freedreno: Only upload the used part of UBO0 to the constant buffer.

We were pessimistically uploading all of it in case of indirection,
but we can just bump that when we encounter indirection.

total constlen in shared programs: 2529623 -> 2485933 (-1.73%)

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Rob Clark <robdclark@gmail.com>

diff --git a/src/freedreno/ir3/ir3_compiler_nir.c b/src/freedreno/ir3/ir3_compiler_nir.c
index e453f33fb91ffe968cc96345b13fc9d9b65904a0..d84c56b01957fd1f7f31bd587cebabda31a31a0e 100644 (file)
--- a/src/freedreno/ir3/ir3_compiler_nir.c
+++ b/src/freedreno/ir3/ir3_compiler_nir.c
@@ -1262,7 +1262,8 @@ emit_intrinsic(struct ir3_context *ctx, nir_intrinsic_instr *intr)
                         * since we don't know in the assembler what the max
                         * addr reg value can be:
                         */
-                       ctx->so->constlen = MAX2(ctx->so->constlen, ctx->s->num_uniforms);
+                       ctx->so->constlen = MAX2(ctx->so->constlen,
+                                       ctx->so->shader->ubo_state.size / 16);
                }
                break;
        case nir_intrinsic_load_ubo:

diff --git a/src/freedreno/ir3/ir3_nir_analyze_ubo_ranges.c b/src/freedreno/ir3/ir3_nir_analyze_ubo_ranges.c
index 46216a6f8621e52490479c8cb691fcaa779bd922..b3191a36c14b8f9a341f1e426f6139368ba5b10d 100644 (file)
--- a/src/freedreno/ir3/ir3_nir_analyze_ubo_ranges.c
+++ b/src/freedreno/ir3/ir3_nir_analyze_ubo_ranges.c
@@ -42,14 +42,22 @@ get_ubo_load_range(nir_intrinsic_instr *instr)
 }
 
 static void
-gather_ubo_ranges(nir_intrinsic_instr *instr,
+gather_ubo_ranges(nir_shader *nir, nir_intrinsic_instr *instr,
                                  struct ir3_ubo_analysis_state *state)
 {
        if (!nir_src_is_const(instr->src[0]))
                return;
 
-       if (!nir_src_is_const(instr->src[1]))
+       if (!nir_src_is_const(instr->src[1])) {
+               if (nir_src_as_uint(instr->src[0]) == 0) {
+                       /* If this is an indirect on UBO 0, we'll still lower it back to
+                        * load_uniform.  Set the range to cover all of UBO 0.
+                        */
+                       state->range[0].end = align(nir->num_uniforms * 16, 16 * 4);
+               }
+
                return;
+       }
 
        const struct ir3_ubo_range r = get_ubo_load_range(instr);
        const uint32_t block = nir_src_as_uint(instr->src[0]);
@@ -132,7 +140,6 @@ ir3_nir_analyze_ubo_ranges(nir_shader *nir, struct ir3_shader *shader)
        struct ir3_ubo_analysis_state *state = &shader->ubo_state;
 
        memset(state, 0, sizeof(*state));
-       state->range[0].end = align(nir->num_uniforms * 16, 16 * 4); /* align to 4*vec4 */
 
        nir_foreach_function(function, nir) {
                if (function->impl) {
@@ -140,7 +147,7 @@ ir3_nir_analyze_ubo_ranges(nir_shader *nir, struct ir3_shader *shader)
                                nir_foreach_instr(instr, block) {
                                        if (instr->type == nir_instr_type_intrinsic &&
                                                nir_instr_as_intrinsic(instr)->intrinsic == nir_intrinsic_load_ubo)
-                                               gather_ubo_ranges(nir_instr_as_intrinsic(instr), state);
+                                               gather_ubo_ranges(nir, nir_instr_as_intrinsic(instr), state);
                                }
                        }
                }