pan/bi: Fix LD_VAR with non-constant index

src0 and src1 were mixed leading to invalid varying indices. In order to
fix that properly, we first extend load_vary to pass the immediate index
through a dedicated field and add a special boolean. This way, we don't
have to make sure src0 always contains the index, and can instead match
the src numbering defined in ISA.xml.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7636>

src/panfrost/bifrost/bi_pack.c

@@ -713,18 +713,16 @@ bi_pack_add_special(bi_clause *clause, bi_instruction *ins, bi_registers *regs)
 static unsigned
 bi_pack_add_ld_var(bi_clause *clause, bi_instruction *ins, bi_registers *regs)
 {
-        bool imm = ins->src[0] & BIR_INDEX_CONSTANT;
-
         if (ins->load_vary.special)
                 return pan_pack_add_ld_var_special(clause, ins, regs);
 
         if (ins->load_vary.flat) {
-                return imm ?
+                return ins->load_vary.immediate ?
                        pan_pack_add_ld_var_flat_imm(clause, ins, regs) :
                        pan_pack_add_ld_var_flat(clause, ins, regs);
         }
 
-        return imm ?
+        return ins->load_vary.immediate ?
                pan_pack_add_ld_var_imm(clause, ins, regs) :
                pan_pack_add_ld_var(clause, ins, regs);
 }

src/panfrost/bifrost/bi_print.c

@@ -255,6 +255,9 @@ bi_print_load_vary(struct bi_load_vary *load, FILE *fp)
         if (load->flat)
                 fprintf(fp, ".flat");
 
+        if (load->immediate)
+                fprintf(fp, ".imm(%d)", load->index);
+
         switch (load->update_mode) {
         case BIFROST_UPDATE_STORE: fprintf(fp, ".store"); break;
         case BIFROST_UPDATE_RETRIEVE: fprintf(fp, ".retrieve"); break;

src/panfrost/bifrost/bifrost_compile.c

@@ -161,12 +161,19 @@ bi_interp_for_intrinsic(nir_intrinsic_op op)
 static void
 bi_emit_ld_vary(bi_context *ctx, nir_intrinsic_instr *instr)
 {
-        bi_instruction ins = bi_load(BI_LOAD_VAR, instr);
-        ins.load_vary.interp_mode = BIFROST_INTERP_CENTER; /* TODO */
-        ins.load_vary.reuse = false; /* TODO */
-        ins.load_vary.flat = instr->intrinsic != nir_intrinsic_load_interpolated_input;
-        ins.dest_type = nir_dest_bit_size(instr->dest);
-        ins.format = ins.dest_type;
+        bi_instruction ins = {
+                .type = BI_LOAD_VAR,
+                .load_vary = {
+                        .interp_mode = BIFROST_INTERP_CENTER,
+                        .update_mode = BIFROST_UPDATE_STORE,
+                        .reuse = false,
+                        .flat = instr->intrinsic != nir_intrinsic_load_interpolated_input,
+                },
+                .dest = pan_dest_index(&instr->dest),
+                .dest_type = nir_dest_bit_size(instr->dest),
+                .src_types = { nir_type_uint32, nir_type_uint32, nir_type_uint32 },
+                .vector_channels = instr->num_components,
+        };
 
         if (instr->intrinsic == nir_intrinsic_load_interpolated_input) {
                 nir_intrinsic_instr *parent = nir_src_as_intrinsic(instr->src[0]);
@@ -176,12 +183,28 @@ bi_emit_ld_vary(bi_context *ctx, nir_intrinsic_instr *instr)
                 }
         }
 
-        if (nir_src_is_const(*nir_get_io_offset_src(instr))) {
-                /* Zero it out for direct */
-                ins.src[1] = BIR_INDEX_ZERO;
+        if (ins.load_vary.interp_mode == BIFROST_INTERP_CENTER) {
+                /* Zero it out for center interpolation */
+                ins.src[0] = BIR_INDEX_ZERO;
         } else {
                 /* R61 contains sample mask stuff, TODO RA XXX */
-                ins.src[1] = BIR_INDEX_REGISTER | 61;
+                ins.src[0] = BIR_INDEX_REGISTER | 61;
+        }
+
+        nir_src *offset = nir_get_io_offset_src(instr);
+        if (nir_src_is_const(*offset)) {
+                unsigned offset_val = nir_intrinsic_base(instr) +
+                                      nir_src_as_uint(*offset);
+
+                if (offset_val < 20) {
+                        ins.load_vary.immediate = true;
+                        ins.load_vary.index = offset_val;
+                } else {
+                        ins.src[1] = BIR_INDEX_CONSTANT | 0;
+                        ins.constant.u64 = offset_val;
+                }
+        } else {
+                ins.src[1] = pan_src_index(offset);
         }
 
         bi_emit(ctx, ins);

src/panfrost/bifrost/compiler.h

@@ -136,6 +136,8 @@ struct bi_load_vary {
         enum bifrost_interp_mode interp_mode;
         enum bifrost_update_mode update_mode;
         enum bifrost_special_var_id var_id;
+        unsigned index;
+        bool immediate;
         bool special;
         bool reuse;
         bool flat;

src/panfrost/bifrost/gen_pack.py

@@ -393,7 +393,7 @@ def pack_derived(pos, exprs, imm_map, body, pack_exprs):
 IMMEDIATE_TABLE = {
         'attribute_index': 'bi_get_immediate(ins, 0)',
         'varying_index': 'bi_get_immediate(ins, 0)',
-        'index': 'bi_get_immediate(ins, 0)',
+        'index': 'ins->load_vary.index',
         'texture_index': 'ins->texture.texture_index',
         'sampler_index': 'ins->texture.sampler_index',
         'table': '63', # Bindless (flat addressing) mode for DTSEL_IMM
@@ -452,7 +452,7 @@ def pack_variant(opname, states):
     if staging in ["r", "rw"]:
         offset += 1
 
-    offset += len(set(["attribute_index", "varying_index", "index"]) & set([x[0] for x in states[0][1].get("immediates", [])]))
+    offset += len(set(["attribute_index", "varying_index"]) & set([x[0] for x in states[0][1].get("immediates", [])]))
 
     pack_sources(states[0][1].get("srcs", []), common_body, pack_exprs, offset)