ac: clean up ac_build_indexed_load function interfaces

Reviewed-by: Nicolai Hähnle <[email protected]>

5 files changed, 61 insertions, 55 deletions

diff --git a/src/amd/common/ac_llvm_build.c b/src/amd/common/ac_llvm_build.c
index 1d97b09cd4f..949f181aace 100644
--- a/src/amd/common/ac_llvm_build.c
+++ b/src/amd/common/ac_llvm_build.c
@@ -717,32 +717,40 @@ ac_build_indexed_store(struct ac_llvm_context *ctx,
  * \param base_ptr  Where the array starts.
  * \param index     The element index into the array.
  * \param uniform   Whether the base_ptr and index can be assumed to be
- *                  dynamically uniform
+ *                  dynamically uniform (i.e. load to an SGPR)
+ * \param invariant Whether the load is invariant (no other opcodes affect it)
  */
-LLVMValueRef
-ac_build_indexed_load(struct ac_llvm_context *ctx,
-		      LLVMValueRef base_ptr, LLVMValueRef index,
-		      bool uniform)
+static LLVMValueRef
+ac_build_load_custom(struct ac_llvm_context *ctx, LLVMValueRef base_ptr,
+		     LLVMValueRef index, bool uniform, bool invariant)
 {
-	LLVMValueRef pointer;
+	LLVMValueRef pointer, result;
 
 	pointer = ac_build_gep0(ctx, base_ptr, index);
 	if (uniform)
 		LLVMSetMetadata(pointer, ctx->uniform_md_kind, ctx->empty_md);
-	return LLVMBuildLoad(ctx->builder, pointer, "");
+	result = LLVMBuildLoad(ctx->builder, pointer, "");
+	if (invariant)
+		LLVMSetMetadata(result, ctx->invariant_load_md_kind, ctx->empty_md);
+	return result;
 }
 
-/**
- * Do a load from &base_ptr[index], but also add a flag that it's loading
- * a constant from a dynamically uniform index.
- */
-LLVMValueRef
-ac_build_indexed_load_const(struct ac_llvm_context *ctx,
-			    LLVMValueRef base_ptr, LLVMValueRef index)
+LLVMValueRef ac_build_load(struct ac_llvm_context *ctx, LLVMValueRef base_ptr,
+			   LLVMValueRef index)
 {
-	LLVMValueRef result = ac_build_indexed_load(ctx, base_ptr, index, true);
-	LLVMSetMetadata(result, ctx->invariant_load_md_kind, ctx->empty_md);
-	return result;
+	return ac_build_load_custom(ctx, base_ptr, index, false, false);
+}
+
+LLVMValueRef ac_build_load_invariant(struct ac_llvm_context *ctx,
+				     LLVMValueRef base_ptr, LLVMValueRef index)
+{
+	return ac_build_load_custom(ctx, base_ptr, index, false, true);
+}
+
+LLVMValueRef ac_build_load_to_sgpr(struct ac_llvm_context *ctx,
+				   LLVMValueRef base_ptr, LLVMValueRef index)
+{
+	return ac_build_load_custom(ctx, base_ptr, index, true, true);
 }
 
 /* TBUFFER_STORE_FORMAT_{X,XY,XYZ,XYZW} <- the suffix is selected by num_channels=1..4.
diff --git a/src/amd/common/ac_llvm_build.h b/src/amd/common/ac_llvm_build.h
index ac8ea9c4152..f0b5875b423 100644
--- a/src/amd/common/ac_llvm_build.h
+++ b/src/amd/common/ac_llvm_build.h
@@ -150,14 +150,12 @@ ac_build_indexed_store(struct ac_llvm_context *ctx,
 		       LLVMValueRef base_ptr, LLVMValueRef index,
 		       LLVMValueRef value);
 
-LLVMValueRef
-ac_build_indexed_load(struct ac_llvm_context *ctx,
-		      LLVMValueRef base_ptr, LLVMValueRef index,
-		      bool uniform);
-
-LLVMValueRef
-ac_build_indexed_load_const(struct ac_llvm_context *ctx,
-			    LLVMValueRef base_ptr, LLVMValueRef index);
+LLVMValueRef ac_build_load(struct ac_llvm_context *ctx, LLVMValueRef base_ptr,
+			   LLVMValueRef index);
+LLVMValueRef ac_build_load_invariant(struct ac_llvm_context *ctx,
+				     LLVMValueRef base_ptr, LLVMValueRef index);
+LLVMValueRef ac_build_load_to_sgpr(struct ac_llvm_context *ctx,
+				   LLVMValueRef base_ptr, LLVMValueRef index);
 
 void
 ac_build_buffer_store_dword(struct ac_llvm_context *ctx,
diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index 11ba487f375..3ba3ebf051e 100644
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -798,7 +798,7 @@ static void create_function(struct nir_to_llvm_context *ctx)
 		for (unsigned i = 0; i < num_sets; ++i) {
 			if (ctx->options->layout->set[i].layout->shader_stages & (1 << ctx->stage)) {
 				set_userdata_location_indirect(&ctx->shader_info->user_sgprs_locs.descriptor_sets[i], desc_sgpr_idx, 2, i * 8);
-				ctx->descriptor_sets[i] = ac_build_indexed_load_const(&ctx->ac, desc_sets, LLVMConstInt(ctx->i32, i, false));
+				ctx->descriptor_sets[i] = ac_build_load_to_sgpr(&ctx->ac, desc_sets, LLVMConstInt(ctx->i32, i, false));
 
 			} else
 				ctx->descriptor_sets[i] = NULL;
@@ -2524,7 +2524,7 @@ lds_load(struct nir_to_llvm_context *ctx,
 	 LLVMValueRef dw_addr)
 {
 	LLVMValueRef value;
-	value = ac_build_indexed_load(&ctx->ac, ctx->lds, dw_addr, false);
+	value = ac_build_load(&ctx->ac, ctx->lds, dw_addr);
 	return value;
 }
 
@@ -3739,7 +3739,7 @@ static LLVMValueRef load_sample_position(struct nir_to_llvm_context *ctx,
 			       const_array(ctx->v2f32, 64), "");
 
 	sample_id = LLVMBuildAdd(ctx->builder, sample_id, ctx->sample_pos_offset, "");
-	result = ac_build_indexed_load(&ctx->ac, ptr, sample_id, false);
+	result = ac_build_load_invariant(&ctx->ac, ptr, sample_id);
 
 	return result;
 }
@@ -4230,7 +4230,7 @@ static LLVMValueRef radv_get_sampler_desc(struct ac_shader_abi *abi,
 	list = ac_build_gep0(&ctx->ac, list, LLVMConstInt(ctx->i32, offset, 0));
 	list = LLVMBuildPointerCast(builder, list, const_array(type, 0), "");
 
-	return ac_build_indexed_load_const(&ctx->ac, list, index);
+	return ac_build_load_to_sgpr(&ctx->ac, list, index);
 }
 
 static LLVMValueRef get_sampler_desc(struct ac_nir_context *ctx,
@@ -4927,7 +4927,7 @@ handle_vs_input_decl(struct nir_to_llvm_context *ctx,
 	for (unsigned i = 0; i < attrib_count; ++i, ++idx) {
 		t_offset = LLVMConstInt(ctx->i32, index + i, false);
 
-		t_list = ac_build_indexed_load_const(&ctx->ac, t_list_ptr, t_offset);
+		t_list = ac_build_load_to_sgpr(&ctx->ac, t_list_ptr, t_offset);
 
 		input = ac_build_buffer_load_format(&ctx->ac, t_list,
 						    buffer_index,
@@ -6217,16 +6217,16 @@ ac_setup_rings(struct nir_to_llvm_context *ctx)
 {
 	if ((ctx->stage == MESA_SHADER_VERTEX && ctx->options->key.vs.as_es) ||
 	    (ctx->stage == MESA_SHADER_TESS_EVAL && ctx->options->key.tes.as_es)) {
-		ctx->esgs_ring = ac_build_indexed_load_const(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->i32, RING_ESGS_VS, false));
+		ctx->esgs_ring = ac_build_load_to_sgpr(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->i32, RING_ESGS_VS, false));
 	}
 
 	if (ctx->is_gs_copy_shader) {
-		ctx->gsvs_ring = ac_build_indexed_load_const(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->i32, RING_GSVS_VS, false));
+		ctx->gsvs_ring = ac_build_load_to_sgpr(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->i32, RING_GSVS_VS, false));
 	}
 	if (ctx->stage == MESA_SHADER_GEOMETRY) {
 		LLVMValueRef tmp;
-		ctx->esgs_ring = ac_build_indexed_load_const(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->i32, RING_ESGS_GS, false));
-		ctx->gsvs_ring = ac_build_indexed_load_const(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->i32, RING_GSVS_GS, false));
+		ctx->esgs_ring = ac_build_load_to_sgpr(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->i32, RING_ESGS_GS, false));
+		ctx->gsvs_ring = ac_build_load_to_sgpr(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->i32, RING_GSVS_GS, false));
 
 		ctx->gsvs_ring = LLVMBuildBitCast(ctx->builder, ctx->gsvs_ring, ctx->v4i32, "");
 
@@ -6238,8 +6238,8 @@ ac_setup_rings(struct nir_to_llvm_context *ctx)
 
 	if (ctx->stage == MESA_SHADER_TESS_CTRL ||
 	    ctx->stage == MESA_SHADER_TESS_EVAL) {
-		ctx->hs_ring_tess_offchip = ac_build_indexed_load_const(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->i32, RING_HS_TESS_OFFCHIP, false));
-		ctx->hs_ring_tess_factor = ac_build_indexed_load_const(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->i32, RING_HS_TESS_FACTOR, false));
+		ctx->hs_ring_tess_offchip = ac_build_load_to_sgpr(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->i32, RING_HS_TESS_OFFCHIP, false));
+		ctx->hs_ring_tess_factor = ac_build_load_to_sgpr(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->i32, RING_HS_TESS_FACTOR, false));
 	}
 }
 
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index ff372ae0531..f72e5af31fd 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -561,7 +561,7 @@ void si_llvm_load_input_vs(
 
 	t_offset = LLVMConstInt(ctx->i32, input_index, 0);
 
-	t_list = ac_build_indexed_load_const(&ctx->ac, t_list_ptr, t_offset);
+	t_list = ac_build_load_to_sgpr(&ctx->ac, t_list_ptr, t_offset);
 
 	vertex_index = LLVMGetParam(ctx->main_fn,
 				    ctx->param_vertex_index0 +
@@ -1099,12 +1099,12 @@ static LLVMValueRef lds_load(struct lp_build_tgsi_context *bld_base,
 	dw_addr = lp_build_add(&bld_base->uint_bld, dw_addr,
 			    LLVMConstInt(ctx->i32, swizzle, 0));
 
-	value = ac_build_indexed_load(&ctx->ac, ctx->lds, dw_addr, false);
+	value = ac_build_load(&ctx->ac, ctx->lds, dw_addr);
 	if (tgsi_type_is_64bit(type)) {
 		LLVMValueRef value2;
 		dw_addr = lp_build_add(&bld_base->uint_bld, dw_addr,
 				       ctx->i32_1);
-		value2 = ac_build_indexed_load(&ctx->ac, ctx->lds, dw_addr, false);
+		value2 = ac_build_load(&ctx->ac, ctx->lds, dw_addr);
 		return si_llvm_emit_fetch_64bit(bld_base, type, value, value2);
 	}
 
@@ -1617,7 +1617,7 @@ static LLVMValueRef load_sample_position(struct si_shader_context *ctx, LLVMValu
 	struct lp_build_context *uint_bld = &ctx->bld_base.uint_bld;
 	LLVMValueRef desc = LLVMGetParam(ctx->main_fn, ctx->param_rw_buffers);
 	LLVMValueRef buf_index = LLVMConstInt(ctx->i32, SI_PS_CONST_SAMPLE_POSITIONS, 0);
-	LLVMValueRef resource = ac_build_indexed_load_const(&ctx->ac, desc, buf_index);
+	LLVMValueRef resource = ac_build_load_to_sgpr(&ctx->ac, desc, buf_index);
 
 	/* offset = sample_id * 8  (8 = 2 floats containing samplepos.xy) */
 	LLVMValueRef offset0 = lp_build_mul_imm(uint_bld, sample_id, 8);
@@ -1792,7 +1792,7 @@ void si_load_system_value(struct si_shader_context *ctx,
 
 		slot = LLVMConstInt(ctx->i32, SI_HS_CONST_DEFAULT_TESS_LEVELS, 0);
 		buf = LLVMGetParam(ctx->main_fn, ctx->param_rw_buffers);
-		buf = ac_build_indexed_load_const(&ctx->ac, buf, slot);
+		buf = ac_build_load_to_sgpr(&ctx->ac, buf, slot);
 		offset = decl->Semantic.Name == TGSI_SEMANTIC_DEFAULT_TESSINNER_SI ? 4 : 0;
 
 		for (i = 0; i < 4; i++)
@@ -1935,8 +1935,8 @@ 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));
+	return ac_build_load_to_sgpr(&ctx->ac, list_ptr,
+				     LLVMConstInt(ctx->i32, si_get_constbuf_slot(i), 0));
 }
 
 static LLVMValueRef load_ubo(struct ac_shader_abi *abi, LLVMValueRef index)
@@ -1948,7 +1948,7 @@ static LLVMValueRef load_ubo(struct ac_shader_abi *abi, LLVMValueRef index)
 	index = LLVMBuildAdd(ctx->ac.builder, index,
 			     LLVMConstInt(ctx->i32, SI_NUM_SHADER_BUFFERS, 0), "");
 
-	return ac_build_indexed_load_const(&ctx->ac, ptr, index);
+	return ac_build_load_to_sgpr(&ctx->ac, ptr, index);
 }
 
 static LLVMValueRef
@@ -1963,7 +1963,7 @@ load_ssbo(struct ac_shader_abi *abi, LLVMValueRef index, bool write)
 			     LLVMConstInt(ctx->i32, SI_NUM_SHADER_BUFFERS - 1, 0),
 			     index, "");
 
-	return ac_build_indexed_load_const(&ctx->ac, rsrc_ptr, index);
+	return ac_build_load_to_sgpr(&ctx->ac, rsrc_ptr, index);
 }
 
 static LLVMValueRef fetch_constant(
@@ -2008,7 +2008,7 @@ static LLVMValueRef fetch_constant(
 						      ctx->num_const_buffers);
 		index = LLVMBuildAdd(ctx->ac.builder, index,
 				     LLVMConstInt(ctx->i32, SI_NUM_SHADER_BUFFERS, 0), "");
-		bufp = ac_build_indexed_load_const(&ctx->ac, ptr, index);
+		bufp = ac_build_load_to_sgpr(&ctx->ac, ptr, index);
 	} else
 		bufp = load_const_buffer_desc(ctx, buf);
 
@@ -2283,7 +2283,7 @@ static void si_llvm_emit_clipvertex(struct lp_build_tgsi_context *bld_base,
 	LLVMValueRef ptr = LLVMGetParam(ctx->main_fn, ctx->param_rw_buffers);
 	LLVMValueRef constbuf_index = LLVMConstInt(ctx->i32,
 						   SI_VS_CONST_CLIP_PLANES, 0);
-	LLVMValueRef const_resource = ac_build_indexed_load_const(&ctx->ac, ptr, constbuf_index);
+	LLVMValueRef const_resource = ac_build_load_to_sgpr(&ctx->ac, ptr, constbuf_index);
 
 	for (reg_index = 0; reg_index < 2; reg_index ++) {
 		struct ac_export_args *args = &pos[2 + reg_index];
@@ -2440,7 +2440,7 @@ static void si_llvm_emit_streamout(struct si_shader_context *ctx,
 			LLVMValueRef offset = LLVMConstInt(ctx->i32,
 							   SI_VS_STREAMOUT_BUF0 + i, 0);
 
-			so_buffers[i] = ac_build_indexed_load_const(&ctx->ac, buf_ptr, offset);
+			so_buffers[i] = ac_build_load_to_sgpr(&ctx->ac, buf_ptr, offset);
 
 			LLVMValueRef so_offset = LLVMGetParam(ctx->main_fn,
 							      ctx->param_streamout_offset[i]);
@@ -4722,20 +4722,20 @@ static void preload_ring_buffers(struct si_shader_context *ctx)
 		LLVMValueRef offset = LLVMConstInt(ctx->i32, ring, 0);
 
 		ctx->esgs_ring =
-			ac_build_indexed_load_const(&ctx->ac, buf_ptr, offset);
+			ac_build_load_to_sgpr(&ctx->ac, buf_ptr, offset);
 	}
 
 	if (ctx->shader->is_gs_copy_shader) {
 		LLVMValueRef offset = LLVMConstInt(ctx->i32, SI_RING_GSVS, 0);
 
 		ctx->gsvs_ring[0] =
-			ac_build_indexed_load_const(&ctx->ac, buf_ptr, offset);
+			ac_build_load_to_sgpr(&ctx->ac, buf_ptr, offset);
 	} else if (ctx->type == PIPE_SHADER_GEOMETRY) {
 		const struct si_shader_selector *sel = ctx->shader->selector;
 		LLVMValueRef offset = LLVMConstInt(ctx->i32, SI_RING_GSVS, 0);
 		LLVMValueRef base_ring;
 
-		base_ring = ac_build_indexed_load_const(&ctx->ac, buf_ptr, offset);
+		base_ring = ac_build_load_to_sgpr(&ctx->ac, buf_ptr, offset);
 
 		/* The conceptual layout of the GSVS ring is
 		 *   v0c0 .. vLv0 v0c1 .. vLc1 ..
@@ -4818,7 +4818,7 @@ static void si_llvm_emit_polygon_stipple(struct si_shader_context *ctx,
 
 	/* Load the buffer descriptor. */
 	slot = LLVMConstInt(ctx->i32, SI_PS_CONST_POLY_STIPPLE, 0);
-	desc = ac_build_indexed_load_const(&ctx->ac, param_rw_buffers, slot);
+	desc = ac_build_load_to_sgpr(&ctx->ac, param_rw_buffers, slot);
 
 	/* The stipple pattern is 32x32, each row has 32 bits. */
 	offset = LLVMBuildMul(builder, address[1],
@@ -6887,7 +6887,7 @@ static void si_build_vs_prolog_function(struct si_shader_context *ctx,
 		LLVMValueRef buf_index =
 			LLVMConstInt(ctx->i32, SI_VS_CONST_INSTANCE_DIVISORS, 0);
 		instance_divisor_constbuf =
-			ac_build_indexed_load_const(&ctx->ac, list, buf_index);
+			ac_build_load_to_sgpr(&ctx->ac, list, buf_index);
 	}
 
 	for (i = 0; i <= key->vs_prolog.last_input; i++) {
diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
index a2b2b874d71..ec11c75b888 100644
--- a/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
+++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
@@ -145,7 +145,7 @@ LLVMValueRef si_load_image_desc(struct si_shader_context *ctx,
 		assert(desc_type == AC_DESC_IMAGE);
 	}
 
-	rsrc = ac_build_indexed_load_const(&ctx->ac, list, index);
+	rsrc = ac_build_load_to_sgpr(&ctx->ac, list, index);
 	if (dcc_off)
 		rsrc = force_dcc_off(ctx, rsrc);
 	return rsrc;
@@ -1134,7 +1134,7 @@ LLVMValueRef si_load_sampler_desc(struct si_shader_context *ctx,
 		break;
 	}
 
-	return ac_build_indexed_load_const(&ctx->ac, list, index);
+	return ac_build_load_to_sgpr(&ctx->ac, list, index);
 }
 
 /* Disable anisotropic filtering if BASE_LEVEL == LAST_LEVEL.