freedreno/ir3: use instr flag to mark unused instructions

Rather than magic depth value, which won't be available in later stages.

Signed-off-by: Rob Clark <robclark@freedesktop.org>

diff --git a/src/gallium/drivers/freedreno/ir3/ir3.h b/src/gallium/drivers/freedreno/ir3/ir3.h
index 12f2ebe18db175cee0975369008fbcd7af57ee9d..cac7d4d6e58286461049569f9f2509a6fdafc8fb 100644 (file)
--- a/src/gallium/drivers/freedreno/ir3/ir3.h
+++ b/src/gallium/drivers/freedreno/ir3/ir3.h
@@ -177,6 +177,7 @@ struct ir3_instruction {
                 * before register assignment is done:
                 */
                IR3_INSTR_MARK  = 0x1000,
+               IR3_INSTR_UNUSED= 0x2000,
        } flags;
        int repeat;
 #ifdef DEBUG
@@ -243,11 +244,7 @@ struct ir3_instruction {
                 * result of moving a const to a reg would have a low cost,  so to
                 * it could make sense to duplicate the instruction at various
                 * points where the result is needed to reduce register footprint.
-                *
-                * DEPTH_UNUSED used to mark unused instructions after depth
-                * calculation pass.
                 */
-#define DEPTH_UNUSED  ~0
                unsigned depth;
                /* When we get to the RA stage, we no longer need depth, but
                 * we do need instruction's position/name:

diff --git a/src/gallium/drivers/freedreno/ir3/ir3_depth.c b/src/gallium/drivers/freedreno/ir3/ir3_depth.c
index 97df0c2ac99fbf61d652c5ad789b8d36c52a83de..4bbc045879075e77b78a1a4b6703fc1b203dc646 100644 (file)
--- a/src/gallium/drivers/freedreno/ir3/ir3_depth.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_depth.c
@@ -139,7 +139,7 @@ remove_unused_by_block(struct ir3_block *block)
                        /* mark it, in case it is input, so we can
                         * remove unused inputs:
                         */
-                       instr->depth = DEPTH_UNUSED;
+                       instr->flags |= IR3_INSTR_UNUSED;
                        /* and remove from instruction list: */
                        list_delinit(&instr->node);
                }
@@ -175,14 +175,14 @@ ir3_depth(struct ir3 *ir)
         */
        for (i = 0; i < ir->indirects_count; i++) {
                struct ir3_instruction *instr = ir->indirects[i];
-               if (instr->depth == DEPTH_UNUSED)
+               if (instr->flags & IR3_INSTR_UNUSED)
                        ir->indirects[i] = NULL;
        }
 
        /* cleanup unused inputs: */
        for (i = 0; i < ir->ninputs; i++) {
                struct ir3_instruction *in = ir->inputs[i];
-               if (in && (in->depth == DEPTH_UNUSED))
+               if (in && (in->flags & IR3_INSTR_UNUSED))
                        ir->inputs[i] = NULL;
        }
 }

diff --git a/src/gallium/drivers/freedreno/ir3/ir3_ra.c b/src/gallium/drivers/freedreno/ir3/ir3_ra.c
index 88018398e968aa51ff4842eafbcb1dae43527bff..ae7282e0b5bc52ca0545837dbbb1412ffad1bc64 100644 (file)
--- a/src/gallium/drivers/freedreno/ir3/ir3_ra.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_ra.c
@@ -314,6 +314,14 @@ writes_gpr(struct ir3_instruction *instr)
        return is_temp(instr->regs[0]);
 }
 
+static bool
+instr_before(struct ir3_instruction *a, struct ir3_instruction *b)
+{
+       if (a->flags & IR3_INSTR_UNUSED)
+               return false;
+       return (a->ip < b->ip);
+}
+
 static struct ir3_instruction *
 get_definer(struct ir3_ra_ctx *ctx, struct ir3_instruction *instr,
                int *sz, int *off)
@@ -348,7 +356,7 @@ get_definer(struct ir3_ra_ctx *ctx, struct ir3_instruction *instr,
 
                        dd = get_definer(ctx, src->instr, &dsz, &doff);
 
-                       if ((!d) || (dd->ip < d->ip)) {
+                       if ((!d) || instr_before(dd, d)) {
                                d = dd;
                                *sz = dsz;
                                *off = doff - n;
@@ -369,9 +377,14 @@ get_definer(struct ir3_ra_ctx *ctx, struct ir3_instruction *instr,
                 */
                int cnt = 0;
 
-               d = f;
+               /* need to skip over unused in the group: */
+               while (f && (f->flags & IR3_INSTR_UNUSED)) {
+                       f = f->cp.right;
+                       cnt++;
+               }
+
                while (f) {
-                       if (f->ip < d->ip)
+                       if ((!d) || instr_before(f, d))
                                d = f;
                        if (f == instr)
                                *off = cnt;
@@ -414,7 +427,7 @@ get_definer(struct ir3_ra_ctx *ctx, struct ir3_instruction *instr,
                *sz = MAX2(*sz, dsz);
                *off = doff;
 
-               if (dd->ip < d->ip) {
+               if (instr_before(dd, d)) {
                        d = dd;
                }
        }
@@ -432,7 +445,7 @@ get_definer(struct ir3_ra_ctx *ctx, struct ir3_instruction *instr,
                foreach_src(src, d) {
                        if (!src->instr)
                                continue;
-                       if (src->instr->ip < dd->ip)
+                       if (instr_before(src->instr, dd))
                                dd = src->instr;
                }
 
@@ -446,7 +459,7 @@ get_definer(struct ir3_ra_ctx *ctx, struct ir3_instruction *instr,
                dd = get_definer(ctx, d->regs[1]->instr, &dsz, &doff);
 
                /* by definition, should come before: */
-               debug_assert(dd->ip < d->ip);
+               debug_assert(instr_before(dd, d));
 
                *sz = MAX2(*sz, dsz);
 

diff --git a/src/gallium/drivers/freedreno/ir3/ir3_sched.c b/src/gallium/drivers/freedreno/ir3/ir3_sched.c
index 2ee325518f729a3b459724a343aca7edd29e7a96..08f5cac0cf4135cffd7de7e4f340fb70e49c777c 100644 (file)
--- a/src/gallium/drivers/freedreno/ir3/ir3_sched.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_sched.c
@@ -246,7 +246,7 @@ instr_eligibility(struct ir3_sched_ctx *ctx, struct ir3_sched_notes *notes,
 
                for (unsigned i = 0; i < ir->baryfs_count; i++) {
                        struct ir3_instruction *baryf = ir->baryfs[i];
-                       if (baryf->depth == DEPTH_UNUSED)
+                       if (baryf->flags & IR3_INSTR_UNUSED)
                                continue;
                        if (!is_scheduled(baryf)) {
                                notes->blocked_kill = true;