radeonsi: move tessellation shader code into si_shader_llvm_tess.c

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3399>

src/gallium/drivers/radeonsi/Makefile.sources

@@ -38,6 +38,7 @@ C_SOURCES := \
 	si_shader_llvm.c \
 	si_shader_llvm_build.c \
 	si_shader_llvm_ps.c \
+	si_shader_llvm_tess.c \
 	si_shader_nir.c \
 	si_shaderlib_tgsi.c \
 	si_state.c \

src/gallium/drivers/radeonsi/meson.build

@@ -53,6 +53,7 @@ files_libradeonsi = files(
   'si_shader_llvm.c',
   'si_shader_llvm_build.c',
   'si_shader_llvm_ps.c',
+  'si_shader_llvm_tess.c',
   'si_shader_nir.c',
   'si_shaderlib_tgsi.c',
   'si_state.c',

src/gallium/drivers/radeonsi/si_shader_internal.h

@@ -256,6 +256,10 @@ LLVMValueRef si_buffer_load_const(struct si_shader_context *ctx,
 				  LLVMValueRef resource, LLVMValueRef offset);
 void si_llvm_build_ret(struct si_shader_context *ctx, LLVMValueRef ret);
 LLVMValueRef si_prolog_get_rw_buffers(struct si_shader_context *ctx);
+LLVMValueRef si_build_gather_64bit(struct si_shader_context *ctx,
+				   LLVMTypeRef type, LLVMValueRef val1,
+				   LLVMValueRef val2);
+void si_llvm_emit_barrier(struct si_shader_context *ctx);
 void si_declare_compute_memory(struct si_shader_context *ctx);
 LLVMValueRef si_get_primitive_id(struct si_shader_context *ctx,
 				 unsigned swizzle);
@@ -311,6 +315,15 @@ void gfx10_ngg_gs_emit_prologue(struct si_shader_context *ctx);
 void gfx10_ngg_gs_emit_epilogue(struct si_shader_context *ctx);
 void gfx10_ngg_calculate_subgroup_info(struct si_shader *shader);
 
+/* si_shader_llvm_tess.c */
+void si_llvm_preload_tes_rings(struct si_shader_context *ctx);
+void si_llvm_emit_ls_epilogue(struct ac_shader_abi *abi, unsigned max_outputs,
+			      LLVMValueRef *addrs);
+void si_llvm_build_tcs_epilog(struct si_shader_context *ctx,
+			      union si_shader_part_key *key);
+void si_llvm_init_tcs_callbacks(struct si_shader_context *ctx);
+void si_llvm_init_tes_callbacks(struct si_shader_context *ctx);
+
 /* si_shader_llvm_ps.c */
 void si_llvm_build_ps_prolog(struct si_shader_context *ctx,
 			     union si_shader_part_key *key);
@@ -320,4 +333,14 @@ void si_llvm_build_monolithic_ps(struct si_shader_context *ctx,
 				 struct si_shader *shader);
 void si_llvm_init_ps_callbacks(struct si_shader_context *ctx);
 
+/* TODO: remove */
+static inline bool llvm_type_is_64bit(struct si_shader_context *ctx,
+				      LLVMTypeRef type)
+{
+	if (type == ctx->ac.i64 || type == ctx->ac.f64)
+		return true;
+
+	return false;
+}
+
 #endif

src/gallium/drivers/radeonsi/si_shader_llvm_build.c

@@ -215,3 +215,30 @@ LLVMValueRef si_prolog_get_rw_buffers(struct si_shader_context *ctx)
 				 ac_array_in_const32_addr_space(ctx->v4i32), "");
 	return list;
 }
+
+LLVMValueRef si_build_gather_64bit(struct si_shader_context *ctx,
+				   LLVMTypeRef type, LLVMValueRef val1,
+				   LLVMValueRef val2)
+{
+	LLVMValueRef values[2] = {
+		ac_to_integer(&ctx->ac, val1),
+		ac_to_integer(&ctx->ac, val2),
+	};
+	LLVMValueRef result = ac_build_gather_values(&ctx->ac, values, 2);
+	return LLVMBuildBitCast(ctx->ac.builder, result, type, "");
+}
+
+void si_llvm_emit_barrier(struct si_shader_context *ctx)
+{
+	/* GFX6 only (thanks to a hw bug workaround):
+	 * The real barrier instruction isn’t needed, because an entire patch
+	 * always fits into a single wave.
+	 */
+	if (ctx->screen->info.chip_class == GFX6 &&
+	    ctx->type == PIPE_SHADER_TESS_CTRL) {
+		ac_build_waitcnt(&ctx->ac, AC_WAIT_LGKM | AC_WAIT_VLOAD | AC_WAIT_VSTORE);
+		return;
+	}
+
+	ac_build_s_barrier(&ctx->ac);
+}