radv,aco: remove old streamout code
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com> Reviewed-by: Timur Kristóf <timur.kristof@gmail.com> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18898>
This commit is contained in:
@@ -727,126 +727,6 @@ radv_load_output(struct radv_shader_context *ctx, unsigned index, unsigned chan)
|
||||
return LLVMBuildLoad2(ctx->ac.builder, type, output, "");
|
||||
}
|
||||
|
||||
static void
|
||||
radv_emit_stream_output(struct radv_shader_context *ctx, LLVMValueRef const *so_buffers,
|
||||
LLVMValueRef const *so_write_offsets,
|
||||
const struct radv_stream_output *output,
|
||||
struct radv_shader_output_values *shader_out)
|
||||
{
|
||||
unsigned num_comps = util_bitcount(output->component_mask);
|
||||
unsigned buf = output->buffer;
|
||||
unsigned offset = output->offset;
|
||||
unsigned start;
|
||||
LLVMValueRef out[4];
|
||||
|
||||
assert(num_comps && num_comps <= 4);
|
||||
if (!num_comps || num_comps > 4)
|
||||
return;
|
||||
|
||||
/* Get the first component. */
|
||||
start = ffs(output->component_mask) - 1;
|
||||
|
||||
/* Load the output as int. */
|
||||
for (int i = 0; i < num_comps; i++) {
|
||||
out[i] = ac_to_integer(&ctx->ac, shader_out->values[start + i]);
|
||||
}
|
||||
|
||||
/* Pack the output. */
|
||||
LLVMValueRef vdata = NULL;
|
||||
|
||||
switch (num_comps) {
|
||||
case 1: /* as i32 */
|
||||
vdata = out[0];
|
||||
break;
|
||||
case 2: /* as v2i32 */
|
||||
case 3: /* as v3i32 */
|
||||
case 4: /* as v4i32 */
|
||||
vdata = ac_build_gather_values(&ctx->ac, out, num_comps);
|
||||
break;
|
||||
}
|
||||
|
||||
LLVMValueRef voffset = LLVMBuildAdd(ctx->ac.builder, so_write_offsets[buf],
|
||||
LLVMConstInt(ctx->ac.i32, offset, 0), "");
|
||||
ac_build_buffer_store_dword(&ctx->ac, so_buffers[buf], vdata, NULL, voffset, ctx->ac.i32_0,
|
||||
ac_glc | ac_slc);
|
||||
}
|
||||
|
||||
static void
|
||||
radv_emit_streamout(struct radv_shader_context *ctx, unsigned stream)
|
||||
{
|
||||
int i;
|
||||
|
||||
/* Get bits [22:16], i.e. (so_param >> 16) & 127; */
|
||||
assert(ctx->args->ac.streamout_config.used);
|
||||
LLVMValueRef so_vtx_count = ac_build_bfe(
|
||||
&ctx->ac, ac_get_arg(&ctx->ac, ctx->args->ac.streamout_config),
|
||||
LLVMConstInt(ctx->ac.i32, 16, false), LLVMConstInt(ctx->ac.i32, 7, false), false);
|
||||
|
||||
LLVMValueRef tid = ac_get_thread_id(&ctx->ac);
|
||||
|
||||
/* can_emit = tid < so_vtx_count; */
|
||||
LLVMValueRef can_emit = LLVMBuildICmp(ctx->ac.builder, LLVMIntULT, tid, so_vtx_count, "");
|
||||
|
||||
/* Emit the streamout code conditionally. This actually avoids
|
||||
* out-of-bounds buffer access. The hw tells us via the SGPR
|
||||
* (so_vtx_count) which threads are allowed to emit streamout data.
|
||||
*/
|
||||
ac_build_ifcc(&ctx->ac, can_emit, 6501);
|
||||
{
|
||||
/* The buffer offset is computed as follows:
|
||||
* ByteOffset = streamout_offset[buffer_id]*4 +
|
||||
* (streamout_write_index + thread_id)*stride[buffer_id] +
|
||||
* attrib_offset
|
||||
*/
|
||||
LLVMValueRef so_write_index = ac_get_arg(&ctx->ac, ctx->args->ac.streamout_write_index);
|
||||
|
||||
/* Compute (streamout_write_index + thread_id). */
|
||||
so_write_index = LLVMBuildAdd(ctx->ac.builder, so_write_index, tid, "");
|
||||
|
||||
/* Load the descriptor and compute the write offset for each
|
||||
* enabled buffer.
|
||||
*/
|
||||
LLVMValueRef so_write_offset[4] = {0};
|
||||
LLVMValueRef so_buffers[4] = {0};
|
||||
struct ac_llvm_pointer buf_ptr = ac_get_ptr_arg(&ctx->ac, &ctx->args->ac, ctx->args->streamout_buffers);
|
||||
|
||||
for (i = 0; i < 4; i++) {
|
||||
uint16_t stride = ctx->shader_info->so.strides[i];
|
||||
|
||||
if (!stride)
|
||||
continue;
|
||||
|
||||
LLVMValueRef offset = LLVMConstInt(ctx->ac.i32, i, false);
|
||||
|
||||
so_buffers[i] = ac_build_load_to_sgpr(&ctx->ac, buf_ptr, offset);
|
||||
|
||||
LLVMValueRef so_offset = ac_get_arg(&ctx->ac, ctx->args->ac.streamout_offset[i]);
|
||||
|
||||
so_offset =
|
||||
LLVMBuildMul(ctx->ac.builder, so_offset, LLVMConstInt(ctx->ac.i32, 4, false), "");
|
||||
|
||||
so_write_offset[i] = ac_build_imad(
|
||||
&ctx->ac, so_write_index, LLVMConstInt(ctx->ac.i32, stride * 4, false), so_offset);
|
||||
}
|
||||
|
||||
/* Write streamout data. */
|
||||
for (i = 0; i < ctx->shader_info->so.num_outputs; i++) {
|
||||
struct radv_shader_output_values shader_out = {0};
|
||||
const struct radv_stream_output *output = &ctx->shader_info->so.outputs[i];
|
||||
|
||||
if (stream != output->stream)
|
||||
continue;
|
||||
|
||||
for (int j = 0; j < 4; j++) {
|
||||
shader_out.values[j] = radv_load_output(ctx, output->location, j);
|
||||
}
|
||||
|
||||
radv_emit_stream_output(ctx, so_buffers, so_write_offset, output, &shader_out);
|
||||
}
|
||||
}
|
||||
ac_build_endif(&ctx->ac, 6501);
|
||||
}
|
||||
|
||||
static void
|
||||
radv_build_param_exports(struct radv_shader_context *ctx, struct radv_shader_output_values *outputs,
|
||||
unsigned noutput, const struct radv_vs_output_info *outinfo,
|
||||
|
Reference in New Issue
Block a user