[Mesa-dev] [PATCH 3/6] radeonsi: use buffer_store_format_x & xy

[Marek Olšák]

From: Marek Olšák <marek.olsak at amd.com>

---
 .../drivers/radeonsi/si_shader_tgsi_mem.c     | 25 +++++++++++++------
 1 file changed, 17 insertions(+), 8 deletions(-)

diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
index 6decedc4cce..727def56f65 100644
--- a/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
+++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
@@ -694,79 +694,88 @@ static void store_emit(
 
 	bool writeonly_memory = is_oneway_access_only(inst, info,
 						      info->shader_buffers_load |
 						      info->shader_buffers_atomic,
 						      info->images_load |
 						      info->images_atomic,
 						      info->uses_bindless_buffer_load |
 						      info->uses_bindless_buffer_atomic,
 						      info->uses_bindless_image_load |
 						      info->uses_bindless_image_atomic);
-	LLVMValueRef chans[4], value;
+	LLVMValueRef chans[4];
 	LLVMValueRef vindex = ctx->i32_0;
 	LLVMValueRef voffset = ctx->i32_0;
 	struct ac_image_args args = {};
 
 	for (unsigned chan = 0; chan < 4; ++chan)
 		chans[chan] = lp_build_emit_fetch(bld_base, inst, 1, chan);
 
-	value = ac_build_gather_values(&ctx->ac, chans, 4);
-
 	if (inst->Dst[0].Register.File == TGSI_FILE_BUFFER) {
 		args.resource = shader_buffer_fetch_rsrc(ctx, &resource_reg, false);
 		voffset = ac_to_integer(&ctx->ac, lp_build_emit_fetch(bld_base, inst, 0, 0));
 	} else {
 		image_fetch_rsrc(bld_base, &resource_reg, true, target, &args.resource);
 		image_fetch_coords(bld_base, inst, 0, args.resource, args.coords);
 		vindex = args.coords[0]; /* for buffers only */
 	}
 
 	if (inst->Memory.Qualifier & TGSI_MEMORY_VOLATILE)
 		ac_build_waitcnt(&ctx->ac, VM_CNT);
 
 	bool is_image = inst->Dst[0].Register.File != TGSI_FILE_BUFFER;
 	args.cache_policy = get_cache_policy(ctx, inst,
 					     false, /* atomic */
 					     is_image, /* may_store_unaligned */
 					     writeonly_memory);
 
 	if (inst->Dst[0].Register.File == TGSI_FILE_BUFFER) {
 		store_emit_buffer(ctx, args.resource, inst->Dst[0].Register.WriteMask,
-				  value, voffset, args.cache_policy, writeonly_memory);
+				  ac_build_gather_values(&ctx->ac, chans, 4),
+				  voffset, args.cache_policy, writeonly_memory);
 		return;
 	}
 
 	if (target == TGSI_TEXTURE_BUFFER) {
+		unsigned num_channels = util_last_bit(inst->Dst[0].Register.WriteMask);
+		num_channels = util_next_power_of_two(num_channels);
+
 		LLVMValueRef buf_args[6] = {
-			value,
+			ac_build_gather_values(&ctx->ac, chans, 4),
 			args.resource,
 			vindex,
 			ctx->i32_0, /* voffset */
 		};
 
 		if (HAVE_LLVM >= 0x0800) {
 			buf_args[4] = ctx->i32_0; /* soffset */
 			buf_args[5] = LLVMConstInt(ctx->i1, args.cache_policy, 0);
 		} else {
 			buf_args[4] = LLVMConstInt(ctx->i1, !!(args.cache_policy & ac_glc), 0);
 			buf_args[5] = LLVMConstInt(ctx->i1, !!(args.cache_policy & ac_slc), 0);
 		}
 
+		const char *types[] = { "f32", "v2f32", "v4f32" };
+		char name[128];
+
+		snprintf(name, sizeof(name), "%s.%s",
+			 HAVE_LLVM >= 0x0800 ? "llvm.amdgcn.struct.buffer.store.format" :
+					       "llvm.amdgcn.buffer.store.format",
+			 types[CLAMP(num_channels, 1, 3) - 1]);
+
 		emit_data->output[emit_data->chan] = ac_build_intrinsic(
 			&ctx->ac,
-			HAVE_LLVM >= 0x0800 ? "llvm.amdgcn.struct.buffer.store.format.v4f32" :
-					      "llvm.amdgcn.buffer.store.format.v4f32",
+			name,
 			ctx->voidt, buf_args, 6,
 			ac_get_store_intr_attribs(writeonly_memory));
 	} else {
 		args.opcode = ac_image_store;
-		args.data[0] = value;
+		args.data[0] = ac_build_gather_values(&ctx->ac, chans, 4);
 		args.dim = ac_image_dim_from_tgsi_target(ctx->screen, inst->Memory.Texture);
 		args.attributes = ac_get_store_intr_attribs(writeonly_memory);
 		args.dmask = 0xf;
 
 		emit_data->output[emit_data->chan] =
 			ac_build_image_opcode(&ctx->ac, &args);
 	}
 }
 
 static void atomic_emit_memory(struct si_shader_context *ctx,
-- 
2.17.1