+ return nir_swizzle(b, def, &c, 1);
+}
+
+static inline nir_ssa_def *
+nir_channels(nir_builder *b, nir_ssa_def *def, nir_component_mask_t mask)
+{
+ unsigned num_channels = 0, swizzle[NIR_MAX_VEC_COMPONENTS] = { 0 };
+
+ for (unsigned i = 0; i < NIR_MAX_VEC_COMPONENTS; i++) {
+ if ((mask & (1 << i)) == 0)
+ continue;
+ swizzle[num_channels++] = i;
+ }
+
+ return nir_swizzle(b, def, swizzle, num_channels);
+}
+
+static inline nir_ssa_def *
+_nir_vector_extract_helper(nir_builder *b, nir_ssa_def *vec, nir_ssa_def *c,
+ unsigned start, unsigned end)
+{
+ if (start == end - 1) {
+ return nir_channel(b, vec, start);
+ } else {
+ unsigned mid = start + (end - start) / 2;
+ return nir_bcsel(b, nir_ilt(b, c, nir_imm_int(b, mid)),
+ _nir_vector_extract_helper(b, vec, c, start, mid),
+ _nir_vector_extract_helper(b, vec, c, mid, end));
+ }
+}
+
+static inline nir_ssa_def *
+nir_vector_extract(nir_builder *b, nir_ssa_def *vec, nir_ssa_def *c)
+{
+ nir_src c_src = nir_src_for_ssa(c);
+ if (nir_src_is_const(c_src)) {
+ unsigned c_const = nir_src_as_uint(c_src);
+ if (c_const < vec->num_components)
+ return nir_channel(b, vec, c_const);
+ else
+ return nir_ssa_undef(b, 1, vec->bit_size);
+ } else {
+ return _nir_vector_extract_helper(b, vec, c, 0, vec->num_components);
+ }
+}
+
+static inline nir_ssa_def *
+nir_i2i(nir_builder *build, nir_ssa_def *x, unsigned dest_bit_size)
+{
+ if (x->bit_size == dest_bit_size)
+ return x;
+
+ switch (dest_bit_size) {
+ case 64: return nir_i2i64(build, x);
+ case 32: return nir_i2i32(build, x);
+ case 16: return nir_i2i16(build, x);
+ case 8: return nir_i2i8(build, x);
+ default: unreachable("Invalid bit size");
+ }
+}
+
+static inline nir_ssa_def *
+nir_u2u(nir_builder *build, nir_ssa_def *x, unsigned dest_bit_size)
+{
+ if (x->bit_size == dest_bit_size)
+ return x;
+
+ switch (dest_bit_size) {
+ case 64: return nir_u2u64(build, x);
+ case 32: return nir_u2u32(build, x);
+ case 16: return nir_u2u16(build, x);
+ case 8: return nir_u2u8(build, x);
+ default: unreachable("Invalid bit size");
+ }
+}
+
+static inline nir_ssa_def *
+nir_iadd_imm(nir_builder *build, nir_ssa_def *x, uint64_t y)
+{
+ assert(x->bit_size <= 64);
+ if (x->bit_size < 64)
+ y &= (1ull << x->bit_size) - 1;
+
+ if (y == 0) {
+ return x;
+ } else {
+ return nir_iadd(build, x, nir_imm_intN_t(build, y, x->bit_size));
+ }
+}
+
+static inline nir_ssa_def *
+nir_imul_imm(nir_builder *build, nir_ssa_def *x, uint64_t y)
+{
+ assert(x->bit_size <= 64);
+ if (x->bit_size < 64)
+ y &= (1ull << x->bit_size) - 1;
+
+ if (y == 0) {
+ return nir_imm_intN_t(build, 0, x->bit_size);
+ } else if (y == 1) {
+ return x;
+ } else if (util_is_power_of_two_or_zero64(y)) {
+ return nir_ishl(build, x, nir_imm_int(build, ffsll(y) - 1));
+ } else {
+ return nir_imul(build, x, nir_imm_intN_t(build, y, x->bit_size));
+ }
+}
+
+static inline nir_ssa_def *
+nir_fadd_imm(nir_builder *build, nir_ssa_def *x, double y)
+{
+ return nir_fadd(build, x, nir_imm_floatN_t(build, y, x->bit_size));
+}
+
+static inline nir_ssa_def *
+nir_fmul_imm(nir_builder *build, nir_ssa_def *x, double y)
+{
+ return nir_fmul(build, x, nir_imm_floatN_t(build, y, x->bit_size));
+}
+
+static inline nir_ssa_def *
+nir_pack_bits(nir_builder *b, nir_ssa_def *src, unsigned dest_bit_size)
+{
+ assert(src->num_components * src->bit_size == dest_bit_size);
+
+ switch (dest_bit_size) {
+ case 64:
+ switch (src->bit_size) {
+ case 32: return nir_pack_64_2x32(b, src);
+ case 16: return nir_pack_64_4x16(b, src);
+ default: break;
+ }
+ break;
+
+ case 32:
+ if (src->bit_size == 16)
+ return nir_pack_32_2x16(b, src);
+ break;
+
+ default:
+ break;
+ }
+
+ /* If we got here, we have no dedicated unpack opcode. */
+ nir_ssa_def *dest = nir_imm_intN_t(b, 0, dest_bit_size);
+ for (unsigned i = 0; i < src->num_components; i++) {
+ nir_ssa_def *val = nir_u2u(b, nir_channel(b, src, i), dest_bit_size);
+ val = nir_ishl(b, val, nir_imm_int(b, i * src->bit_size));
+ dest = nir_ior(b, dest, val);
+ }
+ return dest;
+}
+
+static inline nir_ssa_def *
+nir_unpack_bits(nir_builder *b, nir_ssa_def *src, unsigned dest_bit_size)
+{
+ assert(src->num_components == 1);
+ assert(src->bit_size > dest_bit_size);
+ const unsigned dest_num_components = src->bit_size / dest_bit_size;
+ assert(dest_num_components <= NIR_MAX_VEC_COMPONENTS);
+
+ switch (src->bit_size) {
+ case 64:
+ switch (dest_bit_size) {
+ case 32: return nir_unpack_64_2x32(b, src);
+ case 16: return nir_unpack_64_4x16(b, src);
+ default: break;
+ }
+ break;
+
+ case 32:
+ if (dest_bit_size == 16)
+ return nir_unpack_32_2x16(b, src);
+ break;
+
+ default:
+ break;
+ }
+
+ /* If we got here, we have no dedicated unpack opcode. */
+ nir_ssa_def *dest_comps[NIR_MAX_VEC_COMPONENTS];
+ for (unsigned i = 0; i < dest_num_components; i++) {
+ nir_ssa_def *val = nir_ushr(b, src, nir_imm_int(b, i * dest_bit_size));
+ dest_comps[i] = nir_u2u(b, val, dest_bit_size);
+ }
+ return nir_vec(b, dest_comps, dest_num_components);
+}
+
+static inline nir_ssa_def *
+nir_bitcast_vector(nir_builder *b, nir_ssa_def *src, unsigned dest_bit_size)
+{
+ assert((src->bit_size * src->num_components) % dest_bit_size == 0);
+ const unsigned dest_num_components =
+ (src->bit_size * src->num_components) / dest_bit_size;
+ assert(dest_num_components <= NIR_MAX_VEC_COMPONENTS);
+
+ if (src->bit_size > dest_bit_size) {
+ assert(src->bit_size % dest_bit_size == 0);
+ if (src->num_components == 1) {
+ return nir_unpack_bits(b, src, dest_bit_size);
+ } else {
+ const unsigned divisor = src->bit_size / dest_bit_size;
+ assert(src->num_components * divisor == dest_num_components);
+ nir_ssa_def *dest[NIR_MAX_VEC_COMPONENTS];
+ for (unsigned i = 0; i < src->num_components; i++) {
+ nir_ssa_def *unpacked =
+ nir_unpack_bits(b, nir_channel(b, src, i), dest_bit_size);
+ assert(unpacked->num_components == divisor);
+ for (unsigned j = 0; j < divisor; j++)
+ dest[i * divisor + j] = nir_channel(b, unpacked, j);
+ }
+ return nir_vec(b, dest, dest_num_components);
+ }
+ } else if (src->bit_size < dest_bit_size) {
+ assert(dest_bit_size % src->bit_size == 0);
+ if (dest_num_components == 1) {
+ return nir_pack_bits(b, src, dest_bit_size);
+ } else {
+ const unsigned divisor = dest_bit_size / src->bit_size;
+ assert(src->num_components == dest_num_components * divisor);
+ nir_ssa_def *dest[NIR_MAX_VEC_COMPONENTS];
+ for (unsigned i = 0; i < dest_num_components; i++) {
+ nir_component_mask_t src_mask =
+ ((1 << divisor) - 1) << (i * divisor);
+ dest[i] = nir_pack_bits(b, nir_channels(b, src, src_mask),
+ dest_bit_size);
+ }
+ return nir_vec(b, dest, dest_num_components);
+ }
+ } else {
+ assert(src->bit_size == dest_bit_size);
+ return src;
+ }