v3d,v3dv: fix compute for V3D 7.1.6+

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25450>

3 files changed, 24 insertions, 6 deletions

diff --git a/src/broadcom/vulkan/v3dv_cmd_buffer.c b/src/broadcom/vulkan/v3dv_cmd_buffer.c
index d22d7a081c6..dc01a0fa06e 100644
--- a/src/broadcom/vulkan/v3dv_cmd_buffer.c
+++ b/src/broadcom/vulkan/v3dv_cmd_buffer.c
@@ -4124,6 +4124,7 @@ cmd_buffer_emit_pre_dispatch(struct v3dv_cmd_buffer *cmd_buffer)
 
 void
 v3dv_cmd_buffer_rewrite_indirect_csd_job(
+   struct v3dv_device *device,
    struct v3dv_csd_indirect_cpu_job_info *info,
    const uint32_t *wg_counts)
 {
@@ -4143,8 +4144,15 @@ v3dv_cmd_buffer_rewrite_indirect_csd_job(
    submit->cfg[1] = wg_counts[1] << V3D_CSD_CFG012_WG_COUNT_SHIFT;
    submit->cfg[2] = wg_counts[2] << V3D_CSD_CFG012_WG_COUNT_SHIFT;
 
-   submit->cfg[4] = DIV_ROUND_UP(info->wg_size, 16) *
-                    (wg_counts[0] * wg_counts[1] * wg_counts[2]) - 1;
+   uint32_t num_batches = DIV_ROUND_UP(info->wg_size, 16) *
+                          (wg_counts[0] * wg_counts[1] * wg_counts[2]);
+   /* V3D 7.1.6 and later don't subtract 1 from the number of batches */
+   if (device->devinfo.ver < 71 ||
+       (device->devinfo.ver == 71 && device->devinfo.rev < 6)) {
+      submit->cfg[4] = num_batches - 1;
+   } else {
+      submit->cfg[4] = num_batches;
+   }
    assert(submit->cfg[4] != ~0);
 
    if (info->needs_wg_uniform_rewrite) {
@@ -4177,6 +4185,7 @@ cmd_buffer_create_csd_job(struct v3dv_cmd_buffer *cmd_buffer,
                           uint32_t **wg_uniform_offsets_out,
                           uint32_t *wg_size_out)
 {
+   struct v3dv_device *device = cmd_buffer->device;
    struct v3dv_pipeline *pipeline = cmd_buffer->state.compute.pipeline;
    assert(pipeline && pipeline->shared_data->variants[BROADCOM_SHADER_COMPUTE]);
    struct v3dv_shader_variant *cs_variant =
@@ -4235,18 +4244,26 @@ cmd_buffer_create_csd_job(struct v3dv_cmd_buffer *cmd_buffer,
    if (wg_size_out)
       *wg_size_out = wg_size;
 
-   submit->cfg[4] = num_batches - 1;
+   /* V3D 7.1.6 and later don't subtract 1 from the number of batches */
+   if (device->devinfo.ver < 71 ||
+       (device->devinfo.ver == 71 && device->devinfo.rev < 6)) {
+      submit->cfg[4] = num_batches - 1;
+   } else {
+      submit->cfg[4] = num_batches;
+   }
    assert(submit->cfg[4] != ~0);
 
    assert(pipeline->shared_data->assembly_bo);
    struct v3dv_bo *cs_assembly_bo = pipeline->shared_data->assembly_bo;
 
    submit->cfg[5] = cs_assembly_bo->offset + cs_variant->assembly_offset;
-   submit->cfg[5] |= V3D_CSD_CFG5_PROPAGATE_NANS;
    if (cs_variant->prog_data.base->single_seg)
       submit->cfg[5] |= V3D_CSD_CFG5_SINGLE_SEG;
    if (cs_variant->prog_data.base->threads == 4)
       submit->cfg[5] |= V3D_CSD_CFG5_THREADING;
+   /* V3D 7.x has made the PROPAGATE_NANS bit in CFG5 reserved  */
+   if (device->devinfo.ver < 71)
+      submit->cfg[5] |= V3D_CSD_CFG5_PROPAGATE_NANS;
 
    if (cs_variant->prog_data.cs->shared_size > 0) {
       job->csd.shared_memory =
diff --git a/src/broadcom/vulkan/v3dv_private.h b/src/broadcom/vulkan/v3dv_private.h
index c46ffbcf9be..01651534b12 100644
--- a/src/broadcom/vulkan/v3dv_private.h
+++ b/src/broadcom/vulkan/v3dv_private.h
@@ -1816,7 +1816,8 @@ void v3dv_cmd_buffer_copy_query_results(struct v3dv_cmd_buffer *cmd_buffer,
 void v3dv_cmd_buffer_add_tfu_job(struct v3dv_cmd_buffer *cmd_buffer,
                                  struct drm_v3d_submit_tfu *tfu);
 
-void v3dv_cmd_buffer_rewrite_indirect_csd_job(struct v3dv_csd_indirect_cpu_job_info *info,
+void v3dv_cmd_buffer_rewrite_indirect_csd_job(struct v3dv_device *device,
+                                              struct v3dv_csd_indirect_cpu_job_info *info,
                                               const uint32_t *wg_counts);
 
 void v3dv_cmd_buffer_add_private_obj(struct v3dv_cmd_buffer *cmd_buffer,
diff --git a/src/broadcom/vulkan/v3dv_queue.c b/src/broadcom/vulkan/v3dv_queue.c
index bb7880a33c1..a0942cf1c96 100644
--- a/src/broadcom/vulkan/v3dv_queue.c
+++ b/src/broadcom/vulkan/v3dv_queue.c
@@ -354,7 +354,7 @@ handle_csd_indirect_cpu_job(struct v3dv_queue *queue,
 
    if (memcmp(group_counts, info->csd_job->csd.wg_count,
               sizeof(info->csd_job->csd.wg_count)) != 0) {
-      v3dv_cmd_buffer_rewrite_indirect_csd_job(info, group_counts);
+      v3dv_cmd_buffer_rewrite_indirect_csd_job(queue->device, info, group_counts);
    }
 
    return VK_SUCCESS;