nir/algebraic: shrink 64-bit bitwise operations with 0/-1 constant half

fossil-db (navi21):
Totals from 457 (0.34% of 135636) affected shaders:
Instrs: 259349 -> 250383 (-3.46%)
CodeSize: 1411976 -> 1369136 (-3.03%)
Latency: 2175961 -> 2148158 (-1.28%)
InvThroughput: 502206 -> 490244 (-2.38%)
Copies: 15238 -> 15232 (-0.04%); split: -0.07%, +0.03%

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19748>

src/compiler/nir/nir_opt_algebraic.py

@@ -1265,6 +1265,40 @@ optimizations.extend([
    (('ior', ('ior', a, b), b), ('ior', a, b)),
    (('iand', ('ior', a, b), b), b),
    (('iand', ('iand', a, b), b), ('iand', a, b)),
+
+   (('iand@64', a, '#b(is_lower_half_zero)'),
+    ('pack_64_2x32_split', 0,
+                           ('iand', ('unpack_64_2x32_split_y', a), ('unpack_64_2x32_split_y', b)))),
+   (('iand@64', a, '#b(is_upper_half_zero)'),
+    ('pack_64_2x32_split', ('iand', ('unpack_64_2x32_split_x', a), ('unpack_64_2x32_split_x', b)),
+                           0)),
+   (('iand@64', a, '#b(is_lower_half_negative_one)'),
+    ('pack_64_2x32_split', ('unpack_64_2x32_split_x', a),
+                           ('iand', ('unpack_64_2x32_split_y', a), ('unpack_64_2x32_split_y', b)))),
+   (('iand@64', a, '#b(is_upper_half_negative_one)'),
+    ('pack_64_2x32_split', ('iand', ('unpack_64_2x32_split_x', a), ('unpack_64_2x32_split_x', b)),
+                           ('unpack_64_2x32_split_y', a))),
+
+   (('ior@64', a, '#b(is_lower_half_zero)'),
+    ('pack_64_2x32_split', ('unpack_64_2x32_split_x', a),
+                           ('ior', ('unpack_64_2x32_split_y', a), ('unpack_64_2x32_split_y', b)))),
+   (('ior@64', a, '#b(is_upper_half_zero)'),
+    ('pack_64_2x32_split', ('ior', ('unpack_64_2x32_split_x', a), ('unpack_64_2x32_split_x', b)),
+                           ('unpack_64_2x32_split_y', a))),
+   (('ior@64', a, '#b(is_lower_half_negative_one)'),
+    ('pack_64_2x32_split', -1,
+                           ('ior', ('unpack_64_2x32_split_y', a), ('unpack_64_2x32_split_y', b)))),
+   (('ior@64', a, '#b(is_upper_half_negative_one)'),
+    ('pack_64_2x32_split', ('ior', ('unpack_64_2x32_split_x', a), ('unpack_64_2x32_split_x', b)),
+                           -1)),
+
+   (('ixor@64', a, '#b(is_lower_half_zero)'),
+    ('pack_64_2x32_split', ('unpack_64_2x32_split_x', a),
+                           ('ixor', ('unpack_64_2x32_split_y', a), ('unpack_64_2x32_split_y', b)))),
+   (('ixor@64', a, '#b(is_upper_half_zero)'),
+    ('pack_64_2x32_split', ('ixor', ('unpack_64_2x32_split_x', a), ('unpack_64_2x32_split_x', b)),
+                           ('unpack_64_2x32_split_y', a))),
+
    # DeMorgan's Laws
    (('iand', ('inot', a), ('inot', b)), ('inot', ('ior',  a, b))),
    (('ior',  ('inot', a), ('inot', b)), ('inot', ('iand', a, b))),

src/compiler/nir/nir_search_helpers.h

@@ -471,7 +471,7 @@ is_upper_half_zero(UNUSED struct hash_table *ht, const nir_alu_instr *instr,
 
    for (unsigned i = 0; i < num_components; i++) {
       unsigned half_bit_size = nir_src_bit_size(instr->src[src].src) / 2;
-      uint32_t high_bits = ((1 << half_bit_size) - 1) << half_bit_size;
+      uint64_t high_bits = u_bit_consecutive64(half_bit_size, half_bit_size);
       if ((nir_src_comp_as_uint(instr->src[src].src,
                                 swizzle[i]) & high_bits) != 0) {
          return false;
@@ -495,9 +495,45 @@ is_lower_half_zero(UNUSED struct hash_table *ht, const nir_alu_instr *instr,
       return false;
 
    for (unsigned i = 0; i < num_components; i++) {
-      uint32_t low_bits =
+      uint64_t low_bits = u_bit_consecutive64(0, nir_src_bit_size(instr->src[src].src) / 2);
-         (1 << (nir_src_bit_size(instr->src[src].src) / 2)) - 1;
+      if ((nir_src_comp_as_uint(instr->src[src].src, swizzle[i]) & low_bits) != 0)
-      if ((nir_src_comp_as_int(instr->src[src].src, swizzle[i]) & low_bits) != 0)
+         return false;
+   }
+
+   return true;
+}
+
+static inline bool
+is_upper_half_negative_one(UNUSED struct hash_table *ht, const nir_alu_instr *instr,
+                           unsigned src, unsigned num_components,
+                           const uint8_t *swizzle)
+{
+   if (nir_src_as_const_value(instr->src[src].src) == NULL)
+      return false;
+
+   for (unsigned i = 0; i < num_components; i++) {
+      unsigned half_bit_size = nir_src_bit_size(instr->src[src].src) / 2;
+      uint64_t high_bits = u_bit_consecutive64(half_bit_size, half_bit_size);
+      if ((nir_src_comp_as_uint(instr->src[src].src,
+                                swizzle[i]) & high_bits) != high_bits) {
+         return false;
+      }
+   }
+
+   return true;
+}
+
+static inline bool
+is_lower_half_negative_one(UNUSED struct hash_table *ht, const nir_alu_instr *instr,
+                           unsigned src, unsigned num_components,
+                           const uint8_t *swizzle)
+{
+   if (nir_src_as_const_value(instr->src[src].src) == NULL)
+      return false;
+
+   for (unsigned i = 0; i < num_components; i++) {
+      uint64_t low_bits = u_bit_consecutive64(0, nir_src_bit_size(instr->src[src].src) / 2);
+      if ((nir_src_comp_as_uint(instr->src[src].src, swizzle[i]) & low_bits) != low_bits)
          return false;
    }