summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorMarek Olšák <marek.olsak@amd.com>2017-09-05 17:46:09 +0200
committerMarek Olšák <marek.olsak@amd.com>2017-09-07 13:00:07 +0200
commitf974bb768b8a9def7effea561fad6972376fe35c (patch)
tree90f2923284feba59bc66b321e133aa585f8667a9
parent22f5dfd300f791ecd2c79565731a72074bc7562f (diff)
radeonsi: don't read the LS output vertex stride from an SGPR in LS
Now it's able to generate ds_write2_b64 instead of ds_write2_b32. -20 bytes in one shader binary. (having only 1 output) Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de> Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
-rw-r--r--src/gallium/drivers/radeonsi/si_shader.c25
1 files changed, 21 insertions, 4 deletions
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 61ee040ec7..362ee038e1 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -398,6 +398,24 @@ static LLVMValueRef get_num_tcs_out_vertices(struct si_shader_context *ctx)
return unpack_param(ctx, ctx->param_tcs_offchip_layout, 6, 6);
}
+static LLVMValueRef get_tcs_in_vertex_dw_stride(struct si_shader_context *ctx)
+{
+ unsigned stride;
+
+ switch (ctx->type) {
+ case PIPE_SHADER_VERTEX:
+ stride = util_last_bit64(ctx->shader->selector->outputs_written);
+ return LLVMConstInt(ctx->i32, stride * 4, 0);
+
+ case PIPE_SHADER_TESS_CTRL:
+ return unpack_param(ctx, ctx->param_vs_state_bits, 24, 8);
+
+ default:
+ assert(0);
+ return NULL;
+ }
+}
+
static LLVMValueRef get_instance_index_for_fetch(
struct si_shader_context *ctx,
unsigned param_start_instance, LLVMValueRef divisor)
@@ -1047,7 +1065,7 @@ static LLVMValueRef fetch_input_tcs(
struct si_shader_context *ctx = si_shader_context(bld_base);
LLVMValueRef dw_addr, stride;
- stride = unpack_param(ctx, ctx->param_vs_state_bits, 24, 8);
+ stride = get_tcs_in_vertex_dw_stride(ctx);
dw_addr = get_tcs_in_current_patch_offset(ctx);
dw_addr = get_dw_address(ctx, NULL, reg, stride, dw_addr);
@@ -2595,7 +2613,7 @@ static void si_copy_tcs_inputs(struct lp_build_tgsi_context *bld_base)
buffer = desc_from_addr_base64k(ctx, ctx->param_tcs_offchip_addr_base64k);
buffer_offset = LLVMGetParam(ctx->main_fn, ctx->param_tcs_offchip_offset);
- lds_vertex_stride = unpack_param(ctx, ctx->param_vs_state_bits, 24, 8);
+ lds_vertex_stride = get_tcs_in_vertex_dw_stride(ctx);
lds_vertex_offset = LLVMBuildMul(gallivm->builder, invocation_id,
lds_vertex_stride, "");
lds_base = get_tcs_in_current_patch_offset(ctx);
@@ -2982,8 +3000,7 @@ static void si_llvm_emit_ls_epilogue(struct lp_build_tgsi_context *bld_base)
unsigned i, chan;
LLVMValueRef vertex_id = LLVMGetParam(ctx->main_fn,
ctx->param_rel_auto_id);
- LLVMValueRef vertex_dw_stride =
- unpack_param(ctx, ctx->param_vs_state_bits, 24, 8);
+ LLVMValueRef vertex_dw_stride = get_tcs_in_vertex_dw_stride(ctx);
LLVMValueRef base_dw_addr = LLVMBuildMul(gallivm->builder, vertex_id,
vertex_dw_stride, "");