nir: add a base offset for printf indexing

This will allow a driver to use a single table of printf strings
across all shaders.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25814>

src/compiler/nir/nir.h

@@ -6444,6 +6444,7 @@ bool nir_lower_helper_writes(nir_shader *shader, bool lower_plain_stores);
 
 typedef struct nir_lower_printf_options {
    unsigned max_buffer_size;
+   bool     use_printf_base_identifier;
 } nir_lower_printf_options;
 
 bool nir_lower_printf(nir_shader *nir, const nir_lower_printf_options *options);

src/compiler/nir/nir_divergence_analysis.c

@@ -267,6 +267,7 @@ visit_intrinsic(nir_intrinsic_instr *instr, struct divergence_state *state)
    case nir_intrinsic_preamble_start_ir3:
    case nir_intrinsic_optimization_barrier_sgpr_amd:
    case nir_intrinsic_load_printf_buffer_address:
+   case nir_intrinsic_load_printf_base_identifier:
       is_divergent = false;
       break;
 

src/compiler/nir/nir_intrinsics.py

@@ -1227,6 +1227,10 @@ intrinsic("printf", src_comp=[1, 1], dest_comp=1, bit_sizes=[32])
 # in a buffer, nir_lower_printf will do that, but requires
 # the driver to at least provide a base location
 system_value("printf_buffer_address", 1, bit_sizes=[32,64])
+# If driver wants to have all printfs from various shaders merged into a
+# single output buffer, it needs each shader to have its own base identifier
+# from which each printf is indexed.
+system_value("printf_base_identifier", 1, bit_sizes=[32])
 
 # Mesh shading MultiView intrinsics
 system_value("mesh_view_count", 1)

src/compiler/nir/nir_lower_printf.c

@@ -34,7 +34,15 @@ lower_printf_intrin(nir_builder *b, nir_intrinsic_instr *prntf, void *_options)
    if (prntf->intrinsic != nir_intrinsic_printf)
       return false;
 
+   b->cursor = nir_before_instr(&prntf->instr);
+
    nir_def *fmt_str_id = prntf->src[0].ssa;
+   if (options && options->use_printf_base_identifier) {
+      fmt_str_id = nir_iadd(b,
+                            nir_load_printf_base_identifier(b),
+                            fmt_str_id);
+   }
+
    nir_deref_instr *args = nir_src_as_deref(prntf->src[1]);
    assert(args->deref_type == nir_deref_type_var);
 
@@ -43,7 +51,6 @@ lower_printf_intrin(nir_builder *b, nir_intrinsic_instr *prntf, void *_options)
    /* Atomic add a buffer size counter to determine where to write.  If
     * overflowed, return -1, otherwise, store the arguments and return 0.
     */
-   b->cursor = nir_before_instr(&prntf->instr);
    nir_def *buffer_addr = nir_load_printf_buffer_address(b, ptr_bit_size);
    nir_deref_instr *buffer =
       nir_build_deref_cast(b, buffer_addr, nir_var_mem_global,

src/gallium/frontends/rusticl/core/kernel.rs

@@ -463,6 +463,7 @@ fn lower_and_optimize_nir(
     nir_pass!(nir, nir_dedup_inline_samplers);
 
     let printf_opts = nir_lower_printf_options {
+        use_printf_base_identifier: false,
         max_buffer_size: dev.printf_buffer_size() as u32,
     };
     nir_pass!(nir, nir_lower_printf, &printf_opts);