[Mesa-dev] [PATCH 53/92] ac/nir, radeonsi: add ac_shader_abi::load_ubo

[Nicolai Hähnle]

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

---
 src/amd/common/ac_nir_to_llvm.c          |  3 +++
 src/amd/common/ac_shader_abi.h           |  2 ++
 src/gallium/drivers/radeonsi/si_shader.c | 14 ++++++++++++++
 3 files changed, 19 insertions(+)

diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index d4b52b4..9e7f51a 100644
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -2392,20 +2392,23 @@ static LLVMValueRef visit_load_buffer(struct nir_to_llvm_context *ctx,
 }
 
 static LLVMValueRef visit_load_ubo_buffer(struct ac_nir_context *ctx,
                                           const nir_intrinsic_instr *instr)
 {
 	LLVMValueRef results[8], ret;
 	LLVMValueRef rsrc = get_src(ctx, instr->src[0]);
 	LLVMValueRef offset = get_src(ctx, instr->src[1]);
 	int num_components = instr->num_components;
 
+	if (ctx->abi->load_ubo)
+		rsrc = ctx->abi->load_ubo(ctx->abi, rsrc);
+
 	rsrc = LLVMBuildBitCast(ctx->ac.builder, rsrc, LLVMVectorType(ctx->ac.i8, 16), "");
 
 	if (instr->dest.ssa.bit_size == 64)
 		num_components *= 2;
 
 	for (unsigned i = 0; i < num_components; ++i) {
 		LLVMValueRef params[] = {
 			rsrc,
 			LLVMBuildAdd(ctx->ac.builder, LLVMConstInt(ctx->ac.i32, 4 * i, 0),
 				     offset, "")
diff --git a/src/amd/common/ac_shader_abi.h b/src/amd/common/ac_shader_abi.h
index 0d91dd2..30dc5d4 100644
--- a/src/amd/common/ac_shader_abi.h
+++ b/src/amd/common/ac_shader_abi.h
@@ -37,13 +37,15 @@ struct ac_shader_abi {
 	/* For VS and PS: pre-loaded shader inputs.
 	 *
 	 * Currently only used for NIR shaders; indexed by variables'
 	 * driver_location.
 	 */
 	LLVMValueRef *inputs;
 
 	void (*emit_outputs)(struct ac_shader_abi *abi,
 			     unsigned max_outputs,
 			     LLVMValueRef *addrs);
+
+	LLVMValueRef (*load_ubo)(struct ac_shader_abi *abi, LLVMValueRef index);
 };
 
 #endif /* AC_SHADER_ABI_H */
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index c3ea666..e97e0ee 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -1720,20 +1720,32 @@ static void declare_compute_memory(struct si_shader_context *ctx,
 
 static LLVMValueRef load_const_buffer_desc(struct si_shader_context *ctx, int i)
 {
 	LLVMValueRef list_ptr = LLVMGetParam(ctx->main_fn,
 					     ctx->param_const_and_shader_buffers);
 
 	return ac_build_indexed_load_const(&ctx->ac, list_ptr,
 			LLVMConstInt(ctx->i32, si_get_constbuf_slot(i), 0));
 }
 
+static LLVMValueRef load_ubo(struct ac_shader_abi *abi, LLVMValueRef index)
+{
+	struct si_shader_context *ctx = si_shader_context_from_abi(abi);
+	LLVMValueRef ptr = LLVMGetParam(ctx->main_fn, ctx->param_const_and_shader_buffers);
+
+	index = si_llvm_bound_index(ctx, index, ctx->num_const_buffers);
+	index = LLVMBuildAdd(ctx->gallivm.builder, index,
+			     LLVMConstInt(ctx->i32, SI_NUM_SHADER_BUFFERS, 0), "");
+
+	return ac_build_indexed_load_const(&ctx->ac, ptr, index);
+}
+
 static LLVMValueRef fetch_constant(
 	struct lp_build_tgsi_context *bld_base,
 	const struct tgsi_full_src_register *reg,
 	enum tgsi_opcode_type type,
 	unsigned swizzle)
 {
 	struct si_shader_context *ctx = si_shader_context(bld_base);
 	struct lp_build_context *base = &bld_base->base;
 	const struct tgsi_ind_register *ireg = &reg->Indirect;
 	unsigned buf, idx;
@@ -5539,20 +5551,22 @@ static bool si_compile_tgsi_main(struct si_shader_context *ctx,
 		bld_base->emit_epilogue = si_llvm_return_fs_outputs;
 		break;
 	case PIPE_SHADER_COMPUTE:
 		ctx->declare_memory_region = declare_compute_memory;
 		break;
 	default:
 		assert(!"Unsupported shader type");
 		return false;
 	}
 
+	ctx->abi.load_ubo = load_ubo;
+
 	create_function(ctx);
 	preload_ring_buffers(ctx);
 
 	/* For GFX9 merged shaders:
 	 * - Set EXEC. If the prolog is present, set EXEC there instead.
 	 * - Add a barrier before the second shader.
 	 *
 	 * The same thing for monolithic shaders is done in
 	 * si_build_wrapper_function.
 	 */
-- 
2.9.3