diff --git a/src/panfrost/bifrost/bi_builder.h.py b/src/panfrost/bifrost/bi_builder.h.py
index 5fbb48f180a..4791a1a30aa 100644
--- a/src/panfrost/bifrost/bi_builder.h.py
+++ b/src/panfrost/bifrost/bi_builder.h.py
@@ -40,15 +40,21 @@ def typesize(opcode):
             return int(opcode[-2:])
         except:
             return None
+
+def to_suffix(op):
+    return "_to" if op["dests"] > 0 else ""
+
 %>
 
 % for opcode in ops:
 static inline
-bi_instr * bi_${opcode.replace('.', '_').lower()}_to(${signature(ops[opcode], 1, modifiers)})
+bi_instr * bi_${opcode.replace('.', '_').lower()}${to_suffix(ops[opcode])}(${signature(ops[opcode], modifiers)})
 {
     bi_instr *I = rzalloc(b->shader, bi_instr);
     I->op = BI_OPCODE_${opcode.replace('.', '_').upper()};
-    I->dest[0] = dest0;
+% for dest in range(ops[opcode]["dests"]):
+    I->dest[${dest}] = dest${dest};
+% endfor
 % for src in range(src_count(ops[opcode])):
     I->src[${src}] = src${src};
 % endfor
@@ -64,31 +70,31 @@ bi_instr * bi_${opcode.replace('.', '_').lower()}_to(${signature(ops[opcode], 1,
     return I;
 }
 
-% if opcode.split(".")[0] not in ["JUMP", "BRANCHZ", "BRANCH"]:
+% if ops[opcode]["dests"] == 1:
 static inline
-bi_index bi_${opcode.replace('.', '_').lower()}(${signature(ops[opcode], 0, modifiers)})
+bi_index bi_${opcode.replace('.', '_').lower()}(${signature(ops[opcode], modifiers, no_dests=True)})
 {
-    return (bi_${opcode.replace('.', '_').lower()}_to(${arguments(ops[opcode], 1)}))->dest[0];
+    return (bi_${opcode.replace('.', '_').lower()}_to(${arguments(ops[opcode])}))->dest[0];
 }
 
 %endif
 <%
     common_op = opcode.split('.')[0]
     variants = [a for a in ops.keys() if a.split('.')[0] == common_op]
-    signatures = [signature(ops[op], 0, modifiers, sized=True) for op in variants]
+    signatures = [signature(ops[op], modifiers, sized=True, no_dests=True) for op in variants]
     homogenous = all([sig == signatures[0] for sig in signatures])
     sizes = [typesize(x) for x in variants]
     last = opcode == variants[-1]
 %>
 % if homogenous and len(variants) > 1 and last:
 % for (suffix, temp, dests, ret) in (('_to', False, 1, 'instr *'), ('', True, 0, 'index')):
-% if not temp or common_op not in ["JUMP", "BRANCHZ", "BRANCH"]:
+% if not temp or ops[opcode]["dests"] > 0:
 static inline
-bi_${ret} bi_${common_op.replace('.', '_').lower()}${suffix}(${signature(ops[opcode], dests, modifiers, sized=True)})
+bi_${ret} bi_${common_op.replace('.', '_').lower()}${suffix if ops[opcode]['dests'] > 0 else ''}(${signature(ops[opcode], modifiers, sized=True, no_dests=not dests)})
 {
 % for i, (variant, size) in enumerate(zip(variants, sizes)):
     ${"else " if i > 0 else ""} if (bitsize == ${size})
-        return (bi_${variant.replace('.', '_').lower()}_to(${arguments(ops[opcode], 1, temp_dest = temp)}))${"->dest[0]" if temp else ""};
+        return (bi_${variant.replace('.', '_').lower()}${to_suffix(ops[opcode])}(${arguments(ops[opcode], temp_dest = temp)}))${"->dest[0]" if temp else ""};
 % endfor
     else
         unreachable("Invalid bitsize for ${common_op}");
@@ -116,11 +122,11 @@ def should_skip(mod):
 def modifier_signature(op):
     return sorted([m for m in op["modifiers"].keys() if not should_skip(m)])
 
-def signature(op, dest_count, modifiers, sized = False):
+def signature(op, modifiers, sized = False, no_dests = False):
     return ", ".join(
         ["bi_builder *b"] +
         (["unsigned bitsize"] if sized else []) +
-        ["bi_index dest{}".format(i) for i in range(dest_count)] +
+        ["bi_index dest{}".format(i) for i in range(0 if no_dests else op["dests"])] +
         ["bi_index src{}".format(i) for i in range(src_count(op))] +
         ["{} {}".format(
         "bool" if len(modifiers[T[0:-1]] if T[-1] in "0123" else modifiers[T]) == 2 else
@@ -129,10 +135,10 @@ def signature(op, dest_count, modifiers, sized = False):
         T) for T in modifier_signature(op)] +
         ["uint32_t {}".format(imm) for imm in op["immediates"]])
 
-def arguments(op, dest_count, temp_dest = True):
+def arguments(op, temp_dest = True):
     return ", ".join(
         ["b"] +
-        ["bi_temp(b->shader)" if temp_dest else 'dest{}'.format(i) for i in range(dest_count)] +
+        ["bi_temp(b->shader)" if temp_dest else 'dest{}'.format(i) for i in range(op["dests"])] +
         ["src{}".format(i) for i in range(src_count(op))] +
         modifier_signature(op) +
         op["immediates"])
diff --git a/src/panfrost/bifrost/bi_ra.c b/src/panfrost/bifrost/bi_ra.c
index 742bd1de3a1..2843b04d219 100644
--- a/src/panfrost/bifrost/bi_ra.c
+++ b/src/panfrost/bifrost/bi_ra.c
@@ -256,7 +256,7 @@ bi_spill_dest(bi_builder *b, bi_index index, bi_index temp, uint32_t offset,
         b->cursor = bi_after_clause(clause);
 
         /* setup FAU as [offset][0] */
-        bi_instr *st = bi_store_to(b, channels * 32, bi_null(), temp,
+        bi_instr *st = bi_store(b, channels * 32, temp,
                         bi_passthrough(BIFROST_SRC_FAU_LO),
                         bi_passthrough(BIFROST_SRC_FAU_HI),
                         BI_SEG_TL);
diff --git a/src/panfrost/bifrost/bi_schedule.c b/src/panfrost/bifrost/bi_schedule.c
index 3bb7b87dcb6..4104d00c47c 100644
--- a/src/panfrost/bifrost/bi_schedule.c
+++ b/src/panfrost/bifrost/bi_schedule.c
@@ -182,7 +182,7 @@ bi_lower_atom_c(bi_context *ctx, struct bi_clause_state *clause, struct
 {
         bi_instr *pinstr = tuple->add;
         bi_builder b = bi_init_builder(ctx, bi_before_instr(pinstr));
-        bi_instr *atom_c = bi_atom_c_return_i32_to(&b, bi_null(),
+        bi_instr *atom_c = bi_atom_c_return_i32(&b, 
                         pinstr->src[1], pinstr->src[2], pinstr->src[0],
                         pinstr->atom_opc);
 
@@ -201,7 +201,7 @@ bi_lower_atom_c1(bi_context *ctx, struct bi_clause_state *clause, struct
 {
         bi_instr *pinstr = tuple->add;
         bi_builder b = bi_init_builder(ctx, bi_before_instr(pinstr));
-        bi_instr *atom_c = bi_atom_c1_return_i32_to(&b, bi_null(),
+        bi_instr *atom_c = bi_atom_c1_return_i32(&b,
                         pinstr->src[0], pinstr->src[1], pinstr->atom_opc);
 
         if (bi_is_null(pinstr->dest[0]))
diff --git a/src/panfrost/bifrost/bifrost_compile.c b/src/panfrost/bifrost/bifrost_compile.c
index c7fde1ce20f..a15dd327a0f 100644
--- a/src/panfrost/bifrost/bifrost_compile.c
+++ b/src/panfrost/bifrost/bifrost_compile.c
@@ -65,7 +65,7 @@ static bi_block *emit_cf_list(bi_context *ctx, struct exec_list *list);
 static void
 bi_emit_jump(bi_builder *b, nir_jump_instr *instr)
 {
-        bi_instr *branch = bi_jump_to(b, bi_null(), bi_zero());
+        bi_instr *branch = bi_jump(b, bi_zero());
 
         switch (instr->type) {
         case nir_jump_break:
@@ -528,7 +528,7 @@ bi_emit_fragment_out(bi_builder *b, nir_intrinsic_instr *instr)
                 /* Jump back to the fragment shader, return address is stored
                  * in r48 (see above).
                  */
-                bi_jump_to(b, bi_null(), bi_register(48));
+                bi_jump(b, bi_register(48));
         }
 }
 
@@ -568,8 +568,8 @@ bi_emit_store_vary(bi_builder *b, nir_intrinsic_instr *instr)
         unsigned nr = util_last_bit(nir_intrinsic_write_mask(instr));
         assert(nr > 0 && nr <= nir_intrinsic_src_components(instr, 0));
 
-        bi_st_cvt_to(b, bi_null(), bi_src_index(&instr->src[0]),
-                        address, bi_word(address, 1), bi_word(address, 2),
+        bi_st_cvt(b, bi_src_index(&instr->src[0]), address,
+                        bi_word(address, 1), bi_word(address, 2),
                         regfmt, nr - 1);
 }
 
@@ -609,8 +609,7 @@ bi_emit_load(bi_builder *b, nir_intrinsic_instr *instr, enum bi_seg seg)
 static void
 bi_emit_store(bi_builder *b, nir_intrinsic_instr *instr, enum bi_seg seg)
 {
-        bi_store_to(b, instr->num_components * nir_src_bit_size(instr->src[0]),
-                    bi_null(),
+        bi_store(b, instr->num_components * nir_src_bit_size(instr->src[0]),
                     bi_src_index(&instr->src[0]),
                     bi_src_index(&instr->src[1]), bi_addr_high(&instr->src[1]),
                     seg);
@@ -831,8 +830,7 @@ bi_emit_image_store(bi_builder *b, nir_intrinsic_instr *instr)
 {
         bi_index addr = bi_emit_lea_image(b, instr);
 
-        bi_st_cvt_to(b, bi_null(),
-                     bi_src_index(&instr->src[3]),
+        bi_st_cvt(b, bi_src_index(&instr->src[3]),
                      addr, bi_word(addr, 1), bi_word(addr, 2),
                      bi_reg_fmt_for_nir(nir_intrinsic_src_type(instr)),
                      instr->num_components - 1);
@@ -1014,7 +1012,7 @@ bi_emit_intrinsic(bi_builder *b, nir_intrinsic_instr *instr)
 
         case nir_intrinsic_control_barrier:
                 assert(b->shader->stage != MESA_SHADER_FRAGMENT);
-                bi_barrier_to(b, bi_null());
+                bi_barrier(b);
                 break;
 
         case nir_intrinsic_shared_atomic_add:
@@ -1122,13 +1120,12 @@ bi_emit_intrinsic(bi_builder *b, nir_intrinsic_instr *instr)
                 if (sz == 16)
                         src = bi_half(src, false);
 
-                bi_discard_f32_to(b, bi_null(), src, bi_zero(), BI_CMPF_NE);
+                bi_discard_f32(b, src, bi_zero(), BI_CMPF_NE);
                 break;
         }
 
         case nir_intrinsic_discard:
-                bi_discard_f32_to(b, bi_null(), bi_zero(), bi_zero(),
-                                BI_CMPF_EQ);
+                bi_discard_f32(b, bi_zero(), bi_zero(), BI_CMPF_EQ);
                 break;
 
         case nir_intrinsic_load_ssbo_address:
@@ -2071,9 +2068,9 @@ bi_emit_cube_coord(bi_builder *b, bi_index coord,
                     bi_index *face, bi_index *s, bi_index *t)
 {
         /* Compute max { |x|, |y|, |z| } */
-        bi_instr *cubeface = bi_cubeface_to(b, bi_temp(b->shader), coord,
+        bi_instr *cubeface = bi_cubeface_to(b, bi_temp(b->shader),
+                        bi_temp(b->shader), coord,
                         bi_word(coord, 1), bi_word(coord, 2));
-        cubeface->dest[1] = bi_temp(b->shader);
 
         /* Select coordinates */
 
@@ -2553,24 +2550,6 @@ emit_block(bi_context *ctx, nir_block *block)
         return ctx->current_block;
 }
 
-/* Emits a direct branch based on a given condition. TODO: try to unwrap the
- * condition to optimize */
-
-static bi_instr *
-bi_branch(bi_builder *b, nir_src *condition, bool invert)
-{
-        return bi_branchz_i32_to(b, bi_null(), bi_src_index(condition),
-                        bi_zero(), invert ? BI_CMPF_EQ : BI_CMPF_NE);
-}
-
-static bi_instr *
-bi_jump(bi_builder *b, bi_block *target)
-{
-        bi_instr *I = bi_jump_to(b, bi_null(), bi_zero());
-        I->branch_target = target;
-        return I;
-}
-
 static void
 emit_if(bi_context *ctx, nir_if *nif)
 {
@@ -2578,7 +2557,8 @@ emit_if(bi_context *ctx, nir_if *nif)
 
         /* Speculatively emit the branch, but we can't fill it in until later */
         bi_builder _b = bi_init_builder(ctx, bi_after_block(ctx->current_block));
-        bi_instr *then_branch = bi_branch(&_b, &nif->condition, true);
+        bi_instr *then_branch = bi_branchz_i32(&_b,
+                        bi_src_index(&nif->condition), bi_zero(), BI_CMPF_EQ);
 
         /* Emit the two subblocks. */
         bi_block *then_block = emit_cf_list(ctx, &nif->then_list);
@@ -2604,7 +2584,8 @@ emit_if(bi_context *ctx, nir_if *nif)
 
                 /* Emit a jump from the end of the then block to the end of the else */
                 _b.cursor = bi_after_block(end_then_block);
-                bi_instr *then_exit = bi_jump(&_b, ctx->after_block);
+                bi_instr *then_exit = bi_jump(&_b, bi_zero());
+                then_exit->branch_target = ctx->after_block;
 
                 pan_block_add_successor(&end_then_block->base, &then_exit->branch_target->base);
                 pan_block_add_successor(&end_else_block->base, &ctx->after_block->base); /* fallthrough */
@@ -2632,7 +2613,8 @@ emit_loop(bi_context *ctx, nir_loop *nloop)
 
         /* Branch back to loop back */
         bi_builder _b = bi_init_builder(ctx, bi_after_block(ctx->current_block));
-        bi_jump(&_b, ctx->continue_block);
+        bi_instr *I = bi_jump(&_b, bi_zero());
+        I->branch_target = ctx->continue_block;
         pan_block_add_successor(&start_block->base, &ctx->continue_block->base);
         pan_block_add_successor(&ctx->current_block->base, &ctx->continue_block->base);
 
diff --git a/src/panfrost/bifrost/bifrost_isa.py b/src/panfrost/bifrost/bifrost_isa.py
index a2d7f882e39..ae97795f3ae 100644
--- a/src/panfrost/bifrost/bifrost_isa.py
+++ b/src/panfrost/bifrost/bifrost_isa.py
@@ -131,6 +131,7 @@ def parse_instruction(ins, include_pseudo):
             'derived': [],
             'staging': ins.attrib.get('staging', '').split('=')[0],
             'staging_count': ins.attrib.get('staging', '=0').split('=')[1],
+            'dests': int(ins.attrib.get('dests', '1')),
             'unused': ins.attrib.get('unused', False),
             'pseudo': ins.attrib.get('pseudo', False),
             'message': ins.attrib.get('message', 'none'),
@@ -238,6 +239,7 @@ def simplify_to_ir(ins):
     return {
             'staging': ins['staging'],
             'srcs': len(ins['srcs']),
+            'dests': ins['dests'],
             'modifiers': [[m[0][0], m[2]] for m in ins['modifiers']],
             'immediates': [m[0] for m in ins['immediates']]
         }
@@ -252,6 +254,7 @@ def combine_ir_variants(instructions, v):
     for s in variants:
         # Check consistency
         assert(s['srcs'] == variants[0]['srcs'])
+        assert(s['dests'] == variants[0]['dests'])
         assert(s['immediates'] == variants[0]['immediates'])
         assert(s['staging'] == variants[0]['staging'])
 
@@ -265,6 +268,7 @@ def combine_ir_variants(instructions, v):
     # modifiers
     return {
             'srcs': variants[0]['srcs'],
+            'dests': variants[0]['dests'],
             'staging': variants[0]['staging'],
             'immediates': sorted(variants[0]['immediates']),
             'modifiers': { k: modifiers[k] for k in modifiers }