vc4: Split two notions of instructions having side effects.

Some ops can't be DCEd, while some of the ops that are just important due
to the args they have can be.

5 files changed, 15 insertions, 4 deletions

diff --git a/src/gallium/drivers/vc4/vc4_opt_cse.c b/src/gallium/drivers/vc4/vc4_opt_cse.c
index 2ca736fe4d5..aff777beb47 100644
--- a/src/gallium/drivers/vc4/vc4_opt_cse.c
+++ b/src/gallium/drivers/vc4/vc4_opt_cse.c
@@ -133,7 +133,8 @@ qir_opt_cse(struct vc4_compile *c)
         foreach_s(node, t, &c->instructions) {
                 struct qinst *inst = (struct qinst *)node;
 
-                if (qir_has_side_effects(c, inst)) {
+                if (qir_has_side_effects(c, inst) ||
+                    qir_has_side_effect_reads(c, inst)) {
                         if (inst->op == QOP_TLB_DISCARD_SETUP)
                                 last_sf = NULL;
                         continue;
diff --git a/src/gallium/drivers/vc4/vc4_opt_dead_code.c b/src/gallium/drivers/vc4/vc4_opt_dead_code.c
index 408bd4302b4..f555fcb600e 100644
--- a/src/gallium/drivers/vc4/vc4_opt_dead_code.c
+++ b/src/gallium/drivers/vc4/vc4_opt_dead_code.c
@@ -64,7 +64,8 @@ qir_opt_dead_code(struct vc4_compile *c)
                 if (inst->dst.file == QFILE_TEMP &&
                     !used[inst->dst.index] &&
                     (!qir_has_side_effects(c, inst) ||
-                     inst->op == QOP_TEX_RESULT)) {
+                     inst->op == QOP_TEX_RESULT) &&
+                    !(qir_has_side_effect_reads(c, inst))) {
                         if (inst->op == QOP_TEX_RESULT) {
                                 dce_tex = true;
                                 c->num_texture_samples--;
diff --git a/src/gallium/drivers/vc4/vc4_opt_vpm_writes.c b/src/gallium/drivers/vc4/vc4_opt_vpm_writes.c
index 477d32605a4..0269e32494a 100644
--- a/src/gallium/drivers/vc4/vc4_opt_vpm_writes.c
+++ b/src/gallium/drivers/vc4/vc4_opt_vpm_writes.c
@@ -82,8 +82,10 @@ qir_opt_vpm_writes(struct vc4_compile *c)
                 if (qir_depends_on_flags(inst))
                         continue;
 
-                if (qir_has_side_effects(c, inst))
+                if (qir_has_side_effects(c, inst) ||
+                    qir_has_side_effect_reads(c, inst)) {
                         continue;
+                }
 
                 /* A QOP_TEX_RESULT destination is r4, so we can't move
                  * accesses to it past another QOP_TEX_RESULT which would
diff --git a/src/gallium/drivers/vc4/vc4_qir.c b/src/gallium/drivers/vc4/vc4_qir.c
index 4f73932e233..40356c99ff3 100644
--- a/src/gallium/drivers/vc4/vc4_qir.c
+++ b/src/gallium/drivers/vc4/vc4_qir.c
@@ -141,6 +141,12 @@ qir_get_op_nsrc(enum qop qop)
 bool
 qir_has_side_effects(struct vc4_compile *c, struct qinst *inst)
 {
+        return qir_op_info[inst->op].has_side_effects;
+}
+
+bool
+qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst)
+{
         /* We can dead-code eliminate varyings, because we only tell the VS
          * about the live ones at the end.  But we have to preserve the
          * point/line coordinates reads, because they're generated by
@@ -159,7 +165,7 @@ qir_has_side_effects(struct vc4_compile *c, struct qinst *inst)
         if (inst->dst.file == QFILE_VPM)
                 return true;
 
-        return qir_op_info[inst->op].has_side_effects;
+        return false;
 }
 
 bool
diff --git a/src/gallium/drivers/vc4/vc4_qir.h b/src/gallium/drivers/vc4/vc4_qir.h
index 0c2bca93d8b..ebec7ccfbe3 100644
--- a/src/gallium/drivers/vc4/vc4_qir.h
+++ b/src/gallium/drivers/vc4/vc4_qir.h
@@ -377,6 +377,7 @@ struct qreg qir_get_temp(struct vc4_compile *c);
 int qir_get_op_nsrc(enum qop qop);
 bool qir_reg_equals(struct qreg a, struct qreg b);
 bool qir_has_side_effects(struct vc4_compile *c, struct qinst *inst);
+bool qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst);
 bool qir_is_multi_instruction(struct qinst *inst);
 bool qir_depends_on_flags(struct qinst *inst);
 bool qir_writes_r4(struct qinst *inst);