pan/va: Replace MKVEC.v4i8 with MKVEC.v2i8

This is the instruction that the hardware actually supports. Do the rename, use
the more specific accurate model in the IR, and rework the Valhall texturing
code to emit MKVEC.v2i8 instead of MKVEC.v4i8.

Will fix:

   dEQP-GLES31.functional.texture.gather.offset_dynamic.implementation_offset.*

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17101>

diff --git a/src/panfrost/bifrost/bifrost_compile.c b/src/panfrost/bifrost/bifrost_compile.c
index 15f5924..6328c80 100644 (file)
--- a/src/panfrost/bifrost/bifrost_compile.c
+++ b/src/panfrost/bifrost/bifrost_compile.c
@@ -3198,11 +3198,16 @@ bi_emit_valhall_offsets(bi_builder *b, nir_tex_instr *instr)
                 /* No multisample index with 3D */
                 assert((nr <= 2) || (ms_idx < 0));
 
-                dest = bi_mkvec_v4i8(b,
+                /* Zero extend the Z byte so we can use it with MKVEC.v2i8 */
+                bi_index z = (nr > 2) ?
+                             bi_mkvec_v2i8(b, bi_byte(bi_extract(b, idx, 2), 0),
+                                              bi_imm_u8(0), bi_zero()) :
+                             bi_zero();
+
+                dest = bi_mkvec_v2i8(b,
                                 (nr > 0) ? bi_byte(bi_extract(b, idx, 0), 0) : bi_imm_u8(0),
                                 (nr > 1) ? bi_byte(bi_extract(b, idx, 1), 0) : bi_imm_u8(0),
-                                (nr > 2) ? bi_byte(bi_extract(b, idx, 2), 0) : bi_imm_u8(0),
-                                bi_imm_u8(0));
+                                z);
         }
 
         /* Component 2: multisample index */

diff --git a/src/panfrost/bifrost/valhall/ISA.xml b/src/panfrost/bifrost/valhall/ISA.xml
index a5e35f9..495bac4 100644 (file)
--- a/src/panfrost/bifrost/valhall/ISA.xml
+++ b/src/panfrost/bifrost/valhall/ISA.xml
@@ -2110,15 +2110,15 @@
     <src>Cube face index</src>
   </ins>
 
-  <ins name="MKVEC.v4i8" title="Make 8-bit vector" dests="1" opcode="0xBD" unit="CVT">
+  <ins name="MKVEC.v2i8" title="Make 8-bit vector" dests="1" opcode="0xBD" unit="CVT">
     <desc>
       Calculates $A | (B \ll 8) | (CD \ll 16)$ for 8-bit A and B and 16-bit CD.
 
       To implement `(uchar4) (A, B, C, D)` in full generality, use the sequence
-      `MKVEC.v4i8 CD, C, D, #0; MKVEC.v4i8 out, A, B, CD`
+      `MKVEC.v2i8 CD, C, D, #0; MKVEC.v2i8 out, A, B, CD`
 
-      `MKVEC.v4i8` also allows zero extending arbitrary 8-bit lanes. For
-      example, to extend `r0.b3` to `r1`, use `MKVEC.v4i8 r1, r0.b3, 0x0.b0, 0x0`.
+      `MKVEC.v2i8` also allows zero extending arbitrary 8-bit lanes. For
+      example, to extend `r0.b3` to `r1`, use `MKVEC.v2i8 r1, r0.b3, 0x0.b0, 0x0`.
     </desc>
     <src lane="true">A</src>
     <src lane="true">B</src>

diff --git a/src/panfrost/bifrost/valhall/test/assembler-cases.txt b/src/panfrost/bifrost/valhall/test/assembler-cases.txt
index 44d21a1..59a4d7c 100644 (file)
--- a/src/panfrost/bifrost/valhall/test/assembler-cases.txt
+++ b/src/panfrost/bifrost/valhall/test/assembler-cases.txt
@@ -152,10 +152,10 @@ c0 77 05 04 00 c2 a1 00    MKVEC.v2i16 r2, 0x0.h00, ^r55.h10
 77 00 04 10 00 c2 90 00    S16_TO_S32 r2, ^r55.h10
 c0 77 01 08 00 c2 a8 00    ISUB.s32 r2, 0x0, ^r55.h0
 c0 77 01 0c 00 c2 a8 00    ISUB.s32 r2, 0x0, ^r55.h1
-00 c0 c0 00 c0 c7 bd 00    MKVEC.v4i8 r7, r0.b3, 0x0.b0, 0x0
-00 c0 c0 00 80 c6 bd 00    MKVEC.v4i8 r6, r0.b2, 0x0.b0, 0x0
-00 c0 c0 00 00 c4 bd 00    MKVEC.v4i8 r4, r0.b0, 0x0.b0, 0x0
-40 c0 c0 00 40 c5 bd 00    MKVEC.v4i8 r5, ^r0.b1, 0x0.b0, 0x0
+00 c0 c0 00 c0 c7 bd 00    MKVEC.v2i8 r7, r0.b3, 0x0.b0, 0x0
+00 c0 c0 00 80 c6 bd 00    MKVEC.v2i8 r6, r0.b2, 0x0.b0, 0x0
+00 c0 c0 00 00 c4 bd 00    MKVEC.v2i8 r4, r0.b0, 0x0.b0, 0x0
+40 c0 c0 00 40 c5 bd 00    MKVEC.v2i8 r5, ^r0.b1, 0x0.b0, 0x0
 00 00 11 30 00 c7 90 00    U8_TO_F32 r7, r0.b3
 00 00 11 20 00 c6 90 00    U8_TO_F32 r6, r0.b2
 00 00 11 00 00 c4 90 00    U8_TO_F32 r4, r0.b0