ac/nir: don't write tcs outputs to LDS that aren't read back.

If the TCS doesn't read back the outputs, no need to store them
to LDS in the first place. (except for tess factors).

This seems to give about 50fps (3290->3330) with tessellation demo.

I haven't tested if it impacts DoW3 at all.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index 1ecdeca063c3fb9883c7bf3cf11c9a430062079d..b58bae90c5dc3bf5ba962f3f66d1c9dce7d60928 100644 (file)
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -149,6 +149,9 @@ struct nir_to_llvm_context {
        unsigned tes_primitive_mode;
        uint64_t tess_outputs_written;
        uint64_t tess_patch_outputs_written;
+
+       uint32_t tcs_patch_outputs_read;
+       uint64_t tcs_outputs_read;
 };
 
 static inline struct nir_to_llvm_context *
@@ -2789,7 +2792,15 @@ store_tcs_output(struct nir_to_llvm_context *ctx,
        const unsigned comp = instr->variables[0]->var->data.location_frac;
        const bool per_vertex = nir_is_per_vertex_io(instr->variables[0]->var, ctx->stage);
        const bool is_compact = instr->variables[0]->var->data.compact;
+       bool store_lds = true;
 
+       if (instr->variables[0]->var->data.patch) {
+               if (!(ctx->tcs_patch_outputs_read & (1U << instr->variables[0]->var->data.location)))
+                       store_lds = false;
+       } else {
+               if (!(ctx->tcs_outputs_read & (1ULL << instr->variables[0]->var->data.location)))
+                       store_lds = false;
+       }
        get_deref_offset(ctx->nir, instr->variables[0],
                         false, NULL, per_vertex ? &vertex_index : NULL,
                         &const_index, &indir_index);
@@ -2826,7 +2837,8 @@ store_tcs_output(struct nir_to_llvm_context *ctx,
                        continue;
                LLVMValueRef value = llvm_extract_elem(&ctx->ac, src, chan - comp);
 
-               ac_lds_store(&ctx->ac, dw_addr, value);
+               if (store_lds || is_tess_factor)
+                       ac_lds_store(&ctx->ac, dw_addr, value);
 
                if (!is_tess_factor && writemask != 0xF)
                        ac_build_buffer_store_dword(&ctx->ac, ctx->hs_ring_tess_offchip, value, 1,
@@ -6550,6 +6562,9 @@ LLVMModuleRef ac_translate_nir_to_llvm(LLVMTargetMachineRef tm,
                        ctx.gs_next_vertex = ac_build_alloca(&ctx.ac, ctx.ac.i32, "gs_next_vertex");
 
                        ctx.gs_max_out_vertices = shaders[i]->info.gs.vertices_out;
+               } else if (shaders[i]->info.stage == MESA_SHADER_TESS_CTRL) {
+                       ctx.tcs_outputs_read = shaders[i]->info.outputs_read;
+                       ctx.tcs_patch_outputs_read = shaders[i]->info.patch_outputs_read;
                } else if (shaders[i]->info.stage == MESA_SHADER_TESS_EVAL) {
                        ctx.tes_primitive_mode = shaders[i]->info.tess.primitive_mode;
                } else if (shaders[i]->info.stage == MESA_SHADER_VERTEX) {