Mesa (master): nir: add opcodes for *find_msb_rev and lowering

[GitLab Mirror]

Module: Mesa
Branch: master
Commit: a5747f8ab357ff00c630b937b221e5fb59d90289
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=a5747f8ab357ff00c630b937b221e5fb59d90289

Author: Gert Wollny <gert.wollny at collabora.com>
Date:   Wed Mar 10 09:38:18 2021 +0100

nir: add opcodes for *find_msb_rev and lowering

Some hardware supports a version of find_msb where the bits are counted
starting at the high bit, and this needs some lowering to obtain the
value that is expected by *find_msb

Signed-off-by: Gert Wollny <gert.wollny at collabora.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg at gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9452>

---

 src/compiler/nir/nir.h                |  2 ++
 src/compiler/nir/nir_opcodes.py       | 26 ++++++++++++++++++++++++++
 src/compiler/nir/nir_opt_algebraic.py | 12 ++++++++++++
 3 files changed, 40 insertions(+)

diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
index 3620dde1a30..ec841c43536 100644
--- a/src/compiler/nir/nir.h
+++ b/src/compiler/nir/nir.h
@@ -3160,6 +3160,8 @@ typedef struct nir_shader_compiler_options {
    bool lower_bit_count;
    /** Lowers ifind_msb to compare and ufind_msb */
    bool lower_ifind_msb;
+   /** Lowers ifind_msb and ufind_msb to reverse variants */
+   bool lower_find_msb_to_reverse;
    /** Lowers find_lsb to ufind_msb and logic ops */
    bool lower_find_lsb;
    bool lower_uadd_carry;
diff --git a/src/compiler/nir/nir_opcodes.py b/src/compiler/nir/nir_opcodes.py
index 854f7562635..012c7bfea09 100644
--- a/src/compiler/nir/nir_opcodes.py
+++ b/src/compiler/nir/nir_opcodes.py
@@ -454,6 +454,16 @@ for (int bit = bit_size - 1; bit >= 0; bit--) {
 }
 """)
 
+unop_convert("ufind_msb_rev", tint32, tuint, """
+dst = -1;
+for (int bit = 0; bit < bit_size; bit++) {
+   if ((src0 << bit) & 0x80000000) {
+      dst = bit;
+      break;
+   }
+}
+""")
+
 unop("uclz", tuint32, """
 int bit;
 for (bit = bit_size - 1; bit >= 0; bit--) {
@@ -477,6 +487,22 @@ for (int bit = 31; bit >= 0; bit--) {
 }
 """)
 
+unop_convert("ifind_msb_rev", tint32, tuint, """
+dst = -1;
+if (src0 != 0 || src0 != -1) {
+   for (int bit = 0; bit < 31; bit++) {
+      /* If src0 < 0, we're looking for the first 0 bit.
+       * if src0 >= 0, we're looking for the first 1 bit.
+       */
+      if ((((src0 << bit) & 0x40000000) && (src0 >= 0)) ||
+          ((!((src0 << bit) & 0x40000000)) && (src0 < 0))) {
+         dst = bit;
+         break;
+      }
+   }
+}
+""")
+
 unop_convert("find_lsb", tint32, tint, """
 dst = -1;
 for (unsigned bit = 0; bit < bit_size; bit++) {
diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py
index 5938561735e..32d1dab1743 100644
--- a/src/compiler/nir/nir_opt_algebraic.py
+++ b/src/compiler/nir/nir_opt_algebraic.py
@@ -1523,6 +1523,18 @@ optimizations.extend([
     ('ufind_msb', ('bcsel', ('ilt', 'value', 0), ('inot', 'value'), 'value')),
     'options->lower_ifind_msb'),
 
+   (('ifind_msb', 'value'),
+    ('bcsel', ('ige', ('ifind_msb_rev', 'value'), 0),
+     ('isub', 31, ('ifind_msb_rev', 'value')),
+     ('ifind_msb_rev', 'value')),
+    'options->lower_find_msb_to_reverse'),
+
+    (('ufind_msb', 'value'),
+     ('bcsel', ('ige', ('ufind_msb_rev', 'value'), 0),
+      ('isub', 31, ('ufind_msb_rev', 'value')),
+      ('ufind_msb_rev', 'value')),
+     'options->lower_find_msb_to_reverse'),
+
    (('find_lsb', 'value'),
     ('ufind_msb', ('iand', 'value', ('ineg', 'value'))),
     'options->lower_find_lsb'),