nir/lower_wrmasks: drop callback

All drivers use the same callback and it is unlikely that new drivers will use
this pass since it has better replacements today (lower_mem_bit_sizes for
memory, and it never worked for I/O). This should discourage as much.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38533>

src/broadcom/compiler/vir.c

@@ -1399,21 +1399,6 @@ v3d_instr_delay_cb(nir_instr *instr, void *data)
    return 0;
 }
 
-static bool
-should_split_wrmask(const nir_instr *instr, const void *data)
-{
-        nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
-        switch (intr->intrinsic) {
-        case nir_intrinsic_store_ssbo:
-        case nir_intrinsic_store_shared:
-        case nir_intrinsic_store_global:
-        case nir_intrinsic_store_scratch:
-                return true;
-        default:
-                return false;
-        }
-}
-
 static nir_intrinsic_instr *
 nir_instr_as_constant_ubo_load(nir_instr *inst)
 {
@@ -1857,7 +1842,7 @@ v3d_attempt_compile(struct v3d_compile *c)
                  glsl_get_natural_size_align_bytes);
 
         NIR_PASS(_, c->s, v3d_nir_lower_global_2x32);
-        NIR_PASS(_, c->s, nir_lower_wrmasks, should_split_wrmask, c->s);
+        NIR_PASS(_, c->s, nir_lower_wrmasks);
         NIR_PASS(_, c->s, v3d_nir_lower_load_store_bitsize);
         NIR_PASS(_, c->s, v3d_nir_lower_scratch);
 

src/compiler/nir/nir.h

@@ -6092,7 +6092,7 @@ bool nir_lower_wpos_center(nir_shader *shader);
 bool nir_lower_pntc_ytransform(nir_shader *shader,
                                const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH]);
 
-bool nir_lower_wrmasks(nir_shader *shader, nir_instr_filter_cb cb, const void *data);
+bool nir_lower_wrmasks(nir_shader *shader);
 
 bool nir_lower_fb_read(nir_shader *shader);
 

src/compiler/nir/nir_lower_wrmasks.c

@@ -165,21 +165,9 @@ split_wrmask(nir_builder *b, nir_intrinsic_instr *intr)
    nir_instr_remove(&intr->instr);
 }
 
-struct nir_lower_wrmasks_state {
-   nir_instr_filter_cb cb;
-   const void *data;
-};
-
 static bool
-nir_lower_wrmasks_instr(nir_builder *b, nir_instr *instr, void *data)
+lower(nir_builder *b, nir_intrinsic_instr *intr, void *data)
 {
-   struct nir_lower_wrmasks_state *state = data;
-
-   if (instr->type != nir_instr_type_intrinsic)
-      return false;
-
-   nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
-
    /* if no wrmask, then skip it: */
    if (!nir_intrinsic_has_write_mask(intr))
       return false;
@@ -194,25 +182,14 @@ nir_lower_wrmasks_instr(nir_builder *b, nir_instr *instr, void *data)
 
    assert(offset_src(intr->intrinsic) >= 0);
 
-   /* does backend need us to lower this intrinsic? */
-   if (state->cb && !state->cb(instr, state->data))
-      return false;
-
    split_wrmask(b, intr);
 
    return true;
 }
 
 bool
-nir_lower_wrmasks(nir_shader *shader, nir_instr_filter_cb cb, const void *data)
+nir_lower_wrmasks(nir_shader *shader)
 {
-   struct nir_lower_wrmasks_state state = {
-      .cb = cb,
-      .data = data,
-   };
-
-   return nir_shader_instructions_pass(shader,
-                                       nir_lower_wrmasks_instr,
-                                       nir_metadata_control_flow,
-                                       &state);
+   return nir_shader_intrinsics_pass(shader, lower, nir_metadata_control_flow,
+                                     NULL);
 }

src/freedreno/ir3/ir3_nir.c

@@ -426,22 +426,6 @@ ir3_optimize_loop(struct ir3_compiler *compiler,
    return did_progress;
 }
 
-static bool
-should_split_wrmask(const nir_instr *instr, const void *data)
-{
-   nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
-
-   switch (intr->intrinsic) {
-   case nir_intrinsic_store_ssbo:
-   case nir_intrinsic_store_shared:
-   case nir_intrinsic_store_global:
-   case nir_intrinsic_store_scratch:
-      return true;
-   default:
-      return false;
-   }
-}
-
 static bool
 ir3_nir_lower_ssbo_size_filter(const nir_instr *instr, const void *data)
 {
@@ -705,7 +689,7 @@ ir3_finalize_nir(struct ir3_compiler *compiler,
    NIR_PASS(_, s, nir_lower_frexp);
    NIR_PASS(_, s, nir_lower_amul, ir3_glsl_type_size);
 
-   OPT(s, nir_lower_wrmasks, should_split_wrmask, s);
+   OPT(s, nir_lower_wrmasks);
 
    OPT(s, nir_lower_tex, &tex_options);
    OPT(s, nir_lower_load_const_to_scalar);
@@ -1223,7 +1207,7 @@ ir3_nir_lower_variant(struct ir3_shader_variant *so,
    }
 
    /* Lower scratch writemasks */
-   progress |= OPT(s, nir_lower_wrmasks, should_split_wrmask, s);
+   progress |= OPT(s, nir_lower_wrmasks);
    progress |= OPT(s, nir_lower_atomics, atomic_supported);
 
    if (OPT(s, nir_lower_locals_to_regs, 1)) {

src/panfrost/compiler/bifrost_compile.c

@@ -5413,26 +5413,6 @@ va_gather_stats(bi_context *ctx, unsigned size, struct valhall_stats *out)
                       MAX3(stats.v, stats.t, stats.ls));
 }
 
-/* Split stores to memory. We don't split stores to vertex outputs, since
- * nir_lower_io_vars_to_temporaries will ensure there's only a single write.
- */
-
-static bool
-should_split_wrmask(const nir_instr *instr, UNUSED const void *data)
-{
-   nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
-
-   switch (intr->intrinsic) {
-   case nir_intrinsic_store_ssbo:
-   case nir_intrinsic_store_shared:
-   case nir_intrinsic_store_global:
-   case nir_intrinsic_store_scratch:
-      return true;
-   default:
-      return false;
-   }
-}
-
 /*
  * Some operations are only available as 32-bit instructions. 64-bit floats are
  * unsupported and ints are lowered with nir_lower_int64.  Certain 8-bit and
@@ -5704,7 +5684,7 @@ bi_optimize_loop_nir(nir_shader *nir, unsigned gpu_id, bool allow_copies)
       NIR_PASS(progress, nir, nir_opt_deref);
 
       NIR_PASS(progress, nir, nir_lower_vars_to_ssa);
-      NIR_PASS(progress, nir, nir_lower_wrmasks, should_split_wrmask, NULL);
+      NIR_PASS(progress, nir, nir_lower_wrmasks);
 
       if (allow_copies) {
          /* Only run this pass in the first call to bi_optimize_nir. Later