freedreno/ir3: un-embed const_state

Make it an rzalloc'd ptr instead of embedded struct, so it can serve as
the mem ctx for immediates.  This gets rid of needing to explicitly free
the immediates, so one less thing to deal with when moving const_state.
(Also, after we move const_state to the shader variant, we won't need
one for binning pass variants)

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5508>

diff --git a/src/freedreno/ir3/ir3_assembler.c b/src/freedreno/ir3/ir3_assembler.c
index 005f528..e81eda6 100644 (file)
--- a/src/freedreno/ir3/ir3_assembler.c
+++ b/src/freedreno/ir3/ir3_assembler.c
@@ -37,6 +37,7 @@ ir3_parse_asm(struct ir3_compiler *c, struct ir3_kernel_info *info, FILE *in)
        struct ir3_shader *shader = rzalloc_size(NULL, sizeof(*shader));
        shader->compiler = c;
        shader->type = MESA_SHADER_COMPUTE;
+       shader->const_state = rzalloc_size(shader, sizeof(*shader->const_state));
        mtx_init(&shader->variants_lock, mtx_plain);
 
        struct ir3_shader_variant *v = rzalloc_size(shader, sizeof(*v));

diff --git a/src/freedreno/ir3/ir3_cp.c b/src/freedreno/ir3/ir3_cp.c
index a9562af..e739f79 100644 (file)
--- a/src/freedreno/ir3/ir3_cp.c
+++ b/src/freedreno/ir3/ir3_cp.c
@@ -199,9 +199,12 @@ lower_immed(struct ir3_cp_ctx *ctx, struct ir3_instruction *instr, unsigned n,
        /* Reallocate for 4 more elements whenever it's necessary */
        struct ir3_const_state *const_state = ir3_const_state(ctx->so);
        if (const_state->immediate_idx == const_state->immediates_size * 4) {
+               const_state->immediates = rerzalloc(const_state,
+                               const_state->immediates,
+                               __typeof__(const_state->immediates[0]),
+                               const_state->immediates_size,
+                               const_state->immediates_size + 4);
                const_state->immediates_size += 4;
-               const_state->immediates = realloc (const_state->immediates,
-                       const_state->immediates_size * sizeof(const_state->immediates[0]));
 
                for (int i = const_state->immediate_idx; i < const_state->immediates_size * 4; i++)
                        const_state->immediates[i / 4].val[i % 4] = 0xd0d0d0d0;

diff --git a/src/freedreno/ir3/ir3_nir.c b/src/freedreno/ir3/ir3_nir.c
index 2df5715..de76cb7 100644 (file)
--- a/src/freedreno/ir3/ir3_nir.c
+++ b/src/freedreno/ir3/ir3_nir.c
@@ -277,7 +277,7 @@ ir3_optimize_nir(struct ir3_shader *shader, nir_shader *s)
         * const_state layout setup.  This should be done after ubo range
         * analysis.
         */
-       ir3_setup_const_state(shader, s, &shader->const_state);
+       ir3_setup_const_state(shader, s, shader->const_state);
 }
 
 void

diff --git a/src/freedreno/ir3/ir3_nir_analyze_ubo_ranges.c b/src/freedreno/ir3/ir3_nir_analyze_ubo_ranges.c
index b41bfe4..0ae11a3 100644 (file)
--- a/src/freedreno/ir3/ir3_nir_analyze_ubo_ranges.c
+++ b/src/freedreno/ir3/ir3_nir_analyze_ubo_ranges.c
@@ -304,7 +304,8 @@ instr_is_load_ubo(nir_instr *instr)
 bool
 ir3_nir_analyze_ubo_ranges(nir_shader *nir, struct ir3_shader *shader)
 {
-       struct ir3_ubo_analysis_state *state = &shader->const_state.ubo_state;
+       struct ir3_const_state *const_state = shader->const_state;
+       struct ir3_ubo_analysis_state *state = &const_state->ubo_state;
 
        memset(state, 0, sizeof(*state));
        for (int i = 0; i < IR3_MAX_UBO_PUSH_RANGES; i++) {

diff --git a/src/freedreno/ir3/ir3_parser.y b/src/freedreno/ir3/ir3_parser.y
index 2e89d77..27c52bd 100644 (file)
--- a/src/freedreno/ir3/ir3_parser.y
+++ b/src/freedreno/ir3/ir3_parser.y
@@ -157,9 +157,12 @@ static void add_const(unsigned reg, unsigned c0, unsigned c1, unsigned c2, unsig
        assert((reg & 0x7) == 0);
        int idx = reg >> (1 + 2); /* low bit is half vs full, next two bits are swiz */
        if (const_state->immediate_idx == const_state->immediates_size * 4) {
+               const_state->immediates = rerzalloc(const_state,
+                               const_state->immediates,
+                               __typeof__(const_state->immediates[0]),
+                               const_state->immediates_size,
+                               const_state->immediates_size + 4);
                const_state->immediates_size += 4;
-               const_state->immediates = realloc (const_state->immediates,
-                       const_state->immediates_size * sizeof(const_state->immediates[0]));
        }
        const_state->immediates[idx].val[0] = c0;
        const_state->immediates[idx].val[1] = c1;

diff --git a/src/freedreno/ir3/ir3_shader.c b/src/freedreno/ir3/ir3_shader.c
index c7d06a9..66f72c9 100644 (file)
--- a/src/freedreno/ir3/ir3_shader.c
+++ b/src/freedreno/ir3/ir3_shader.c
@@ -271,7 +271,6 @@ ir3_shader_get_variant(struct ir3_shader *shader, const struct ir3_shader_key *k
 void
 ir3_shader_destroy(struct ir3_shader *shader)
 {
-       free(shader->const_state.immediates);
        ralloc_free(shader->nir);
        mtx_destroy(&shader->variants_lock);
        ralloc_free(shader);
@@ -346,6 +345,7 @@ ir3_shader_from_nir(struct ir3_compiler *compiler, nir_shader *nir,
        if (stream_output)
                memcpy(&shader->stream_output, stream_output, sizeof(shader->stream_output));
        shader->num_reserved_user_consts = reserved_user_consts;
+       shader->const_state = rzalloc_size(shader, sizeof(*shader->const_state));
 
        if (nir->info.stage == MESA_SHADER_GEOMETRY)
                NIR_PASS_V(nir, ir3_nir_lower_gs);

diff --git a/src/freedreno/ir3/ir3_shader.h b/src/freedreno/ir3/ir3_shader.h
index b9dd4a4..ff131cc 100644 (file)
--- a/src/freedreno/ir3/ir3_shader.h
+++ b/src/freedreno/ir3/ir3_shader.h
@@ -621,7 +621,7 @@ struct ir3_shader {
 
        struct ir3_compiler *compiler;
 
-       struct ir3_const_state const_state;
+       struct ir3_const_state *const_state;
        unsigned num_reserved_user_consts;
 
        struct nir_shader *nir;
@@ -644,7 +644,7 @@ struct ir3_shader {
 static inline struct ir3_const_state *
 ir3_const_state(const struct ir3_shader_variant *v)
 {
-       return &v->shader->const_state;
+       return v->shader->const_state;
 }
 
 void * ir3_shader_assemble(struct ir3_shader_variant *v);

diff --git a/src/freedreno/vulkan/tu_clear_blit.c b/src/freedreno/vulkan/tu_clear_blit.c
index df2359f..e6e7daf 100644 (file)
--- a/src/freedreno/vulkan/tu_clear_blit.c
+++ b/src/freedreno/vulkan/tu_clear_blit.c
@@ -321,7 +321,10 @@ static void
 r3d_common(struct tu_cmd_buffer *cmd, struct tu_cs *cs, bool blit, uint32_t num_rts,
            bool layered_clear)
 {
-   struct ir3_shader dummy_shader = {};
+   struct ir3_const_state dummy_const_state = {};
+   struct ir3_shader dummy_shader = {
+      .const_state = &dummy_const_state,
+   };
 
    struct ir3_shader_variant vs = {
       .type = MESA_SHADER_VERTEX,