freedreno/ir3: Add wide load/store lowering

Lower load/store for vectors wider than 4.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13300>

diff --git a/src/freedreno/ir3/ir3_nir.c b/src/freedreno/ir3/ir3_nir.c
index 885d4a5..19e41ac 100644 (file)
--- a/src/freedreno/ir3/ir3_nir.c
+++ b/src/freedreno/ir3/ir3_nir.c
@@ -707,6 +707,7 @@ ir3_nir_lower_variant(struct ir3_shader_variant *so, nir_shader *s)
    /* Lower scratch writemasks */
    progress |= OPT(s, nir_lower_wrmasks, should_split_wrmask, s);
 
+   progress |= OPT(s, ir3_nir_lower_wide_load_store);
    progress |= OPT(s, ir3_nir_lower_64b_global);
    progress |= OPT(s, ir3_nir_lower_64b_intrinsics);
    progress |= OPT(s, ir3_nir_lower_64b_undef);

diff --git a/src/freedreno/ir3/ir3_nir.h b/src/freedreno/ir3/ir3_nir.h
index 2de6adf..dccd8bc 100644 (file)
--- a/src/freedreno/ir3/ir3_nir.h
+++ b/src/freedreno/ir3/ir3_nir.h
@@ -42,6 +42,7 @@ bool ir3_nir_lower_load_barycentric_at_offset(nir_shader *shader);
 bool ir3_nir_move_varying_inputs(nir_shader *shader);
 int ir3_nir_coord_offset(nir_ssa_def *ssa);
 bool ir3_nir_lower_tex_prefetch(nir_shader *shader);
+bool ir3_nir_lower_wide_load_store(nir_shader *shader);
 
 void ir3_nir_lower_to_explicit_output(nir_shader *shader,
                                       struct ir3_shader_variant *v,

diff --git a/src/freedreno/ir3/ir3_nir_lower_wide_load_store.c b/src/freedreno/ir3/ir3_nir_lower_wide_load_store.c
new file mode 100644 (file)
index 0000000..617d9b1
--- /dev/null
+++ b/src/freedreno/ir3/ir3_nir_lower_wide_load_store.c
@@ -0,0 +1,118 @@
+/*
+ * Copyright © 2021 Google, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "ir3_nir.h"
+
+
+/*
+ * Lowering for wide (larger than vec4) load/store
+ */
+
+static bool
+lower_wide_load_store_filter(const nir_instr *instr, const void *unused)
+{
+   (void)unused;
+
+   if (instr->type != nir_instr_type_intrinsic)
+      return false;
+
+   nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
+
+   if (is_intrinsic_store(intr->intrinsic))
+      return nir_intrinsic_src_components(intr, 0) > 4;
+
+   if (is_intrinsic_load(intr->intrinsic))
+      return nir_intrinsic_dest_components(intr) > 4;
+
+   return false;
+}
+
+static nir_ssa_def *
+lower_wide_load_store(nir_builder *b, nir_instr *instr, void *unused)
+{
+   (void)unused;
+
+   nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
+
+   if (is_intrinsic_store(intr->intrinsic)) {
+      unsigned num_comp = nir_intrinsic_src_components(intr, 0);
+      unsigned wrmask = nir_intrinsic_write_mask(intr);
+      nir_ssa_def *val = nir_ssa_for_src(b, intr->src[0], num_comp);
+      nir_ssa_def *addr = nir_ssa_for_src(b, intr->src[1], 1);
+
+      for (unsigned off = 0; off < num_comp; off += 4) {
+         unsigned c = MIN2(num_comp - off, 4);
+         nir_ssa_def *v = nir_channels(b, val, BITFIELD_MASK(c) << off);
+
+         nir_intrinsic_instr *store =
+               nir_intrinsic_instr_create(b->shader, intr->intrinsic);
+         store->num_components = c;
+         store->src[0] = nir_src_for_ssa(v);
+         store->src[1] = nir_src_for_ssa(addr);
+         nir_intrinsic_set_align(store, nir_intrinsic_align(intr), 0);
+         nir_intrinsic_set_write_mask(store, (wrmask >> off) & 0xf);
+         nir_builder_instr_insert(b, &store->instr);
+
+         addr = nir_iadd(b,
+               nir_imm_intN_t(b, (c * val->bit_size) / 8, addr->bit_size),
+               addr);
+      }
+
+      return NIR_LOWER_INSTR_PROGRESS_REPLACE;
+   } else {
+      unsigned num_comp = nir_intrinsic_dest_components(intr);
+      unsigned bit_size = nir_dest_bit_size(intr->dest);
+      nir_ssa_def *addr = nir_ssa_for_src(b, intr->src[0], 1);
+      nir_ssa_def *components[num_comp];
+
+      for (unsigned off = 0; off < num_comp;) {
+         unsigned c = MIN2(num_comp - off, 4);
+
+         nir_intrinsic_instr *load =
+            nir_intrinsic_instr_create(b->shader, intr->intrinsic);
+         load->num_components = c;
+         load->src[0] = nir_src_for_ssa(addr);
+         nir_intrinsic_set_align(load, nir_intrinsic_align(intr), 0);
+         nir_ssa_dest_init(&load->instr, &load->dest, c, bit_size, NULL);
+         nir_builder_instr_insert(b, &load->instr);
+
+         addr = nir_iadd(b,
+               nir_imm_intN_t(b, (c * bit_size) / 8, addr->bit_size),
+               addr);
+
+         for (unsigned i = 0; i < c; i++) {
+            components[off++] = nir_channel(b, &load->dest.ssa, i);
+         }
+      }
+
+      return nir_build_alu_src_arr(b, nir_op_vec(num_comp), components);
+   }
+}
+
+bool
+ir3_nir_lower_wide_load_store(nir_shader *shader)
+{
+   return nir_shader_lower_instructions(
+         shader, lower_wide_load_store_filter,
+         lower_wide_load_store, NULL);
+}

diff --git a/src/freedreno/ir3/meson.build b/src/freedreno/ir3/meson.build
index 127a6a5..55f5ad0 100644 (file)
--- a/src/freedreno/ir3/meson.build
+++ b/src/freedreno/ir3/meson.build
@@ -101,6 +101,7 @@ libfreedreno_ir3_files = files(
   'ir3_nir_lower_tess.c',
   'ir3_nir_lower_tex_prefetch.c',
   'ir3_nir_lower_tg4_to_tex.c',
+  'ir3_nir_lower_wide_load_store.c',
   'ir3_nir_move_varying_inputs.c',
   'ir3_postsched.c',
   'ir3_print.c',