vc4: Fix live intervals analysis for screening defs in if statements.

If a conditional assignment is only conditioned on the exec mask, that's
still screening off the value in the executed channels (and, since we're
not storing to the unexcuted channels, we don't care what's in there).

Fixes a bunch of extra register pressure on Processing's Ribbons demo,
which is failing to allocate.

diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index a91e6200e8a7225d7ad29cfbdc170772c4e7736f..81ac070d4631cd2f0c863c2f1749a5dd572bdc7f 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -169,8 +169,11 @@ ntq_store_dest(struct vc4_compile *c, nir_dest *dest, int chan,
                  * channel is active.
                  */
                 if (c->execute.file != QFILE_NULL) {
+                        struct qinst *mov;
+
                         qir_SF(c, c->execute);
-                        qir_MOV_cond(c, QPU_COND_ZS, qregs[chan], result);
+                        mov = qir_MOV_cond(c, QPU_COND_ZS, qregs[chan], result);
+                        mov->cond_is_exec_mask = true;
                 } else {
                         qir_MOV_dest(c, qregs[chan], result);
                 }

diff --git a/src/gallium/drivers/vc4/vc4_qir.h b/src/gallium/drivers/vc4/vc4_qir.h
index 366521675951a5a6d7b75e521d0a89293cc03121..4d41c427c1032df38a99b5f9a74a4f15030db4c1 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_qir.h
+++ b/src/gallium/drivers/vc4/vc4_qir.h
@@ -195,6 +195,7 @@ struct qinst {
         struct qreg dst;
         struct qreg *src;
         bool sf;
+        bool cond_is_exec_mask;
         uint8_t cond;
 };
 
@@ -835,11 +836,13 @@ qir_ROT_MUL(struct vc4_compile *c, struct qreg val, uint32_t rot)
                                                 QPU_SMALL_IMM_MUL_ROT + rot)));
 }
 
-static inline void
+static inline struct qinst *
 qir_MOV_cond(struct vc4_compile *c, uint8_t cond,
              struct qreg dest, struct qreg src)
 {
-        qir_MOV_dest(c, dest, src)->cond = cond;
+        struct qinst *mov = qir_MOV_dest(c, dest, src);
+        mov->cond = cond;
+        return mov;
 }
 
 static inline struct qinst *

diff --git a/src/gallium/drivers/vc4/vc4_qir_live_variables.c b/src/gallium/drivers/vc4/vc4_qir_live_variables.c
index eac350ab85216e03d2d41238e3f6e28519ca82d8..beefb0d7f8ad75b2d58b09137693e6c554ae4ccf 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_qir_live_variables.c
+++ b/src/gallium/drivers/vc4/vc4_qir_live_variables.c
@@ -113,8 +113,17 @@ qir_setup_def(struct vc4_compile *c, struct qblock *block, int ip,
         if (BITSET_TEST(block->use, var) || BITSET_TEST(block->def, var))
                 return;
 
-        /* Easy, common case: unconditional full register update. */
-        if (inst->cond == QPU_COND_ALWAYS && !inst->dst.pack) {
+        /* Easy, common case: unconditional full register update.
+         *
+         * We treat conditioning on the exec mask as the same as not being
+         * conditional.  This makes sure that if the register gets set on
+         * either side of an if, it is treated as being screened off before
+         * the if.  Otherwise, if there was no intervening def, its live
+         * interval doesn't extend back to the start of he program, and if too
+         * many registers did that we'd fail to register allocate.
+         */
+        if ((inst->cond == QPU_COND_ALWAYS ||
+             inst->cond_is_exec_mask) && !inst->dst.pack) {
                 BITSET_SET(block->def, var);
                 return;
         }