pan/bi: Match CSEL argument order with hw

It turns out ports need to be in order of the arguments of an
instruction (port 3, that is), which breaks on instructions whose IR
argument order is different from the packed order, like csel. So fix
that.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4458>

diff --git a/src/panfrost/bifrost/bi_pack.c b/src/panfrost/bifrost/bi_pack.c
index 8c91c0436d2cce944b6a8a64c399a32e0131b9b4..ea68aa41b37e713b26000837885691965ac67e15 100644 (file)
--- a/src/panfrost/bifrost/bi_pack.c
+++ b/src/panfrost/bifrost/bi_pack.c
@@ -691,10 +691,10 @@ bi_pack_fma_csel(bi_instruction *ins, struct bi_registers *regs)
 
         unsigned size = nir_alu_type_get_type_size(ins->dest_type);
 
-        unsigned cmp_0 = (flip ? 3 : 0);
-        unsigned cmp_1 = (flip ? 0 : 3);
-        unsigned res_0 = (invert ? 2 : 1);
-        unsigned res_1 = (invert ? 1 : 2);
+        unsigned cmp_0 = (flip ? 1 : 0);
+        unsigned cmp_1 = (flip ? 0 : 1);
+        unsigned res_0 = (invert ? 3 : 2);
+        unsigned res_1 = (invert ? 2 : 3);
         
         struct bifrost_csel4 pack = {
                 .src0 = bi_get_src(ins, regs, cmp_0, true),

diff --git a/src/panfrost/bifrost/bifrost_compile.c b/src/panfrost/bifrost/bifrost_compile.c
index 0eced9713dabb08160fba466ab72c86cd20e6e47..869e353624ed384c0f764a24e69dae088bf252a5 100644 (file)
--- a/src/panfrost/bifrost/bifrost_compile.c
+++ b/src/panfrost/bifrost/bifrost_compile.c
@@ -520,7 +520,7 @@ bi_fuse_csel_cond(bi_instruction *csel, nir_alu_src cond,
         /* We found one, let's fuse it in */
         csel->csel_cond = bcond;
         bi_copy_src(csel, alu, 0, 0, constants_left, constant_shift);
-        bi_copy_src(csel, alu, 1, 3, constants_left, constant_shift);
+        bi_copy_src(csel, alu, 1, 1, constants_left, constant_shift);
 }
 
 static void
@@ -571,8 +571,14 @@ emit_alu(bi_context *ctx, nir_alu_instr *instr)
         unsigned num_inputs = nir_op_infos[instr->op].num_inputs;
         assert(num_inputs <= ARRAY_SIZE(alu.src));
 
-        for (unsigned i = 0; i < num_inputs; ++i)
-                bi_copy_src(&alu, instr, i, i, &constants_left, &constant_shift);
+        for (unsigned i = 0; i < num_inputs; ++i) {
+                unsigned f = 0;
+
+                if (i && alu.type == BI_CSEL)
+                        f++;
+
+                bi_copy_src(&alu, instr, i, i + f, &constants_left, &constant_shift);
+        }
 
         /* Op-specific fixup */
         switch (instr->op) {
@@ -642,8 +648,8 @@ emit_alu(bi_context *ctx, nir_alu_instr *instr)
         if (alu.type == BI_CSEL) {
                 /* Default to csel3 */
                 alu.csel_cond = BI_COND_NE;
-                alu.src[3] = BIR_INDEX_ZERO;
-                alu.src_types[3] = alu.src_types[0];
+                alu.src[1] = BIR_INDEX_ZERO;
+                alu.src_types[1] = alu.src_types[0];
 
                 bi_fuse_csel_cond(&alu, instr->src[0],
                                 &constants_left, &constant_shift);