fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-24 02:20:11 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

radv: add multiple streams support for the GS copy shader

Browse source 
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
This commit is contained in:
Samuel Pitoiset 2018-10-05 17:54:49 +02:00
parent 6c21645046
commit f8d0337299
1 changed files with 77 additions and 27 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

104
src/amd/vulkan/radv_nir_to_llvm.c
View file

@ -2723,8 +2723,11 @@ handle_vs_outputs_post(struct radv_shader_context *ctx,
viewport_index_value = radv_load_output(ctx, VARYING_SLOT_VIEWPORT, 0); viewport_index_value = radv_load_output(ctx, VARYING_SLOT_VIEWPORT, 0);
} }
if (ctx->shader_info->info.so.num_outputs) if (ctx->shader_info->info.so.num_outputs &&
!ctx->is_gs_copy_shader) {
/* The GS copy shader emission already emits streamout. */
radv_emit_streamout(ctx, 0); radv_emit_streamout(ctx, 0);
}
if (outinfo->writes_pointsize || if (outinfo->writes_pointsize ||
outinfo->writes_layer || outinfo->writes_layer ||
@ -3829,45 +3832,92 @@ ac_gs_copy_shader_emit(struct radv_shader_context *ctx)
LLVMValueRef vtx_offset = LLVMValueRef vtx_offset =
LLVMBuildMul(ctx->ac.builder, ctx->abi.vertex_id, LLVMBuildMul(ctx->ac.builder, ctx->abi.vertex_id,
LLVMConstInt(ctx->ac.i32, 4, false), ""); LLVMConstInt(ctx->ac.i32, 4, false), "");
unsigned offset = 0; LLVMValueRef stream_id;
for (unsigned i = 0; i < AC_LLVM_MAX_OUTPUTS; ++i) { /* Fetch the vertex stream ID. */
unsigned output_usage_mask = if (ctx->shader_info->info.so.num_outputs) {
ctx->shader_info->info.gs.output_usage_mask[i]; stream_id =
int length = util_last_bit(output_usage_mask); ac_unpack_param(&ctx->ac, ctx->streamout_config, 24, 2);
} else {
stream_id = ctx->ac.i32_0;
}
if (!(ctx->output_mask & (1ull << i))) LLVMBasicBlockRef end_bb;
LLVMValueRef switch_inst;
end_bb = LLVMAppendBasicBlockInContext(ctx->ac.context,
ctx->main_function, "end");
switch_inst = LLVMBuildSwitch(ctx->ac.builder, stream_id, end_bb, 4);
for (unsigned stream = 0; stream < 4; stream++) {
unsigned num_components =
ctx->shader_info->info.gs.num_stream_output_components[stream];
LLVMBasicBlockRef bb;
unsigned offset;
if (!num_components)
continue; continue;
for (unsigned j = 0; j < length; j++) { if (stream > 0 && !ctx->shader_info->info.so.num_outputs)
LLVMValueRef value, soffset; continue;
if (!(output_usage_mask & (1 << j))) bb = LLVMInsertBasicBlockInContext(ctx->ac.context, end_bb, "out");
LLVMAddCase(switch_inst, LLVMConstInt(ctx->ac.i32, stream, 0), bb);
LLVMPositionBuilderAtEnd(ctx->ac.builder, bb);
offset = 0;
for (unsigned i = 0; i < AC_LLVM_MAX_OUTPUTS; ++i) {
unsigned output_usage_mask =
ctx->shader_info->info.gs.output_usage_mask[i];
unsigned output_stream =
ctx->shader_info->info.gs.output_streams[i];
int length = util_last_bit(output_usage_mask);
if (!(ctx->output_mask & (1ull << i)) ||
output_stream != stream)
continue; continue;
soffset = LLVMConstInt(ctx->ac.i32, for (unsigned j = 0; j < length; j++) {
offset * LLVMValueRef value, soffset;
ctx->gs_max_out_vertices * 16 * 4, false);
offset++; if (!(output_usage_mask & (1 << j)))
continue;
value = ac_build_buffer_load(&ctx->ac, soffset = LLVMConstInt(ctx->ac.i32,
ctx->gsvs_ring[0], offset *
1, ctx->ac.i32_0, ctx->gs_max_out_vertices * 16 * 4, false);
vtx_offset, soffset,
0, 1, 1, true, false);
LLVMTypeRef type = LLVMGetAllocatedType(ctx->abi.outputs[ac_llvm_reg_index_soa(i, j)]); offset++;
if (ac_get_type_size(type) == 2) {
value = LLVMBuildBitCast(ctx->ac.builder, value, ctx->ac.i32, ""); value = ac_build_buffer_load(&ctx->ac,
value = LLVMBuildTrunc(ctx->ac.builder, value, ctx->ac.i16, ""); ctx->gsvs_ring[0],
1, ctx->ac.i32_0,
vtx_offset, soffset,
0, 1, 1, true, false);
LLVMTypeRef type = LLVMGetAllocatedType(ctx->abi.outputs[ac_llvm_reg_index_soa(i, j)]);
if (ac_get_type_size(type) == 2) {
value = LLVMBuildBitCast(ctx->ac.builder, value, ctx->ac.i32, "");
value = LLVMBuildTrunc(ctx->ac.builder, value, ctx->ac.i16, "");
}
LLVMBuildStore(ctx->ac.builder,
ac_to_float(&ctx->ac, value), ctx->abi.outputs[ac_llvm_reg_index_soa(i, j)]);
} }
LLVMBuildStore(ctx->ac.builder,
ac_to_float(&ctx->ac, value), ctx->abi.outputs[ac_llvm_reg_index_soa(i, j)]);
} }
if (ctx->shader_info->info.so.num_outputs)
radv_emit_streamout(ctx, stream);
if (stream == 0) {
handle_vs_outputs_post(ctx, false, false,
&ctx->shader_info->vs.outinfo);
}
LLVMBuildBr(ctx->ac.builder, end_bb);
} }
handle_vs_outputs_post(ctx, false, false, &ctx->shader_info->vs.outinfo);
LLVMPositionBuilderAtEnd(ctx->ac.builder, end_bb);
} }
void void