static inline nir_ssa_def *
nir_build_imm(nir_builder *build, unsigned num_components,
- unsigned bit_size, nir_const_value value)
+ unsigned bit_size, const nir_const_value *value)
{
nir_load_const_instr *load_const =
nir_load_const_instr_create(build->shader, num_components, bit_size);
if (!load_const)
return NULL;
- load_const->value = value;
+ memcpy(load_const->value, value, sizeof(nir_const_value) * num_components);
nir_builder_instr_insert(build, &load_const->instr);
}
static inline nir_ssa_def *
-nir_imm_bool(nir_builder *build, bool x)
+nir_imm_zero(nir_builder *build, unsigned num_components, unsigned bit_size)
{
- nir_const_value v;
+ nir_load_const_instr *load_const =
+ nir_load_const_instr_create(build->shader, num_components, bit_size);
+
+ /* nir_load_const_instr_create uses rzalloc so it's already zero */
- memset(&v, 0, sizeof(v));
- v.b[0] = x;
+ nir_builder_instr_insert(build, &load_const->instr);
+
+ return &load_const->def;
+}
- return nir_build_imm(build, 1, 1, v);
+static inline nir_ssa_def *
+nir_imm_boolN_t(nir_builder *build, bool x, unsigned bit_size)
+{
+ nir_const_value v = nir_const_value_for_bool(x, bit_size);
+ return nir_build_imm(build, 1, bit_size, &v);
+}
+
+static inline nir_ssa_def *
+nir_imm_bool(nir_builder *build, bool x)
+{
+ return nir_imm_boolN_t(build, x, 1);
}
static inline nir_ssa_def *
}
static inline nir_ssa_def *
-nir_imm_float16(nir_builder *build, float x)
+nir_imm_floatN_t(nir_builder *build, double x, unsigned bit_size)
{
- nir_const_value v;
-
- memset(&v, 0, sizeof(v));
- v.u16[0] = _mesa_float_to_half(x);
+ nir_const_value v = nir_const_value_for_float(x, bit_size);
+ return nir_build_imm(build, 1, bit_size, &v);
+}
- return nir_build_imm(build, 1, 16, v);
+static inline nir_ssa_def *
+nir_imm_float16(nir_builder *build, float x)
+{
+ return nir_imm_floatN_t(build, x, 16);
}
static inline nir_ssa_def *
nir_imm_float(nir_builder *build, float x)
{
- nir_const_value v;
-
- memset(&v, 0, sizeof(v));
- v.f32[0] = x;
-
- return nir_build_imm(build, 1, 32, v);
+ return nir_imm_floatN_t(build, x, 32);
}
static inline nir_ssa_def *
nir_imm_double(nir_builder *build, double x)
{
- nir_const_value v;
-
- memset(&v, 0, sizeof(v));
- v.f64[0] = x;
-
- return nir_build_imm(build, 1, 64, v);
+ return nir_imm_floatN_t(build, x, 64);
}
static inline nir_ssa_def *
-nir_imm_floatN_t(nir_builder *build, double x, unsigned bit_size)
+nir_imm_vec2(nir_builder *build, float x, float y)
{
- switch (bit_size) {
- case 16:
- return nir_imm_float16(build, x);
- case 32:
- return nir_imm_float(build, x);
- case 64:
- return nir_imm_double(build, x);
- }
-
- unreachable("unknown float immediate bit size");
+ nir_const_value v[2] = {
+ nir_const_value_for_float(x, 32),
+ nir_const_value_for_float(y, 32),
+ };
+ return nir_build_imm(build, 2, 32, v);
}
static inline nir_ssa_def *
nir_imm_vec4(nir_builder *build, float x, float y, float z, float w)
{
- nir_const_value v;
-
- memset(&v, 0, sizeof(v));
- v.f32[0] = x;
- v.f32[1] = y;
- v.f32[2] = z;
- v.f32[3] = w;
+ nir_const_value v[4] = {
+ nir_const_value_for_float(x, 32),
+ nir_const_value_for_float(y, 32),
+ nir_const_value_for_float(z, 32),
+ nir_const_value_for_float(w, 32),
+ };
return nir_build_imm(build, 4, 32, v);
}
static inline nir_ssa_def *
-nir_imm_ivec2(nir_builder *build, int x, int y)
+nir_imm_vec4_16(nir_builder *build, float x, float y, float z, float w)
{
- nir_const_value v;
+ nir_const_value v[4] = {
+ nir_const_value_for_float(x, 16),
+ nir_const_value_for_float(y, 16),
+ nir_const_value_for_float(z, 16),
+ nir_const_value_for_float(w, 16),
+ };
- memset(&v, 0, sizeof(v));
- v.i32[0] = x;
- v.i32[1] = y;
+ return nir_build_imm(build, 4, 16, v);
+}
- return nir_build_imm(build, 2, 32, v);
+static inline nir_ssa_def *
+nir_imm_intN_t(nir_builder *build, uint64_t x, unsigned bit_size)
+{
+ nir_const_value v = nir_const_value_for_raw_uint(x, bit_size);
+ return nir_build_imm(build, 1, bit_size, &v);
}
static inline nir_ssa_def *
nir_imm_int(nir_builder *build, int x)
{
- nir_const_value v;
-
- memset(&v, 0, sizeof(v));
- v.i32[0] = x;
-
- return nir_build_imm(build, 1, 32, v);
+ return nir_imm_intN_t(build, x, 32);
}
static inline nir_ssa_def *
nir_imm_int64(nir_builder *build, int64_t x)
{
- nir_const_value v;
-
- memset(&v, 0, sizeof(v));
- v.i64[0] = x;
-
- return nir_build_imm(build, 1, 64, v);
+ return nir_imm_intN_t(build, x, 64);
}
static inline nir_ssa_def *
-nir_imm_intN_t(nir_builder *build, uint64_t x, unsigned bit_size)
+nir_imm_ivec2(nir_builder *build, int x, int y)
{
- nir_const_value v;
-
- memset(&v, 0, sizeof(v));
- assert(bit_size <= 64);
- if (bit_size == 1)
- v.b[0] = x & 1;
- else
- v.i64[0] = x & (~0ull >> (64 - bit_size));
+ nir_const_value v[2] = {
+ nir_const_value_for_int(x, 32),
+ nir_const_value_for_int(y, 32),
+ };
- return nir_build_imm(build, 1, bit_size, v);
+ return nir_build_imm(build, 2, 32, v);
}
static inline nir_ssa_def *
nir_imm_ivec4(nir_builder *build, int x, int y, int z, int w)
{
- nir_const_value v;
-
- memset(&v, 0, sizeof(v));
- v.i32[0] = x;
- v.i32[1] = y;
- v.i32[2] = z;
- v.i32[3] = w;
+ nir_const_value v[4] = {
+ nir_const_value_for_int(x, 32),
+ nir_const_value_for_int(y, 32),
+ nir_const_value_for_int(z, 32),
+ nir_const_value_for_int(w, 32),
+ };
return nir_build_imm(build, 4, 32, v);
}
static inline nir_ssa_def *
-nir_imm_boolN_t(nir_builder *build, bool x, unsigned bit_size)
-{
- /* We use a 0/-1 convention for all booleans regardless of size */
- return nir_imm_intN_t(build, -(int)x, bit_size);
-}
-
-static inline nir_ssa_def *
-nir_build_alu(nir_builder *build, nir_op op, nir_ssa_def *src0,
- nir_ssa_def *src1, nir_ssa_def *src2, nir_ssa_def *src3)
+nir_builder_alu_instr_finish_and_insert(nir_builder *build, nir_alu_instr *instr)
{
- const nir_op_info *op_info = &nir_op_infos[op];
- nir_alu_instr *instr = nir_alu_instr_create(build->shader, op);
- if (!instr)
- return NULL;
+ const nir_op_info *op_info = &nir_op_infos[instr->op];
instr->exact = build->exact;
- instr->src[0].src = nir_src_for_ssa(src0);
- if (src1)
- instr->src[1].src = nir_src_for_ssa(src1);
- if (src2)
- instr->src[2].src = nir_src_for_ssa(src2);
- if (src3)
- instr->src[3].src = nir_src_for_ssa(src3);
-
/* Guess the number of components the destination temporary should have
* based on our input sizes, if it's not fixed for the op.
*/
return &instr->dest.dest.ssa;
}
-#include "nir_builder_opcodes.h"
-
static inline nir_ssa_def *
-nir_vec(nir_builder *build, nir_ssa_def **comp, unsigned num_components)
+nir_build_alu(nir_builder *build, nir_op op, nir_ssa_def *src0,
+ nir_ssa_def *src1, nir_ssa_def *src2, nir_ssa_def *src3)
{
- switch (num_components) {
- case 4:
- return nir_vec4(build, comp[0], comp[1], comp[2], comp[3]);
- case 3:
- return nir_vec3(build, comp[0], comp[1], comp[2]);
- case 2:
- return nir_vec2(build, comp[0], comp[1]);
- case 1:
- return comp[0];
- default:
- unreachable("bad component count");
+ nir_alu_instr *instr = nir_alu_instr_create(build->shader, op);
+ if (!instr)
return NULL;
- }
+
+ instr->src[0].src = nir_src_for_ssa(src0);
+ if (src1)
+ instr->src[1].src = nir_src_for_ssa(src1);
+ if (src2)
+ instr->src[2].src = nir_src_for_ssa(src2);
+ if (src3)
+ instr->src[3].src = nir_src_for_ssa(src3);
+
+ return nir_builder_alu_instr_finish_and_insert(build, instr);
}
-/**
- * Similar to nir_fmov, but takes a nir_alu_src instead of a nir_ssa_def.
- */
+/* for the couple special cases with more than 4 src args: */
static inline nir_ssa_def *
-nir_fmov_alu(nir_builder *build, nir_alu_src src, unsigned num_components)
+nir_build_alu_src_arr(nir_builder *build, nir_op op, nir_ssa_def **srcs)
{
- nir_alu_instr *mov = nir_alu_instr_create(build->shader, nir_op_fmov);
- nir_ssa_dest_init(&mov->instr, &mov->dest.dest, num_components,
- nir_src_bit_size(src.src), NULL);
- mov->exact = build->exact;
- mov->dest.write_mask = (1 << num_components) - 1;
- mov->src[0] = src;
- nir_builder_instr_insert(build, &mov->instr);
+ const nir_op_info *op_info = &nir_op_infos[op];
+ nir_alu_instr *instr = nir_alu_instr_create(build->shader, op);
+ if (!instr)
+ return NULL;
- return &mov->dest.dest.ssa;
+ for (unsigned i = 0; i < op_info->num_inputs; i++)
+ instr->src[i].src = nir_src_for_ssa(srcs[i]);
+
+ return nir_builder_alu_instr_finish_and_insert(build, instr);
}
+#include "nir_builder_opcodes.h"
+
static inline nir_ssa_def *
-nir_imov_alu(nir_builder *build, nir_alu_src src, unsigned num_components)
+nir_vec(nir_builder *build, nir_ssa_def **comp, unsigned num_components)
{
- nir_alu_instr *mov = nir_alu_instr_create(build->shader, nir_op_imov);
+ return nir_build_alu_src_arr(build, nir_op_vec(num_components), comp);
+}
+
+static inline nir_ssa_def *
+nir_mov_alu(nir_builder *build, nir_alu_src src, unsigned num_components)
+{
+ assert(!src.abs && !src.negate);
+ if (src.src.is_ssa && src.src.ssa->num_components == num_components) {
+ bool any_swizzles = false;
+ for (unsigned i = 0; i < num_components; i++) {
+ if (src.swizzle[i] != i)
+ any_swizzles = true;
+ }
+ if (!any_swizzles)
+ return src.src.ssa;
+ }
+
+ nir_alu_instr *mov = nir_alu_instr_create(build->shader, nir_op_mov);
nir_ssa_dest_init(&mov->instr, &mov->dest.dest, num_components,
nir_src_bit_size(src.src), NULL);
mov->exact = build->exact;
*/
static inline nir_ssa_def *
nir_swizzle(nir_builder *build, nir_ssa_def *src, const unsigned *swiz,
- unsigned num_components, bool use_fmov)
+ unsigned num_components)
{
assert(num_components <= NIR_MAX_VEC_COMPONENTS);
nir_alu_src alu_src = { NIR_SRC_INIT };
if (num_components == src->num_components && is_identity_swizzle)
return src;
- return use_fmov ? nir_fmov_alu(build, alu_src, num_components) :
- nir_imov_alu(build, alu_src, num_components);
+ return nir_mov_alu(build, alu_src, num_components);
}
/* Selects the right fdot given the number of components in each source. */
return NULL;
}
+static inline nir_ssa_def *
+nir_ball_iequal(nir_builder *b, nir_ssa_def *src0, nir_ssa_def *src1)
+{
+ switch (src0->num_components) {
+ case 1: return nir_ieq(b, src0, src1);
+ case 2: return nir_ball_iequal2(b, src0, src1);
+ case 3: return nir_ball_iequal3(b, src0, src1);
+ case 4: return nir_ball_iequal4(b, src0, src1);
+ default:
+ unreachable("bad component size");
+ }
+}
+
static inline nir_ssa_def *
nir_bany_inequal(nir_builder *b, nir_ssa_def *src0, nir_ssa_def *src1)
{
static inline nir_ssa_def *
nir_channel(nir_builder *b, nir_ssa_def *def, unsigned c)
{
- return nir_swizzle(b, def, &c, 1, false);
+ return nir_swizzle(b, def, &c, 1);
}
static inline nir_ssa_def *
swizzle[num_channels++] = i;
}
- return nir_swizzle(b, def, swizzle, num_channels, false);
+ 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 *
}
static inline nir_ssa_def *
-nir_imul_imm(nir_builder *build, nir_ssa_def *x, uint64_t y)
+_nir_mul_imm(nir_builder *build, nir_ssa_def *x, uint64_t y, bool amul)
{
assert(x->bit_size <= 64);
if (x->bit_size < 64)
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 if (amul) {
+ return nir_amul(build, x, nir_imm_intN_t(build, y, x->bit_size));
} else {
return nir_imul(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)
+{
+ return _nir_mul_imm(build, x, y, false);
+}
+
+static inline nir_ssa_def *
+nir_amul_imm(nir_builder *build, nir_ssa_def *x, uint64_t y)
+{
+ return _nir_mul_imm(build, x, y, true);
+}
+
static inline nir_ssa_def *
nir_fadd_imm(nir_builder *build, nir_ssa_def *x, double y)
{
return nir_vec(b, dest_comps, dest_num_components);
}
+/**
+ * Treats srcs as if it's one big blob of bits and extracts the range of bits
+ * given by
+ *
+ * [first_bit, first_bit + dest_num_components * dest_bit_size)
+ *
+ * The range can have any alignment or size as long as it's an integer number
+ * of destination components and fits inside the concatenated sources.
+ *
+ * TODO: The one caveat here is that we can't handle byte alignment if 64-bit
+ * values are involved because that would require pack/unpack to/from a vec8
+ * which NIR currently does not support.
+ */
+static inline nir_ssa_def *
+nir_extract_bits(nir_builder *b, nir_ssa_def **srcs, unsigned num_srcs,
+ unsigned first_bit,
+ unsigned dest_num_components, unsigned dest_bit_size)
+{
+ const unsigned num_bits = dest_num_components * dest_bit_size;
+
+ /* Figure out the common bit size */
+ unsigned common_bit_size = dest_bit_size;
+ for (unsigned i = 0; i < num_srcs; i++)
+ common_bit_size = MIN2(common_bit_size, srcs[i]->bit_size);
+ if (first_bit > 0)
+ common_bit_size = MIN2(common_bit_size, (1u << (ffs(first_bit) - 1)));
+
+ /* We don't want to have to deal with 1-bit values */
+ assert(common_bit_size >= 8);
+
+ nir_ssa_def *common_comps[NIR_MAX_VEC_COMPONENTS * sizeof(uint64_t)];
+ assert(num_bits / common_bit_size <= ARRAY_SIZE(common_comps));
+
+ /* First, unpack to the common bit size and select the components from the
+ * source.
+ */
+ int src_idx = -1;
+ unsigned src_start_bit = 0;
+ unsigned src_end_bit = 0;
+ for (unsigned i = 0; i < num_bits / common_bit_size; i++) {
+ const unsigned bit = first_bit + (i * common_bit_size);
+ while (bit >= src_end_bit) {
+ src_idx++;
+ assert(src_idx < (int) num_srcs);
+ src_start_bit = src_end_bit;
+ src_end_bit += srcs[src_idx]->bit_size *
+ srcs[src_idx]->num_components;
+ }
+ assert(bit >= src_start_bit);
+ assert(bit + common_bit_size <= src_end_bit);
+ const unsigned rel_bit = bit - src_start_bit;
+ const unsigned src_bit_size = srcs[src_idx]->bit_size;
+
+ nir_ssa_def *comp = nir_channel(b, srcs[src_idx],
+ rel_bit / src_bit_size);
+ if (srcs[src_idx]->bit_size > common_bit_size) {
+ nir_ssa_def *unpacked = nir_unpack_bits(b, comp, common_bit_size);
+ comp = nir_channel(b, unpacked, (rel_bit % src_bit_size) /
+ common_bit_size);
+ }
+ common_comps[i] = comp;
+ }
+
+ /* Now, re-pack the destination if we have to */
+ if (dest_bit_size > common_bit_size) {
+ unsigned common_per_dest = dest_bit_size / common_bit_size;
+ nir_ssa_def *dest_comps[NIR_MAX_VEC_COMPONENTS];
+ for (unsigned i = 0; i < dest_num_components; i++) {
+ nir_ssa_def *unpacked = nir_vec(b, common_comps + i * common_per_dest,
+ common_per_dest);
+ dest_comps[i] = nir_pack_bits(b, unpacked, dest_bit_size);
+ }
+ return nir_vec(b, dest_comps, dest_num_components);
+ } else {
+ assert(dest_bit_size == common_bit_size);
+ return nir_vec(b, common_comps, dest_num_components);
+ }
+}
+
static inline nir_ssa_def *
nir_bitcast_vector(nir_builder *b, nir_ssa_def *src, unsigned dest_bit_size)
{
(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;
- }
+ return nir_extract_bits(b, &src, 1, 0, dest_num_components, dest_bit_size);
}
/**
for (int j = 0; j < 4; j++)
alu.swizzle[j] = j;
- return nir_imov_alu(build, alu, num_components);
+ return nir_mov_alu(build, alu, num_components);
}
/**
static inline nir_ssa_def *
nir_ssa_for_alu_src(nir_builder *build, nir_alu_instr *instr, unsigned srcn)
{
- static uint8_t trivial_swizzle[NIR_MAX_VEC_COMPONENTS];
- for (int i = 0; i < NIR_MAX_VEC_COMPONENTS; ++i)
- trivial_swizzle[i] = i;
+ static uint8_t trivial_swizzle[] = { 0, 1, 2, 3 };
+ STATIC_ASSERT(ARRAY_SIZE(trivial_swizzle) == NIR_MAX_VEC_COMPONENTS);
+
nir_alu_src *src = &instr->src[srcn];
unsigned num_components = nir_ssa_alu_instr_src_components(instr, srcn);
(memcmp(src->swizzle, trivial_swizzle, num_components) == 0))
return src->src.ssa;
- return nir_imov_alu(build, *src, num_components);
+ return nir_mov_alu(build, *src, num_components);
+}
+
+static inline unsigned
+nir_get_ptr_bitsize(nir_builder *build)
+{
+ if (build->shader->info.stage == MESA_SHADER_KERNEL)
+ return build->shader->info.cs.ptr_size;
+ return 32;
}
static inline nir_deref_instr *
deref->type = var->type;
deref->var = var;
- nir_ssa_dest_init(&deref->instr, &deref->dest, 1, 32, NULL);
+ nir_ssa_dest_init(&deref->instr, &deref->dest, 1,
+ nir_get_ptr_bitsize(build), NULL);
nir_builder_instr_insert(build, &deref->instr);
glsl_type_is_matrix(parent->type) ||
glsl_type_is_vector(parent->type));
+ assert(index->bit_size == parent->dest.ssa.bit_size);
+
nir_deref_instr *deref =
nir_deref_instr_create(build->shader, nir_deref_type_array);
return deref;
}
+static inline nir_deref_instr *
+nir_build_deref_array_imm(nir_builder *build, nir_deref_instr *parent,
+ int64_t index)
+{
+ assert(parent->dest.is_ssa);
+ nir_ssa_def *idx_ssa = nir_imm_intN_t(build, index,
+ parent->dest.ssa.bit_size);
+
+ return nir_build_deref_array(build, parent, idx_ssa);
+}
+
static inline nir_deref_instr *
nir_build_deref_ptr_as_array(nir_builder *build, nir_deref_instr *parent,
nir_ssa_def *index)
parent->deref_type == nir_deref_type_ptr_as_array ||
parent->deref_type == nir_deref_type_cast);
+ assert(index->bit_size == parent->dest.ssa.bit_size);
+
nir_deref_instr *deref =
nir_deref_instr_create(build->shader, nir_deref_type_ptr_as_array);
if (leader->deref_type == nir_deref_type_array) {
assert(leader->arr.index.is_ssa);
- return nir_build_deref_array(b, parent, leader->arr.index.ssa);
+ nir_ssa_def *index = nir_i2i(b, leader->arr.index.ssa,
+ parent->dest.ssa.bit_size);
+ return nir_build_deref_array(b, parent, index);
} else {
return nir_build_deref_array_wildcard(b, parent);
}
}
static inline nir_ssa_def *
-nir_load_deref(nir_builder *build, nir_deref_instr *deref)
+nir_load_deref_with_access(nir_builder *build, nir_deref_instr *deref,
+ enum gl_access_qualifier access)
{
nir_intrinsic_instr *load =
nir_intrinsic_instr_create(build->shader, nir_intrinsic_load_deref);
load->src[0] = nir_src_for_ssa(&deref->dest.ssa);
nir_ssa_dest_init(&load->instr, &load->dest, load->num_components,
glsl_get_bit_size(deref->type), NULL);
+ nir_intrinsic_set_access(load, access);
nir_builder_instr_insert(build, &load->instr);
return &load->dest.ssa;
}
+static inline nir_ssa_def *
+nir_load_deref(nir_builder *build, nir_deref_instr *deref)
+{
+ return nir_load_deref_with_access(build, deref, (enum gl_access_qualifier)0);
+}
+
static inline void
-nir_store_deref(nir_builder *build, nir_deref_instr *deref,
- nir_ssa_def *value, unsigned writemask)
+nir_store_deref_with_access(nir_builder *build, nir_deref_instr *deref,
+ nir_ssa_def *value, unsigned writemask,
+ enum gl_access_qualifier access)
{
nir_intrinsic_instr *store =
nir_intrinsic_instr_create(build->shader, nir_intrinsic_store_deref);
store->src[1] = nir_src_for_ssa(value);
nir_intrinsic_set_write_mask(store,
writemask & ((1 << store->num_components) - 1));
+ nir_intrinsic_set_access(store, access);
nir_builder_instr_insert(build, &store->instr);
}
static inline void
-nir_copy_deref(nir_builder *build, nir_deref_instr *dest, nir_deref_instr *src)
+nir_store_deref(nir_builder *build, nir_deref_instr *deref,
+ nir_ssa_def *value, unsigned writemask)
+{
+ nir_store_deref_with_access(build, deref, value, writemask,
+ (enum gl_access_qualifier)0);
+}
+
+static inline void
+nir_copy_deref_with_access(nir_builder *build, nir_deref_instr *dest,
+ nir_deref_instr *src,
+ enum gl_access_qualifier dest_access,
+ enum gl_access_qualifier src_access)
{
nir_intrinsic_instr *copy =
nir_intrinsic_instr_create(build->shader, nir_intrinsic_copy_deref);
copy->src[0] = nir_src_for_ssa(&dest->dest.ssa);
copy->src[1] = nir_src_for_ssa(&src->dest.ssa);
+ nir_intrinsic_set_dst_access(copy, dest_access);
+ nir_intrinsic_set_src_access(copy, src_access);
nir_builder_instr_insert(build, ©->instr);
}
+static inline void
+nir_copy_deref(nir_builder *build, nir_deref_instr *dest, nir_deref_instr *src)
+{
+ nir_copy_deref_with_access(build, dest, src,
+ (enum gl_access_qualifier) 0,
+ (enum gl_access_qualifier) 0);
+}
+
static inline nir_ssa_def *
nir_load_var(nir_builder *build, nir_variable *var)
{
};
}
+static inline nir_ssa_def *
+nir_b2i(nir_builder *build, nir_ssa_def *b, uint32_t bit_size)
+{
+ switch (bit_size) {
+ case 64: return nir_b2i64(build, b);
+ case 32: return nir_b2i32(build, b);
+ case 16: return nir_b2i16(build, b);
+ case 8: return nir_b2i8(build, b);
+ default:
+ unreachable("Invalid bit-size");
+ };
+}
static inline nir_ssa_def *
nir_load_barycentric(nir_builder *build, nir_intrinsic_op op,
unsigned interp_mode)