radeonsi: set the calling convention for inlined function calls

otherwise the behavior is undefined

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

src/amd/common/ac_llvm_build.c

@@ -4445,3 +4445,11 @@ ac_build_load_helper_invocation(struct ac_llvm_context *ctx)
 	result = LLVMBuildNot(ctx->builder, result, "");
 	return LLVMBuildSExt(ctx->builder, result, ctx->i32, "");
 }
+
+LLVMValueRef ac_build_call(struct ac_llvm_context *ctx, LLVMValueRef func,
+			   LLVMValueRef *args, unsigned num_args)
+{
+	LLVMValueRef ret = LLVMBuildCall(ctx->builder, func, args, num_args, "");
+	LLVMSetInstructionCallConv(ret, LLVMGetFunctionCallConv(func));
+	return ret;
+}

src/amd/common/ac_llvm_build.h

@@ -720,6 +720,9 @@ ac_build_ddxy_interp(struct ac_llvm_context *ctx, LLVMValueRef interp_ij);
 LLVMValueRef
 ac_build_load_helper_invocation(struct ac_llvm_context *ctx);
 
+LLVMValueRef ac_build_call(struct ac_llvm_context *ctx, LLVMValueRef func,
+			   LLVMValueRef *args, unsigned num_args);
+
 LLVMValueRef ac_build_atomic_rmw(struct ac_llvm_context *ctx, LLVMAtomicRMWBinOp op,
 				 LLVMValueRef ptr, LLVMValueRef val,
 				 const char *sync_scope);

src/gallium/drivers/radeonsi/si_compute_prim_discard.c

@@ -667,7 +667,7 @@ void si_build_prim_discard_compute_shader(struct si_shader_context *ctx)
 		vs_params[param_vertex_id] = index[i];
 		vs_params[param_instance_id] = instance_id;
 
-		LLVMValueRef ret = LLVMBuildCall(builder, vs, vs_params, num_vs_params, "");
+		LLVMValueRef ret = ac_build_call(&ctx->ac, vs, vs_params, num_vs_params);
 		for (unsigned chan = 0; chan < 4; chan++)
 			pos[i][chan] = LLVMBuildExtractValue(builder, ret, chan, "");
 	}

src/gallium/drivers/radeonsi/si_shader.c

@@ -6591,7 +6591,7 @@ static void si_build_wrapper_function(struct si_shader_context *ctx,
 			out_idx += param_size;
 		}
 
-		ret = LLVMBuildCall(builder, parts[part], in, num_params, "");
+		ret = ac_build_call(&ctx->ac, parts[part], in, num_params);
 
 		if (is_merged_shader(ctx) &&
 		    part + 1 == next_shader_first_part) {