aco: fix imod/omod for gfx11 VOP3 opcodes

Fixes: d8d99c3c4f2 ("aco: add GFX11 opcode numbers")
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21170>
(cherry picked from commit c8adf16278fc02207602e4fafa12b95e6ea8dda0)

diff --git a/.pick_status.json b/.pick_status.json
index c9dc699..32a0931 100644 (file)
--- a/.pick_status.json
+++ b/.pick_status.json
@@ -112,7 +112,7 @@
         "description": "aco: fix imod/omod for gfx11 VOP3 opcodes",
         "nominated": true,
         "nomination_type": 1,
-        "resolution": 0,
+        "resolution": 1,
         "main_sha": null,
         "because_sha": "d8d99c3c4f2355d9546f1998da75dd2ab69e67e2"
     },

diff --git a/src/amd/compiler/aco_opcodes.py b/src/amd/compiler/aco_opcodes.py
index 2c11cf2..0fcbb3c 100644 (file)
--- a/src/amd/compiler/aco_opcodes.py
+++ b/src/amd/compiler/aco_opcodes.py
@@ -1173,18 +1173,18 @@ VOP3 = {
    (   -1,    -1,    -1,    -1,    -1, 0x25f, "v_minmax_f32", True, True),
    (   -1,    -1,    -1,    -1,    -1, 0x260, "v_maxmin_f16", True, True),
    (   -1,    -1,    -1,    -1,    -1, 0x261, "v_minmax_f16", True, True),
-   (   -1,    -1,    -1,    -1,    -1, 0x262, "v_maxmin_u32", True, True),
-   (   -1,    -1,    -1,    -1,    -1, 0x263, "v_minmax_u32", True, True),
-   (   -1,    -1,    -1,    -1,    -1, 0x264, "v_maxmin_i32", True, True),
-   (   -1,    -1,    -1,    -1,    -1, 0x265, "v_minmax_i32", True, True),
-   (   -1,    -1,    -1,    -1,    -1, 0x266, "v_dot2_f16_f16", True, True),
-   (   -1,    -1,    -1,    -1,    -1, 0x267, "v_dot2_bf16_bf16", True, True),
-   (   -1,    -1,    -1,    -1,    -1, 0x306, "v_cvt_pk_i16_f32", True, True),
-   (   -1,    -1,    -1,    -1,    -1, 0x307, "v_cvt_pk_u16_f32", True, True),
-   (   -1,    -1,    -1,    -1,    -1, 0x362, "v_and_b16", True, True),
-   (   -1,    -1,    -1,    -1,    -1, 0x363, "v_or_b16", True, True),
-   (   -1,    -1,    -1,    -1,    -1, 0x364, "v_xor_b16", True, True),
-   (   -1,    -1,    -1,    -1,    -1, 0x25d, "v_cndmask_b16", True, True),
+   (   -1,    -1,    -1,    -1,    -1, 0x262, "v_maxmin_u32", False, False),
+   (   -1,    -1,    -1,    -1,    -1, 0x263, "v_minmax_u32", False, False),
+   (   -1,    -1,    -1,    -1,    -1, 0x264, "v_maxmin_i32", False, False),
+   (   -1,    -1,    -1,    -1,    -1, 0x265, "v_minmax_i32", False, False),
+   (   -1,    -1,    -1,    -1,    -1, 0x266, "v_dot2_f16_f16", False, False),
+   (   -1,    -1,    -1,    -1,    -1, 0x267, "v_dot2_bf16_bf16", False, False),
+   (   -1,    -1,    -1,    -1,    -1, 0x306, "v_cvt_pk_i16_f32", True, False),
+   (   -1,    -1,    -1,    -1,    -1, 0x307, "v_cvt_pk_u16_f32", True, False),
+   (   -1,    -1,    -1,    -1,    -1, 0x362, "v_and_b16", False, False),
+   (   -1,    -1,    -1,    -1,    -1, 0x363, "v_or_b16", False, False),
+   (   -1,    -1,    -1,    -1,    -1, 0x364, "v_xor_b16", False, False),
+   (   -1,    -1,    -1,    -1,    -1, 0x25d, "v_cndmask_b16", True, False),
 }
 for (gfx6, gfx7, gfx8, gfx9, gfx10, gfx11, name, in_mod, out_mod, cls) in default_class(VOP3, InstrClass.Valu32):
    opcode(name, gfx7, gfx9, gfx10, gfx11, Format.VOP3, cls, in_mod, out_mod)