swr/rast: Add autogen of helper llvm intrinsics.
[mesa.git] / src / gallium / drivers / radeonsi / si_shader_tgsi_mem.c
index f5fa18fd38a3b374e36d76912a5c4e351cc85753..6a307c4ddba99f8c0a7483f68893a7427a8d9a00 100644 (file)
@@ -1,5 +1,6 @@
 /*
  * Copyright 2017 Advanced Micro Devices, Inc.
+ * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -1873,6 +1874,63 @@ static void si_llvm_emit_txqs(
        emit_data->output[emit_data->chan] = samples;
 }
 
+static void si_llvm_emit_fbfetch(const struct lp_build_tgsi_action *action,
+                                struct lp_build_tgsi_context *bld_base,
+                                struct lp_build_emit_data *emit_data)
+{
+       struct si_shader_context *ctx = si_shader_context(bld_base);
+       struct ac_image_args args = {};
+       LLVMValueRef ptr, image, fmask, addr_vec;
+
+       /* Ignore src0, because KHR_blend_func_extended disallows multiple render
+        * targets.
+        */
+
+       /* Load the image descriptor. */
+       STATIC_ASSERT(SI_PS_IMAGE_COLORBUF0 % 2 == 0);
+       ptr = LLVMGetParam(ctx->main_fn, ctx->param_rw_buffers);
+       ptr = LLVMBuildPointerCast(ctx->ac.builder, ptr,
+                                  ac_array_in_const32_addr_space(ctx->v8i32), "");
+       image = ac_build_load_to_sgpr(&ctx->ac, ptr,
+                       LLVMConstInt(ctx->i32, SI_PS_IMAGE_COLORBUF0 / 2, 0));
+
+       LLVMValueRef addr[4];
+       unsigned chan = 0;
+
+       addr[chan++] = si_unpack_param(ctx, SI_PARAM_POS_FIXED_PT, 0, 16);
+
+       if (!ctx->shader->key.mono.u.ps.fbfetch_is_1D)
+               addr[chan++] = si_unpack_param(ctx, SI_PARAM_POS_FIXED_PT, 16, 16);
+
+       /* Get the current render target layer index. */
+       if (ctx->shader->key.mono.u.ps.fbfetch_layered)
+               addr[chan++] = si_unpack_param(ctx, SI_PARAM_ANCILLARY, 16, 11);
+
+       if (ctx->shader->key.mono.u.ps.fbfetch_msaa)
+               addr[chan++] = si_get_sample_id(ctx);
+
+       while (chan < 4)
+               addr[chan++] = LLVMGetUndef(ctx->i32);
+
+       if (ctx->shader->key.mono.u.ps.fbfetch_msaa) {
+               fmask = ac_build_load_to_sgpr(&ctx->ac, ptr,
+                       LLVMConstInt(ctx->i32, SI_PS_IMAGE_COLORBUF0_FMASK / 2, 0));
+
+               ac_apply_fmask_to_sample(&ctx->ac, fmask, addr, false);
+       }
+
+       addr_vec = ac_build_gather_values(&ctx->ac, addr, ARRAY_SIZE(addr));
+
+       args.opcode = ac_image_load;
+       args.resource = image;
+       args.addr = addr_vec;
+       args.dmask = 0xf;
+       args.da = ctx->shader->key.mono.u.ps.fbfetch_layered;
+
+       emit_data->output[emit_data->chan] =
+               ac_build_image_opcode(&ctx->ac, &args);
+}
+
 static const struct lp_build_tgsi_action tex_action = {
        .fetch_args = tex_fetch_args,
        .emit = build_tex_intrinsic,
@@ -1905,6 +1963,8 @@ void si_shader_context_init_mem(struct si_shader_context *ctx)
        bld_base->op_actions[TGSI_OPCODE_LODQ] = tex_action;
        bld_base->op_actions[TGSI_OPCODE_TXQS].emit = si_llvm_emit_txqs;
 
+       bld_base->op_actions[TGSI_OPCODE_FBFETCH].emit = si_llvm_emit_fbfetch;
+
        bld_base->op_actions[TGSI_OPCODE_LOAD].fetch_args = load_fetch_args;
        bld_base->op_actions[TGSI_OPCODE_LOAD].emit = load_emit;
        bld_base->op_actions[TGSI_OPCODE_STORE].fetch_args = store_fetch_args;