Mesa (main): nir: Add nir_op_sad_u8x4 which corresponds to AMD's v_sad_u8.

[GitLab Mirror]

Module: Mesa
Branch: main
Commit: c92dab8e2b6964b6dbd9ea122d7ff819efd45244
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=c92dab8e2b6964b6dbd9ea122d7ff819efd45244

Author: Timur Kristóf <timur.kristof at gmail.com>
Date:   Fri May 28 21:53:06 2021 +0200

nir: Add nir_op_sad_u8x4 which corresponds to AMD's v_sad_u8.

NIR currently doesn't have any intrinsics for a horizontal packed add,
so this one is modeled after AMD's v_sad_u8.

Signed-off-by: Timur Kristóf <timur.kristof at gmail.com>
Reviewed-by: Tony Wasserka <tony.wasserka at gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11072>

---

 src/compiler/nir/nir_opcodes.py       | 18 ++++++++++++++++++
 src/compiler/nir/nir_range_analysis.c |  3 +++
 2 files changed, 21 insertions(+)

diff --git a/src/compiler/nir/nir_opcodes.py b/src/compiler/nir/nir_opcodes.py
index eda8d11cb3d..b39c7b57498 100644
--- a/src/compiler/nir/nir_opcodes.py
+++ b/src/compiler/nir/nir_opcodes.py
@@ -1097,6 +1097,24 @@ if (bits == 0) {
 }
 """)
 
+triop_horiz("sad_u8x4", 1, 1, 1, 1, """
+uint8_t s0_b0 = (src0.x & 0x000000ff) >> 0;
+uint8_t s0_b1 = (src0.x & 0x0000ff00) >> 8;
+uint8_t s0_b2 = (src0.x & 0x00ff0000) >> 16;
+uint8_t s0_b3 = (src0.x & 0xff000000) >> 24;
+
+uint8_t s1_b0 = (src1.x & 0x000000ff) >> 0;
+uint8_t s1_b1 = (src1.x & 0x0000ff00) >> 8;
+uint8_t s1_b2 = (src1.x & 0x00ff0000) >> 16;
+uint8_t s1_b3 = (src1.x & 0xff000000) >> 24;
+
+dst.x = src2.x +
+        (s0_b0 > s1_b0 ? (s0_b0 - s1_b0) : (s1_b0 - s0_b0)) +
+        (s0_b1 > s1_b1 ? (s0_b1 - s1_b1) : (s1_b1 - s0_b1)) +
+        (s0_b2 > s1_b2 ? (s0_b2 - s1_b2) : (s1_b2 - s0_b2)) +
+        (s0_b3 > s1_b3 ? (s0_b3 - s1_b3) : (s1_b3 - s0_b3));
+""")
+
 # Combines the first component of each input to make a 3-component vector.
 
 triop_horiz("vec3", 3, 1, 1, 1, """
diff --git a/src/compiler/nir/nir_range_analysis.c b/src/compiler/nir/nir_range_analysis.c
index 501084f14f5..e18d0446e3e 100644
--- a/src/compiler/nir/nir_range_analysis.c
+++ b/src/compiler/nir/nir_range_analysis.c
@@ -1579,6 +1579,9 @@ nir_unsigned_upper_bound(nir_shader *shader, struct hash_table *range_ht,
       case nir_op_u2u32:
          res = MIN2(src0, max);
          break;
+      case nir_op_sad_u8x4:
+         res = src2 + 4 * 255;
+         break;
       default:
          res = max;
          break;