From 8c8fc5f8336c8c79e5890265ae6c03271aa94075 Mon Sep 17 00:00:00 2001 From: Jason Ekstrand Date: Wed, 9 Sep 2015 13:18:29 -0700 Subject: [PATCH] nir: Fix a bunch of ralloc parenting errors As of a10d4937, we would really like things associated with an instruction to be allocated out of that instruction and not out of the shader. In particular, you should be passing the instruction that will ultimately be holding the source into nir_src_copy rather than an arbitrary memory context. We also change the prototypes of nir_dest_copy and nir_alu_src/dest_copy to explicitly take an instruction so we catch this earlier in the future. Cc: "11.0" Reviewed-by: Thomas Helland --- src/glsl/nir/nir.c | 20 +++++++++++--------- src/glsl/nir/nir.h | 13 +++++++------ src/glsl/nir/nir_from_ssa.c | 2 +- src/glsl/nir/nir_lower_alu_to_scalar.c | 6 +++--- src/glsl/nir/nir_lower_atomics.c | 2 +- src/glsl/nir/nir_lower_io.c | 2 +- src/glsl/nir/nir_lower_locals_to_regs.c | 7 +++---- src/glsl/nir/nir_lower_vec_to_movs.c | 4 ++-- src/glsl/nir/nir_opt_peephole_ffma.c | 3 +-- src/glsl/nir/nir_opt_peephole_select.c | 4 ++-- 10 files changed, 32 insertions(+), 31 deletions(-) diff --git a/src/glsl/nir/nir.c b/src/glsl/nir/nir.c index 1dc7e12..fd67532 100644 --- a/src/glsl/nir/nir.c +++ b/src/glsl/nir/nir.c @@ -153,7 +153,7 @@ void nir_src_copy(nir_src *dest, const nir_src *src, void *mem_ctx) } } -void nir_dest_copy(nir_dest *dest, const nir_dest *src, void *mem_ctx) +void nir_dest_copy(nir_dest *dest, const nir_dest *src, nir_instr *instr) { /* Copying an SSA definition makes no sense whatsoever. */ assert(!src->is_ssa); @@ -163,17 +163,18 @@ void nir_dest_copy(nir_dest *dest, const nir_dest *src, void *mem_ctx) dest->reg.base_offset = src->reg.base_offset; dest->reg.reg = src->reg.reg; if (src->reg.indirect) { - dest->reg.indirect = ralloc(mem_ctx, nir_src); - nir_src_copy(dest->reg.indirect, src->reg.indirect, mem_ctx); + dest->reg.indirect = ralloc(instr, nir_src); + nir_src_copy(dest->reg.indirect, src->reg.indirect, instr); } else { dest->reg.indirect = NULL; } } void -nir_alu_src_copy(nir_alu_src *dest, const nir_alu_src *src, void *mem_ctx) +nir_alu_src_copy(nir_alu_src *dest, const nir_alu_src *src, + nir_alu_instr *instr) { - nir_src_copy(&dest->src, &src->src, mem_ctx); + nir_src_copy(&dest->src, &src->src, &instr->instr); dest->abs = src->abs; dest->negate = src->negate; for (unsigned i = 0; i < 4; i++) @@ -181,9 +182,10 @@ nir_alu_src_copy(nir_alu_src *dest, const nir_alu_src *src, void *mem_ctx) } void -nir_alu_dest_copy(nir_alu_dest *dest, const nir_alu_dest *src, void *mem_ctx) +nir_alu_dest_copy(nir_alu_dest *dest, const nir_alu_dest *src, + nir_alu_instr *instr) { - nir_dest_copy(&dest->dest, &src->dest, mem_ctx); + nir_dest_copy(&dest->dest, &src->dest, &instr->instr); dest->write_mask = src->write_mask; dest->saturate = src->saturate; } @@ -1210,14 +1212,14 @@ nir_ssa_def_rewrite_uses(nir_ssa_def *def, nir_src new_src, void *mem_ctx) nir_foreach_use_safe(def, use_src) { nir_instr *src_parent_instr = use_src->parent_instr; list_del(&use_src->use_link); - nir_src_copy(use_src, &new_src, mem_ctx); + nir_src_copy(use_src, &new_src, src_parent_instr); src_add_all_uses(use_src, src_parent_instr, NULL); } nir_foreach_if_use_safe(def, use_src) { nir_if *src_parent_if = use_src->parent_if; list_del(&use_src->use_link); - nir_src_copy(use_src, &new_src, mem_ctx); + nir_src_copy(use_src, &new_src, src_parent_if); src_add_all_uses(use_src, NULL, src_parent_if); } } diff --git a/src/glsl/nir/nir.h b/src/glsl/nir/nir.h index f9c8295..92945f9 100644 --- a/src/glsl/nir/nir.h +++ b/src/glsl/nir/nir.h @@ -580,8 +580,8 @@ nir_dest_for_reg(nir_register *reg) return dest; } -void nir_src_copy(nir_src *dest, const nir_src *src, void *mem_ctx); -void nir_dest_copy(nir_dest *dest, const nir_dest *src, void *mem_ctx); +void nir_src_copy(nir_src *dest, const nir_src *src, void *instr_or_if); +void nir_dest_copy(nir_dest *dest, const nir_dest *src, nir_instr *instr); typedef struct { nir_src src; @@ -630,10 +630,6 @@ typedef struct { unsigned write_mask : 4; /* ignored if dest.is_ssa is true */ } nir_alu_dest; -void nir_alu_src_copy(nir_alu_src *dest, const nir_alu_src *src, void *mem_ctx); -void nir_alu_dest_copy(nir_alu_dest *dest, const nir_alu_dest *src, - void *mem_ctx); - typedef enum { nir_type_invalid = 0, /* Not a valid type */ nir_type_float, @@ -702,6 +698,11 @@ typedef struct nir_alu_instr { nir_alu_src src[]; } nir_alu_instr; +void nir_alu_src_copy(nir_alu_src *dest, const nir_alu_src *src, + nir_alu_instr *instr); +void nir_alu_dest_copy(nir_alu_dest *dest, const nir_alu_dest *src, + nir_alu_instr *instr); + /* is this source channel used? */ static inline bool nir_alu_instr_channel_used(nir_alu_instr *instr, unsigned src, unsigned channel) diff --git a/src/glsl/nir/nir_from_ssa.c b/src/glsl/nir/nir_from_ssa.c index 94002f1..d6569d8 100644 --- a/src/glsl/nir/nir_from_ssa.c +++ b/src/glsl/nir/nir_from_ssa.c @@ -556,7 +556,7 @@ emit_copy(nir_parallel_copy_instr *pcopy, nir_src src, nir_src dest_src, assert(src.reg.reg->num_components >= dest_src.reg.reg->num_components); nir_alu_instr *mov = nir_alu_instr_create(mem_ctx, nir_op_imov); - nir_src_copy(&mov->src[0].src, &src, mem_ctx); + nir_src_copy(&mov->src[0].src, &src, mov); mov->dest.dest = nir_dest_for_reg(dest_src.reg.reg); mov->dest.write_mask = (1 << dest_src.reg.reg->num_components) - 1; diff --git a/src/glsl/nir/nir_lower_alu_to_scalar.c b/src/glsl/nir/nir_lower_alu_to_scalar.c index efbe9e7..1607308 100644 --- a/src/glsl/nir/nir_lower_alu_to_scalar.c +++ b/src/glsl/nir/nir_lower_alu_to_scalar.c @@ -46,11 +46,11 @@ lower_reduction(nir_alu_instr *instr, nir_op chan_op, nir_op merge_op, for (unsigned i = 0; i < num_components; i++) { nir_alu_instr *chan = nir_alu_instr_create(mem_ctx, chan_op); nir_alu_ssa_dest_init(chan, 1); - nir_alu_src_copy(&chan->src[0], &instr->src[0], mem_ctx); + nir_alu_src_copy(&chan->src[0], &instr->src[0], chan); chan->src[0].swizzle[0] = chan->src[0].swizzle[i]; if (nir_op_infos[chan_op].num_inputs > 1) { assert(nir_op_infos[chan_op].num_inputs == 2); - nir_alu_src_copy(&chan->src[1], &instr->src[1], mem_ctx); + nir_alu_src_copy(&chan->src[1], &instr->src[1], chan); chan->src[1].swizzle[0] = chan->src[1].swizzle[i]; } @@ -153,7 +153,7 @@ lower_alu_instr_scalar(nir_alu_instr *instr, void *mem_ctx) unsigned src_chan = (nir_op_infos[instr->op].input_sizes[i] == 1 ? 0 : chan); - nir_alu_src_copy(&lower->src[i], &instr->src[i], mem_ctx); + nir_alu_src_copy(&lower->src[i], &instr->src[i], lower); for (int j = 0; j < 4; j++) lower->src[i].swizzle[j] = instr->src[i].swizzle[src_chan]; } diff --git a/src/glsl/nir/nir_lower_atomics.c b/src/glsl/nir/nir_lower_atomics.c index ce3615a..7ae8462 100644 --- a/src/glsl/nir/nir_lower_atomics.c +++ b/src/glsl/nir/nir_lower_atomics.c @@ -91,7 +91,7 @@ lower_instr(nir_intrinsic_instr *instr, nir_function_impl *impl) nir_alu_instr *mul = nir_alu_instr_create(mem_ctx, nir_op_imul); nir_ssa_dest_init(&mul->instr, &mul->dest.dest, 1, NULL); mul->dest.write_mask = 0x1; - nir_src_copy(&mul->src[0].src, &deref_array->indirect, mem_ctx); + nir_src_copy(&mul->src[0].src, &deref_array->indirect, mul); mul->src[1].src.is_ssa = true; mul->src[1].src.ssa = &atomic_counter_size->def; nir_instr_insert_before(&instr->instr, &mul->instr); diff --git a/src/glsl/nir/nir_lower_io.c b/src/glsl/nir/nir_lower_io.c index afb4630..3739fc8 100644 --- a/src/glsl/nir/nir_lower_io.c +++ b/src/glsl/nir/nir_lower_io.c @@ -221,7 +221,7 @@ nir_lower_io_block(nir_block *block, void *void_state) store->const_index[0] = offset; - nir_src_copy(&store->src[0], &intrin->src[0], state->mem_ctx); + nir_src_copy(&store->src[0], &intrin->src[0], store); if (has_indirect) store->src[1] = indirect; diff --git a/src/glsl/nir/nir_lower_locals_to_regs.c b/src/glsl/nir/nir_lower_locals_to_regs.c index 28fdec5..b77d974 100644 --- a/src/glsl/nir/nir_lower_locals_to_regs.c +++ b/src/glsl/nir/nir_lower_locals_to_regs.c @@ -183,8 +183,7 @@ get_deref_reg_src(nir_deref_var *deref, nir_instr *instr, nir_alu_instr *add = nir_alu_instr_create(state->shader, nir_op_iadd); add->src[0].src = *src.reg.indirect; - nir_src_copy(&add->src[1].src, &deref_array->indirect, - state->shader); + nir_src_copy(&add->src[1].src, &deref_array->indirect, add); add->dest.write_mask = 1; nir_ssa_dest_init(&add->instr, &add->dest.dest, 1, NULL); nir_instr_insert_before(instr, &add->instr); @@ -225,7 +224,7 @@ lower_locals_to_regs_block(nir_block *block, void *void_state) nir_src_for_ssa(&mov->dest.dest.ssa), state->shader); } else { - nir_dest_copy(&mov->dest.dest, &intrin->dest, state->shader); + nir_dest_copy(&mov->dest.dest, &intrin->dest, &mov->instr); } nir_instr_insert_before(&intrin->instr, &mov->instr); @@ -241,7 +240,7 @@ lower_locals_to_regs_block(nir_block *block, void *void_state) &intrin->instr, state); nir_alu_instr *mov = nir_alu_instr_create(state->shader, nir_op_imov); - nir_src_copy(&mov->src[0].src, &intrin->src[0], state->shader); + nir_src_copy(&mov->src[0].src, &intrin->src[0], mov); mov->dest.write_mask = (1 << intrin->num_components) - 1; mov->dest.dest.is_ssa = false; mov->dest.dest.reg.reg = reg_src.reg.reg; diff --git a/src/glsl/nir/nir_lower_vec_to_movs.c b/src/glsl/nir/nir_lower_vec_to_movs.c index e6d522f..b7f096d 100644 --- a/src/glsl/nir/nir_lower_vec_to_movs.c +++ b/src/glsl/nir/nir_lower_vec_to_movs.c @@ -60,8 +60,8 @@ insert_mov(nir_alu_instr *vec, unsigned start_channel, assert(src_idx < nir_op_infos[vec->op].num_inputs); nir_alu_instr *mov = nir_alu_instr_create(mem_ctx, nir_op_imov); - nir_alu_src_copy(&mov->src[0], &vec->src[src_idx], mem_ctx); - nir_alu_dest_copy(&mov->dest, &vec->dest, mem_ctx); + nir_alu_src_copy(&mov->src[0], &vec->src[src_idx], mov); + nir_alu_dest_copy(&mov->dest, &vec->dest, mov); mov->dest.write_mask = (1u << start_channel); mov->src[0].swizzle[start_channel] = vec->src[src_idx].swizzle[0]; diff --git a/src/glsl/nir/nir_opt_peephole_ffma.c b/src/glsl/nir/nir_opt_peephole_ffma.c index 97538e5..a23123e 100644 --- a/src/glsl/nir/nir_opt_peephole_ffma.c +++ b/src/glsl/nir/nir_opt_peephole_ffma.c @@ -216,8 +216,7 @@ nir_opt_peephole_ffma_block(nir_block *block, void *void_state) for (unsigned j = 0; j < add->dest.dest.ssa.num_components; j++) ffma->src[i].swizzle[j] = mul->src[i].swizzle[swizzle[j]]; } - nir_alu_src_copy(&ffma->src[2], &add->src[1 - add_mul_src], - state->mem_ctx); + nir_alu_src_copy(&ffma->src[2], &add->src[1 - add_mul_src], ffma); assert(add->dest.dest.is_ssa); diff --git a/src/glsl/nir/nir_opt_peephole_select.c b/src/glsl/nir/nir_opt_peephole_select.c index 26ec4ed..5b6037a 100644 --- a/src/glsl/nir/nir_opt_peephole_select.c +++ b/src/glsl/nir/nir_opt_peephole_select.c @@ -196,7 +196,7 @@ nir_opt_peephole_select_block(nir_block *block, void *void_state) nir_phi_instr *phi = nir_instr_as_phi(instr); nir_alu_instr *sel = nir_alu_instr_create(state->mem_ctx, nir_op_bcsel); - nir_src_copy(&sel->src[0].src, &if_stmt->condition, state->mem_ctx); + nir_src_copy(&sel->src[0].src, &if_stmt->condition, sel); /* Splat the condition to all channels */ memset(sel->src[0].swizzle, 0, sizeof sel->src[0].swizzle); @@ -206,7 +206,7 @@ nir_opt_peephole_select_block(nir_block *block, void *void_state) assert(src->src.is_ssa); unsigned idx = src->pred == then_block ? 1 : 2; - nir_src_copy(&sel->src[idx].src, &src->src, state->mem_ctx); + nir_src_copy(&sel->src[idx].src, &src->src, sel); } nir_ssa_dest_init(&sel->instr, &sel->dest.dest, -- 2.7.4