
+      OUT_RELOC(ring, ctx->pvtmem[so->pvtmem_per_wave].bo, 0, 0, 0);
    } else {
       OUT_RING(ring, 0);
       OUT_RING(ring, 0);
@@ -225,8 +227,6 @@ fd6_emit_shader(struct fd_context *ctx, struct fd_ringbuffer *ring,
                      CP_LOAD_STATE6_0_NUM_UNIT(shader_preload_size));
    OUT_RELOC(ring, so->bo, 0, 0, 0);
 
-   fd_ringbuffer_attach_bo(ring, so->bo);
-
    fd6_emit_immediates(so, ring);
 }
 
@@ -467,6 +467,8 @@ fd6_emit_tess_bos(struct fd_screen *screen, struct fd_ringbuffer *ring,
    if (regid >= s->constlen)
       return;
 
+   fd_ringbuffer_attach_bo(ring, screen->tess_bo);
+
    OUT_PKT7(ring, fd6_stage2opcode(s->type), 7);
    OUT_RING(ring, CP_LOAD_STATE6_0_DST_OFF(regid) |
                      CP_LOAD_STATE6_0_STATE_TYPE(ST6_CONSTANTS) |

diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_query.cc b/src/gallium/drivers/freedreno/a6xx/fd6_query.cc
index a64c6dad381f6277058b7c4d05ee94ba2506e987..8cb81141ebd1d09778a86b1335a35ca2a9bf5030 100644 (file)
--- a/src/gallium/drivers/freedreno/a6xx/fd6_query.cc
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_query.cc
@@ -703,6 +703,9 @@ so_overflow_predicate_result_resource(struct fd_acc_query *aq,
                                       int index, struct fd_resource *dst,
                                       unsigned offset)
 {
+   fd_ringbuffer_attach_bo(ring, dst->bo);
+   fd_ringbuffer_attach_bo(ring, fd_resource(aq->prsc)->bo);
+
    /* result = generated - emitted: */
    OUT_PKT7(ring, CP_MEM_TO_MEM, 7);
    OUT_RING(ring, CP_MEM_TO_MEM_0_NEG_B |

diff --git a/src/gallium/drivers/freedreno/freedreno_batch.c b/src/gallium/drivers/freedreno/freedreno_batch.c
index eff58ccb35bfa74365b7ad717bbe2a71d3788bdc..6af2c7af5b04629f022beff90c177650bce9006f 100644 (file)
--- a/src/gallium/drivers/freedreno/freedreno_batch.c
+++ b/src/gallium/drivers/freedreno/freedreno_batch.c
@@ -529,8 +529,6 @@ fd_batch_resource_write(struct fd_batch *batch, struct fd_resource *rsc)
    if (track->write_batch == batch)
       return;
 
-   fd_batch_write_prep(batch, rsc);
-
    if (rsc->stencil)
       fd_batch_resource_write(batch, rsc->stencil);
 
@@ -574,6 +572,8 @@ fd_batch_resource_write(struct fd_batch *batch, struct fd_resource *rsc)
    fd_batch_reference_locked(&track->write_batch, batch);
 
    fd_batch_add_resource(batch, rsc);
+
+   fd_batch_write_prep(batch, rsc);
 }
 
 void

diff --git a/src/gallium/drivers/freedreno/freedreno_query_acc.c b/src/gallium/drivers/freedreno/freedreno_query_acc.c
index e3f4c69f89fc590d5956358748a243b3db441ea3..6dee86d70f30cf5be8d861310fc4ac9e47225563 100644 (file)
--- a/src/gallium/drivers/freedreno/freedreno_query_acc.c
+++ b/src/gallium/drivers/freedreno/freedreno_query_acc.c
@@ -84,13 +84,13 @@ fd_acc_query_resume(struct fd_acc_query *aq, struct fd_batch *batch) assert_dt
 {
    const struct fd_acc_sample_provider *p = aq->provider;
 
-   aq->batch = batch;
-   fd_batch_needs_flush(aq->batch);
-   p->resume(aq, aq->batch);
-
    fd_screen_lock(batch->ctx->screen);
    fd_batch_resource_write(batch, fd_resource(aq->prsc));
    fd_screen_unlock(batch->ctx->screen);
+
+   aq->batch = batch;
+   fd_batch_needs_flush(aq->batch);
+   p->resume(aq, aq->batch);
 }
 
 static void

diff --git a/src/gallium/drivers/freedreno/freedreno_query_acc.h b/src/gallium/drivers/freedreno/freedreno_query_acc.h
index 8022842f9fc94cb8d9989da00e73c08e571b0f59..eebe8c0809502adc91b97e376ce96afa510f1f8b 100644 (file)
--- a/src/gallium/drivers/freedreno/freedreno_query_acc.h
+++ b/src/gallium/drivers/freedreno/freedreno_query_acc.h
@@ -138,6 +138,9 @@ copy_result(struct fd_ringbuffer *ring, enum pipe_query_value_type result_type,
             struct fd_resource *dst, unsigned dst_offset,
             struct fd_resource *src, unsigned src_offset)
 {
+   fd_ringbuffer_attach_bo(ring, dst->bo);
+   fd_ringbuffer_attach_bo(ring, src->bo);
+
    OUT_PKT7(ring, CP_MEM_TO_MEM, 5);
    OUT_RING(ring, COND(result_type >= PIPE_QUERY_TYPE_I64, CP_MEM_TO_MEM_0_DOUBLE));
    OUT_RELOC(ring, dst->bo, dst_offset, 0, 0);