[Mesa-dev] [PATCH 1/2] nir/search: Don't match inexact expressions with exact subexpressions
Ian Romanick
idr at freedesktop.org
Tue Mar 29 21:50:46 UTC 2016
Does this fix any test cases? Either way, this series is
Reviewed-by: Ian Romanick <ian.d.romanick at intel.com>
On 03/28/2016 04:54 PM, Jason Ekstrand wrote:
> In the first pass of implementing exact handling, I made a mistake with
> search-and-replace. In particular, we only reallly handled exact/inexact
> on the root of the tree. Instead, we need to check every node in the tree
> for an exact/inexact match. As an example of this, consider the following
> GLSL code
>
> precise float a = b + c;
> if (a < 0) {
> do_stuff();
> }
>
> In that case, only the add will be declared "exact" and an expression that
> looks for "b + c < 0" will still match and replace it with "b < -c" which
> may yield different results. The solution is to simply bail if any of the
> values are exact when matching an inexact expression.
> ---
> src/compiler/nir/nir_search.c | 23 ++++++++++++++++++-----
> 1 file changed, 18 insertions(+), 5 deletions(-)
>
> diff --git a/src/compiler/nir/nir_search.c b/src/compiler/nir/nir_search.c
> index 6e63063..b915101 100644
> --- a/src/compiler/nir/nir_search.c
> +++ b/src/compiler/nir/nir_search.c
> @@ -28,6 +28,8 @@
> #include "nir_search.h"
>
> struct match_state {
> + bool inexact_match;
> + bool has_exact_alu;
> unsigned variables_seen;
> nir_alu_src variables[NIR_SEARCH_MAX_VARIABLES];
> };
> @@ -239,7 +241,10 @@ match_expression(const nir_search_expression *expr, nir_alu_instr *instr,
> return false;
>
> assert(instr->dest.dest.is_ssa);
> - if (expr->inexact && instr->exact)
> +
> + state->inexact_match = expr->inexact || state->inexact_match;
> + state->has_exact_alu = instr->exact || state->has_exact_alu;
> + if (state->inexact_match && state->has_exact_alu)
> return false;
>
> assert(!instr->dest.saturate);
> @@ -410,7 +415,7 @@ bitsize_tree_filter_down(bitsize_tree *tree, unsigned size)
>
> static nir_alu_src
> construct_value(const nir_search_value *value,
> - unsigned num_components, bitsize_tree *bitsize, bool exact,
> + unsigned num_components, bitsize_tree *bitsize,
> struct match_state *state,
> nir_instr *instr, void *mem_ctx)
> {
> @@ -424,10 +429,16 @@ construct_value(const nir_search_value *value,
> nir_alu_instr *alu = nir_alu_instr_create(mem_ctx, expr->opcode);
> nir_ssa_dest_init(&alu->instr, &alu->dest.dest, num_components,
> bitsize->dest_size, NULL);
> - alu->exact = exact;
> alu->dest.write_mask = (1 << num_components) - 1;
> alu->dest.saturate = false;
>
> + /* We have no way of knowing what values in a given search expression
> + * map to a particular replacement value. Therefore, if the
> + * expression we are replacing has any exact values, the entire
> + * replacement should be exact.
> + */
> + alu->exact = state->has_exact_alu;
> +
> for (unsigned i = 0; i < nir_op_infos[expr->opcode].num_inputs; i++) {
> /* If the source is an explicitly sized source, then we need to reset
> * the number of components to match.
> @@ -436,7 +447,7 @@ construct_value(const nir_search_value *value,
> num_components = nir_op_infos[alu->op].input_sizes[i];
>
> alu->src[i] = construct_value(expr->srcs[i],
> - num_components, bitsize->srcs[i], exact,
> + num_components, bitsize->srcs[i],
> state, instr, mem_ctx);
> }
>
> @@ -546,6 +557,8 @@ nir_replace_instr(nir_alu_instr *instr, const nir_search_expression *search,
> assert(instr->dest.dest.is_ssa);
>
> struct match_state state;
> + state.inexact_match = false;
> + state.has_exact_alu = false;
> state.variables_seen = 0;
>
> if (!match_expression(search, instr, instr->dest.dest.ssa.num_components,
> @@ -569,7 +582,7 @@ nir_replace_instr(nir_alu_instr *instr, const nir_search_expression *search,
>
> mov->src[0] = construct_value(replace,
> instr->dest.dest.ssa.num_components, tree,
> - instr->exact, &state, &instr->instr, mem_ctx);
> + &state, &instr->instr, mem_ctx);
> nir_instr_insert_before(&instr->instr, &mov->instr);
>
> nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa,
>
More information about the mesa-dev
mailing list