[Mesa-dev] [PATCH 54/92] radeonsi/nir: load VS inputs

[Nicolai Hähnle]

From: Nicolai Hähnle <nicolai.haehnle at amd.com>

---
 src/gallium/drivers/radeonsi/si_shader.c          | 11 ++++++++--
 src/gallium/drivers/radeonsi/si_shader_internal.h |  5 +++++
 src/gallium/drivers/radeonsi/si_shader_nir.c      | 26 +++++++++++++++++++++++
 3 files changed, 40 insertions(+), 2 deletions(-)

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index e97e0ee..b7ea131 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -337,24 +337,23 @@ static LLVMValueRef extract_double_to_float(struct si_shader_context *ctx,
 {
 	LLVMBuilderRef builder = ctx->gallivm.builder;
 	LLVMTypeRef f64 = LLVMDoubleTypeInContext(ctx->gallivm.context);
 	LLVMValueRef dvec2 = LLVMBuildBitCast(builder, vec4,
 					      LLVMVectorType(f64, 2), "");
 	LLVMValueRef index = LLVMConstInt(ctx->i32, double_index, 0);
 	LLVMValueRef value = LLVMBuildExtractElement(builder, dvec2, index, "");
 	return LLVMBuildFPTrunc(builder, value, ctx->f32, "");
 }
 
-static void declare_input_vs(
+void si_llvm_load_input_vs(
 	struct si_shader_context *ctx,
 	unsigned input_index,
-	const struct tgsi_full_declaration *decl,
 	LLVMValueRef out[4])
 {
 	struct gallivm_state *gallivm = &ctx->gallivm;
 
 	unsigned chan;
 	unsigned fix_fetch;
 	unsigned num_fetches;
 	unsigned fetch_stride;
 
 	LLVMValueRef t_list_ptr;
@@ -543,20 +542,28 @@ static void declare_input_vs(
 		    fix_fetch == SI_FIX_FETCH_RGB_16) {
 			out[3] = LLVMConstReal(ctx->f32, 1);
 		} else {
 			out[3] = LLVMBuildBitCast(gallivm->builder, ctx->i32_1,
 						  ctx->f32, "");
 		}
 		break;
 	}
 }
 
+static void declare_input_vs(
+	struct si_shader_context *ctx,
+	unsigned input_index,
+	const struct tgsi_full_declaration *decl,
+	LLVMValueRef out[4])
+{
+	si_llvm_load_input_vs(ctx, input_index, out);
+}
 
 static LLVMValueRef get_primitive_id(struct si_shader_context *ctx,
 				     unsigned swizzle)
 {
 	if (swizzle > 0)
 		return ctx->i32_0;
 
 	switch (ctx->type) {
 	case PIPE_SHADER_VERTEX:
 		return LLVMGetParam(ctx->main_fn,
diff --git a/src/gallium/drivers/radeonsi/si_shader_internal.h b/src/gallium/drivers/radeonsi/si_shader_internal.h
index c8dcd5c..99d17c2 100644
--- a/src/gallium/drivers/radeonsi/si_shader_internal.h
+++ b/src/gallium/drivers/radeonsi/si_shader_internal.h
@@ -306,13 +306,18 @@ void si_emit_waitcnt(struct si_shader_context *ctx, unsigned simm16);
 
 LLVMValueRef si_get_bounded_indirect_index(struct si_shader_context *ctx,
 					   const struct tgsi_ind_register *ind,
 					   int rel_index, unsigned num);
 
 LLVMTypeRef si_const_array(LLVMTypeRef elem_type, int num_elements);
 
 void si_shader_context_init_alu(struct lp_build_tgsi_context *bld_base);
 void si_shader_context_init_mem(struct si_shader_context *ctx);
 
+void si_llvm_load_input_vs(
+	struct si_shader_context *ctx,
+	unsigned input_index,
+	LLVMValueRef out[4]);
+
 bool si_nir_build_llvm(struct si_shader_context *ctx, struct nir_shader *nir);
 
 #endif
diff --git a/src/gallium/drivers/radeonsi/si_shader_nir.c b/src/gallium/drivers/radeonsi/si_shader_nir.c
index cc7517b..8b8baac 100644
--- a/src/gallium/drivers/radeonsi/si_shader_nir.c
+++ b/src/gallium/drivers/radeonsi/si_shader_nir.c
@@ -306,16 +306,42 @@ void si_nir_scan_shader(const struct nir_shader *nir,
 	info->const_buffers_declared = u_bit_consecutive(0, SI_NUM_CONST_BUFFERS);
 	info->shader_buffers_declared = u_bit_consecutive(0, SI_NUM_SHADER_BUFFERS);
 
 	func = (struct nir_function *)exec_list_get_head_const(&nir->functions);
 	nir_foreach_block(block, func->impl) {
 		nir_foreach_instr(instr, block)
 			scan_instruction(info, instr);
 	}
 }
 
+static void declare_nir_input_vs(struct si_shader_context *ctx,
+				 struct nir_variable *variable, unsigned rel,
+				 LLVMValueRef out[4])
+{
+	si_llvm_load_input_vs(ctx, variable->data.driver_location / 4 + rel, out);
+}
+
 bool si_nir_build_llvm(struct si_shader_context *ctx, struct nir_shader *nir)
 {
+	nir_foreach_variable(variable, &nir->inputs) {
+		unsigned attrib_count = glsl_count_attribute_slots(variable->type,
+								   nir->stage == MESA_SHADER_VERTEX);
+		unsigned input_idx = variable->data.driver_location;
+
+		for (unsigned i = 0; i < attrib_count; ++i) {
+			LLVMValueRef data[4];
+
+			declare_nir_input_vs(ctx, variable, i, data);
+
+			for (unsigned chan = 0; chan < 4; chan++) {
+				ctx->inputs[input_idx + chan] =
+					LLVMBuildBitCast(ctx->ac.builder, data[chan], ctx->ac.i32, "");
+			}
+		}
+	}
+
+	ctx->abi.inputs = &ctx->inputs[0];
+
 	ac_nir_translate(&ctx->ac, &ctx->abi, nir, NULL);
 
 	return true;
 }
-- 
2.9.3