KonstantinSeurer
/
mesa
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

freedreno/ir3: Move max-tf-vtx calculation to a .c file 

This much code should probably be a function.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9687>
This commit is contained in:
Eric Anholt 2021-03-17 16:43:52 -07:00 committed by Marge Bot
parent c29c7f5042
commit 4af6fbb965
3 changed files with 50 additions and 47 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

48
src/gallium/drivers/freedreno/ir3/ir3_const.h
View File

@ -409,52 +409,6 @@ emit_tfbos(struct fd_context *ctx, const struct ir3_shader_variant *v,
}
}
static inline uint32_t
max_tf_vtx(struct fd_context *ctx, const struct ir3_shader_variant *v)
{
struct fd_streamout_stateobj *so = &ctx->streamout;
struct ir3_stream_output_info *info = &v->shader->stream_output;
uint32_t maxvtxcnt = 0x7fffffff;
if (ctx->screen->gpu_id >= 500)
return 0;
if (v->binning_pass)
return 0;
if (v->shader->stream_output.num_outputs == 0)
return 0;
if (so->num_targets == 0)
return 0;
/* offset to write to is:
*
* total_vtxcnt = vtxcnt + offsets[i]
* offset = total_vtxcnt * stride[i]
*
* offset = vtxcnt * stride[i] ; calculated in shader
* + offsets[i] * stride[i] ; calculated at emit_tfbos()
*
* assuming for each vtx, each target buffer will have data written
* up to 'offset + stride[i]', that leaves maxvtxcnt as:
*
* buffer_size = (maxvtxcnt * stride[i]) + stride[i]
* maxvtxcnt = (buffer_size - stride[i]) / stride[i]
*
* but shader is actually doing a less-than (rather than less-than-
* equal) check, so we can drop the -stride[i].
*
* TODO is assumption about `offset + stride[i]` legit?
*/
for (unsigned i = 0; i < so->num_targets; i++) {
struct pipe_stream_output_target *target = so->targets[i];
unsigned stride = info->stride[i] * 4; /* convert dwords->bytes */
if (target) {
uint32_t max = target->buffer_size / stride;
maxvtxcnt = MIN2(maxvtxcnt, max);
}
}
return maxvtxcnt;
}
static inline void
emit_common_consts(const struct ir3_shader_variant *v, struct fd_ringbuffer *ring,
@ -520,7 +474,7 @@ ir3_emit_vs_driver_params(const struct ir3_shader_variant *v,
[IR3_DP_VTXID_BASE] = info->index_size ?
info->index_bias : draw->start,
[IR3_DP_INSTID_BASE] = info->start_instance,
[IR3_DP_VTXCNT_MAX] = max_tf_vtx(ctx, v),
[IR3_DP_VTXCNT_MAX] = ir3_max_tf_vtx(ctx, v),
};
if (v->key.ucp_enables) {
struct pipe_clip_state *ucp = &ctx->ucp;

47
src/gallium/drivers/freedreno/ir3/ir3_gallium.c
View File

@ -556,3 +556,50 @@ ir3_screen_fini(struct pipe_screen *pscreen)
ir3_compiler_destroy(screen->compiler);
screen->compiler = NULL;
}
uint32_t
ir3_max_tf_vtx(struct fd_context *ctx, const struct ir3_shader_variant *v)
{
struct fd_streamout_stateobj *so = &ctx->streamout;
struct ir3_stream_output_info *info = &v->shader->stream_output;
uint32_t maxvtxcnt = 0x7fffffff;
if (ctx->screen->gpu_id >= 500)
return 0;
if (v->binning_pass)
return 0;
if (v->shader->stream_output.num_outputs == 0)
return 0;
if (so->num_targets == 0)
return 0;
/* offset to write to is:
*
* total_vtxcnt = vtxcnt + offsets[i]
* offset = total_vtxcnt * stride[i]
*
* offset = vtxcnt * stride[i] ; calculated in shader
* + offsets[i] * stride[i] ; calculated at emit_tfbos()
*
* assuming for each vtx, each target buffer will have data written
* up to 'offset + stride[i]', that leaves maxvtxcnt as:
*
* buffer_size = (maxvtxcnt * stride[i]) + stride[i]
* maxvtxcnt = (buffer_size - stride[i]) / stride[i]
*
* but shader is actually doing a less-than (rather than less-than-
* equal) check, so we can drop the -stride[i].
*
* TODO is assumption about `offset + stride[i]` legit?
*/
for (unsigned i = 0; i < so->num_targets; i++) {
struct pipe_stream_output_target *target = so->targets[i];
unsigned stride = info->stride[i] * 4; /* convert dwords->bytes */
if (target) {
uint32_t max = target->buffer_size / stride;
maxvtxcnt = MIN2(maxvtxcnt, max);
}
}
return maxvtxcnt;
}

2
src/gallium/drivers/freedreno/ir3/ir3_gallium.h
View File

@ -77,4 +77,6 @@ ir3_point_sprite(const struct ir3_shader_variant *fs, int i,
}
}
uint32_t ir3_max_tf_vtx(struct fd_context *ctx, const struct ir3_shader_variant *v) assert_dt;
#endif /* IR3_GALLIUM_H_ */