vc4: Refactor qir_get_op_nsrc(enum qop) to qir_get_nsrc(struct qinst *).

Every caller was dereffing the qinst, and this will let us make the number
of sources vary depending on the destination of the qinst so that we can
have general ALU ops that store to tex_[strb] and get an implicit uniform.

diff --git a/src/gallium/drivers/vc4/vc4_opt_constant_folding.c b/src/gallium/drivers/vc4/vc4_opt_constant_folding.c
index 7ff91615545411214b239097e26f43b890e5742a..de642d46582211c11607efdea03f39a201b907e3 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_opt_constant_folding.c
+++ b/src/gallium/drivers/vc4/vc4_opt_constant_folding.c
@@ -58,7 +58,7 @@ dump_to(struct vc4_compile *c, struct qinst *inst)
 static bool
 constant_fold(struct vc4_compile *c, struct qinst *inst)
 {
-        int nsrc = qir_get_op_nsrc(inst->op);
+        int nsrc = qir_get_nsrc(inst);
         uint32_t ui[nsrc];
 
         for (int i = 0; i < nsrc; i++) {

diff --git a/src/gallium/drivers/vc4/vc4_opt_copy_propagation.c b/src/gallium/drivers/vc4/vc4_opt_copy_propagation.c
index d20ee5e227d5f4195db3c0eba9079918e1eb57c5..9a6320a9a203a924bc0def40afac450655d065e7 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_opt_copy_propagation.c
+++ b/src/gallium/drivers/vc4/vc4_opt_copy_propagation.c
@@ -67,7 +67,7 @@ try_copy_prop(struct vc4_compile *c, struct qinst *inst, struct qinst **movs)
         bool debug = false;
         bool progress = false;
 
-       for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+       for (int i = 0; i < qir_get_nsrc(inst); i++) {
                 if (inst->src[i].file != QFILE_TEMP)
                         continue;
 
@@ -113,7 +113,7 @@ try_copy_prop(struct vc4_compile *c, struct qinst *inst, struct qinst **movs)
                          * this instruction doesn't already use it.
                          */
                         bool already_has_unpack = false;
-                        for (int j = 0; j < qir_get_op_nsrc(inst->op); j++) {
+                        for (int j = 0; j < qir_get_nsrc(inst); j++) {
                                 if (inst->src[j].pack)
                                         already_has_unpack = true;
                         }

diff --git a/src/gallium/drivers/vc4/vc4_opt_dead_code.c b/src/gallium/drivers/vc4/vc4_opt_dead_code.c
index 1838c394ff91aa0f68a8d6c8ffcfab36c4378f17..f04d0ff97ab7cbb2fd27e7de883d98602e79de8b 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_opt_dead_code.c
+++ b/src/gallium/drivers/vc4/vc4_opt_dead_code.c
@@ -54,7 +54,7 @@ dce(struct vc4_compile *c, struct qinst *inst)
 static bool
 has_nonremovable_reads(struct vc4_compile *c, struct qinst *inst)
 {
-        for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+        for (int i = 0; i < qir_get_nsrc(inst); i++) {
                 if (inst->src[i].file == QFILE_VPM) {
                         uint32_t attr = inst->src[i].index / 4;
                         uint32_t offset = (inst->src[i].index % 4) * 4;
@@ -88,7 +88,7 @@ qir_opt_dead_code(struct vc4_compile *c)
         bool *used = calloc(c->num_temps, sizeof(bool));
 
         qir_for_each_inst_inorder(inst, c) {
-                for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+                for (int i = 0; i < qir_get_nsrc(inst); i++) {
                         if (inst->src[i].file == QFILE_TEMP)
                                 used[inst->src[i].index] = true;
                 }
@@ -129,7 +129,7 @@ qir_opt_dead_code(struct vc4_compile *c)
                                 continue;
                         }
 
-                        for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+                        for (int i = 0; i < qir_get_nsrc(inst); i++) {
                                 if (inst->src[i].file != QFILE_VPM)
                                         continue;
                                 uint32_t attr = inst->src[i].index / 4;

diff --git a/src/gallium/drivers/vc4/vc4_opt_peephole_sf.c b/src/gallium/drivers/vc4/vc4_opt_peephole_sf.c
index f4856673ba24e06fefa6631c9860a6b6ceaf116d..577290b1fc46649caf264b753ec8ce1b1ad26c50 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_opt_peephole_sf.c
+++ b/src/gallium/drivers/vc4/vc4_opt_peephole_sf.c
@@ -62,7 +62,7 @@ inst_srcs_updated(struct qinst *inst, struct qinst *writer)
          */
         switch (writer->dst.file) {
         case QFILE_TEMP:
-                for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+                for (int i = 0; i < qir_get_nsrc(inst); i++) {
                         if (inst->src[i].file == QFILE_TEMP &&
                             inst->src[i].index == writer->dst.index) {
                                 return true;
@@ -95,7 +95,7 @@ inst_result_equals(struct qinst *a, struct qinst *b)
                 return false;
         }
 
-        for (int i = 0; i < qir_get_op_nsrc(a->op); i++) {
+        for (int i = 0; i < qir_get_nsrc(a); i++) {
                 if (!qir_reg_equals(a->src[i], b->src[i]) ||
                     src_file_varies_on_reread(a->src[i]) ||
                     src_file_varies_on_reread(b->src[i])) {

diff --git a/src/gallium/drivers/vc4/vc4_opt_small_immediates.c b/src/gallium/drivers/vc4/vc4_opt_small_immediates.c
index 4c105f37344253dd0f6cafc4cbba730f4b6e4cc5..15cbd12773fb883892609ffda500b5f3624e516f 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_opt_small_immediates.c
+++ b/src/gallium/drivers/vc4/vc4_opt_small_immediates.c
@@ -45,14 +45,14 @@ qir_opt_small_immediates(struct vc4_compile *c)
                  * elsewhere).
                  */
                 bool uses_small_imm = false;
-                for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+                for (int i = 0; i < qir_get_nsrc(inst); i++) {
                         if (inst->src[i].file == QFILE_SMALL_IMM)
                                 uses_small_imm = true;
                 }
                 if (uses_small_imm)
                         continue;
 
-                for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+                for (int i = 0; i < qir_get_nsrc(inst); i++) {
                         struct qreg src = qir_follow_movs(c, inst->src[i]);
 
                         if (src.file != QFILE_UNIF ||

diff --git a/src/gallium/drivers/vc4/vc4_opt_vpm.c b/src/gallium/drivers/vc4/vc4_opt_vpm.c
index 83ba11b817f753634777104da3d65e5b91ea058f..b3bef27225423a968145b3d0a29d81ed18f48a51 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_opt_vpm.c
+++ b/src/gallium/drivers/vc4/vc4_opt_vpm.c
@@ -58,7 +58,7 @@ qir_opt_vpm(struct vc4_compile *c)
                         break;
                 }
 
-                for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+                for (int i = 0; i < qir_get_nsrc(inst); i++) {
                         if (inst->src[i].file == QFILE_TEMP) {
                                 uint32_t temp = inst->src[i].index;
                                 use_count[temp]++;
@@ -81,7 +81,7 @@ qir_opt_vpm(struct vc4_compile *c)
                     qir_is_tex(inst))
                         continue;
 
-                for (int j = 0; j < qir_get_op_nsrc(inst->op); j++) {
+                for (int j = 0; j < qir_get_nsrc(inst); j++) {
                         if (inst->src[j].file != QFILE_TEMP ||
                             inst->src[j].pack)
                                 continue;
@@ -106,7 +106,7 @@ qir_opt_vpm(struct vc4_compile *c)
                         }
 
                         uint32_t temps = 0;
-                        for (int k = 0; k < qir_get_op_nsrc(inst->op); k++) {
+                        for (int k = 0; k < qir_get_nsrc(inst); k++) {
                                 if (inst->src[k].file == QFILE_TEMP)
                                         temps++;
                         }

diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index f8c7c4586223650b65348ff81293c641ab190965..97cbabbd511b57d72f256d096c8c3277136f8d39 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -2441,7 +2441,7 @@ vc4_setup_compiled_fs_inputs(struct vc4_context *vc4, struct vc4_compile *c,
 
         memset(input_live, 0, sizeof(input_live));
         qir_for_each_inst_inorder(inst, c) {
-                for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+                for (int i = 0; i < qir_get_nsrc(inst); i++) {
                         if (inst->src[i].file == QFILE_VARY)
                                 input_live[inst->src[i].index] = true;
                 }

diff --git a/src/gallium/drivers/vc4/vc4_qir.c b/src/gallium/drivers/vc4/vc4_qir.c
index 8bd016c5535496b0ec6f71e5ffbdfcca356dc6b5..2c9119d9ccf4c72d9533b139b993ae7b173935b6 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_qir.c
+++ b/src/gallium/drivers/vc4/vc4_qir.c
@@ -104,12 +104,10 @@ qir_get_op_name(enum qop qop)
 }
 
 int
-qir_get_op_nsrc(enum qop qop)
+qir_get_nsrc(struct qinst *inst)
 {
-        if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
-                return qir_op_info[qop].nsrc;
-        else
-                abort();
+        assert(qir_op_info[inst->op].name);
+        return qir_op_info[inst->op].nsrc;
 }
 
 /**
@@ -140,7 +138,7 @@ qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst)
          * point/line coordinates reads, because they're generated by
          * fixed-function hardware.
          */
-        for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+        for (int i = 0; i < qir_get_nsrc(inst); i++) {
                 if (inst->src[i].file == QFILE_VARY &&
                     c->input_slots[inst->src[i].index].slot == 0xff) {
                         return true;
@@ -372,7 +370,7 @@ qir_dump_inst(struct vc4_compile *c, struct qinst *inst)
                 }
         }
 
-        for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+        for (int i = 0; i < qir_get_nsrc(inst); i++) {
                 fprintf(stderr, ", ");
                 qir_print_reg(c, inst->src[i], false);
                 vc4_qpu_disasm_unpack(stderr, inst->src[i].pack);

diff --git a/src/gallium/drivers/vc4/vc4_qir.h b/src/gallium/drivers/vc4/vc4_qir.h
index eec50c3439eec821e03b40b7a7272f2a00e0d2b7..a3b8762951d78d8dcbe87cdb5a0df9f00f5de6e9 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_qir.h
+++ b/src/gallium/drivers/vc4/vc4_qir.h
@@ -577,7 +577,7 @@ struct qinst *qir_emit_nondef(struct vc4_compile *c, struct qinst *inst);
 
 struct qreg qir_get_temp(struct vc4_compile *c);
 void qir_calculate_live_intervals(struct vc4_compile *c);
-int qir_get_op_nsrc(enum qop qop);
+int qir_get_nsrc(struct qinst *inst);
 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);

diff --git a/src/gallium/drivers/vc4/vc4_qir_emit_uniform_stream_resets.c b/src/gallium/drivers/vc4/vc4_qir_emit_uniform_stream_resets.c
index 3fd6358e3d382b9d71a4a81a4fefd4edceb59a57..23ae8ebfa6f49d12d5832447130f936658678c9e 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_qir_emit_uniform_stream_resets.c
+++ b/src/gallium/drivers/vc4/vc4_qir_emit_uniform_stream_resets.c
@@ -41,7 +41,7 @@ inst_reads_a_uniform(struct qinst *inst)
         if (qir_is_tex(inst))
                 return true;
 
-        for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+        for (int i = 0; i < qir_get_nsrc(inst); i++) {
                 if (inst->src[i].file == QFILE_UNIF)
                         return true;
         }

diff --git a/src/gallium/drivers/vc4/vc4_qir_live_variables.c b/src/gallium/drivers/vc4/vc4_qir_live_variables.c
index dc058f5b4a7c19619d4e33f12ec639c2660597b9..330e1c8f7a9d920bccc2d046e49522f1ddcab13a 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_qir_live_variables.c
+++ b/src/gallium/drivers/vc4/vc4_qir_live_variables.c
@@ -205,7 +205,7 @@ qir_setup_def_use(struct vc4_compile *c)
                 _mesa_hash_table_clear(partial_update_ht, NULL);
 
                 qir_for_each_inst(inst, block) {
-                        for (int i = 0; i < qir_get_op_nsrc(inst->op); i++)
+                        for (int i = 0; i < qir_get_nsrc(inst); i++)
                                 qir_setup_use(c, block, ip, inst->src[i]);
 
                         qir_setup_def(c, block, ip, partial_update_ht, inst);

diff --git a/src/gallium/drivers/vc4/vc4_qir_lower_uniforms.c b/src/gallium/drivers/vc4/vc4_qir_lower_uniforms.c
index 8ec6c79739670c8e28b6e47cad51eb01069c7b44..1884cfa5b784c1f015c35d1b3953be5e06d5954c 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_qir_lower_uniforms.c
+++ b/src/gallium/drivers/vc4/vc4_qir_lower_uniforms.c
@@ -89,7 +89,7 @@ qir_get_instruction_uniform_count(struct qinst *inst)
 {
         uint32_t count = 0;
 
-        for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+        for (int i = 0; i < qir_get_nsrc(inst); i++) {
                 if (inst->src[i].file != QFILE_UNIF)
                         continue;
 
@@ -119,7 +119,7 @@ qir_lower_uniforms(struct vc4_compile *c)
          * ht.
          */
         qir_for_each_inst_inorder(inst, c) {
-                uint32_t nsrc = qir_get_op_nsrc(inst->op);
+                uint32_t nsrc = qir_get_nsrc(inst);
 
                 if (qir_get_instruction_uniform_count(inst) <= 1)
                         continue;
@@ -155,7 +155,7 @@ qir_lower_uniforms(struct vc4_compile *c)
                         struct qinst *mov = NULL;
 
                         qir_for_each_inst(inst, block) {
-                                uint32_t nsrc = qir_get_op_nsrc(inst->op);
+                                uint32_t nsrc = qir_get_nsrc(inst);
 
                                 uint32_t count = qir_get_instruction_uniform_count(inst);
 

diff --git a/src/gallium/drivers/vc4/vc4_qir_schedule.c b/src/gallium/drivers/vc4/vc4_qir_schedule.c
index 6eaa2e075acb2064b237008984b8e71074066c49..c1a2db5e3c5fdb226113be1b8ad7d1d843622cbb 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_qir_schedule.c
+++ b/src/gallium/drivers/vc4/vc4_qir_schedule.c
@@ -187,7 +187,7 @@ calculate_deps(struct schedule_setup_state *state, struct schedule_node *n)
          * ignore uniforms accesses, because qir_reorder_uniforms() happens
          * after this.
          */
-        for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+        for (int i = 0; i < qir_get_nsrc(inst); i++) {
                 switch (inst->src[i].file) {
                 case QFILE_TEMP:
                         add_dep(dir,
@@ -305,7 +305,7 @@ calculate_forward_deps(struct vc4_compile *c, void *mem_ctx,
 
                 calculate_deps(&state, n);
 
-                for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+                for (int i = 0; i < qir_get_nsrc(inst); i++) {
                         switch (inst->src[i].file) {
                         case QFILE_UNIF:
                                 add_dep(state.dir, state.last_uniforms_reset, n);
@@ -429,7 +429,7 @@ get_register_pressure_cost(struct schedule_state *state, struct qinst *inst)
             state->temp_writes[inst->dst.index] == 1)
                 cost--;
 
-        for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+        for (int i = 0; i < qir_get_nsrc(inst); i++) {
                 if (inst->src[i].file == QFILE_TEMP &&
                     !BITSET_TEST(state->temp_live, inst->src[i].index)) {
                         cost++;
@@ -648,7 +648,7 @@ schedule_instructions(struct vc4_compile *c,
                 }
 
                 /* Update our tracking of register pressure. */
-                for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+                for (int i = 0; i < qir_get_nsrc(inst); i++) {
                         if (inst->src[i].file == QFILE_TEMP)
                                 BITSET_SET(state->temp_live, inst->src[i].index);
                 }

diff --git a/src/gallium/drivers/vc4/vc4_qir_validate.c b/src/gallium/drivers/vc4/vc4_qir_validate.c
index e7cfe5ad217368bb2f237b86578ecd2590bdf17d..9579f7a15cbdf4f2e8f74fd28281cb86c17ff00a 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_qir_validate.c
+++ b/src/gallium/drivers/vc4/vc4_qir_validate.c
@@ -86,7 +86,7 @@ void qir_validate(struct vc4_compile *c)
                         break;
                 }
 
-                for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+                for (int i = 0; i < qir_get_nsrc(inst); i++) {
                         struct qreg src = inst->src[i];
 
                         switch (src.file) {

diff --git a/src/gallium/drivers/vc4/vc4_qpu_emit.c b/src/gallium/drivers/vc4/vc4_qpu_emit.c
index 2cc0f3013efb6d587416bd4c407844b71bc79cf0..9d9e5d84ecd9178dbbe537a93508dabc5fddb83a 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_qpu_emit.c
+++ b/src/gallium/drivers/vc4/vc4_qpu_emit.c
@@ -289,7 +289,7 @@ vc4_generate_code_block(struct vc4_compile *c,
 
                 uint64_t unpack = 0;
                 struct qpu_reg src[ARRAY_SIZE(qinst->src)];
-                for (int i = 0; i < qir_get_op_nsrc(qinst->op); i++) {
+                for (int i = 0; i < qir_get_nsrc(qinst); i++) {
                         int index = qinst->src[i].index;
                         switch (qinst->src[i].file) {
                         case QFILE_NULL:
@@ -538,7 +538,7 @@ vc4_generate_code_block(struct vc4_compile *c,
                          * argument slot as well so that we don't take up
                          * another raddr just to get unused data.
                          */
-                        if (qir_get_op_nsrc(qinst->op) == 1)
+                        if (qir_get_nsrc(qinst) == 1)
                                 src[1] = src[0];
 
                         fixup_raddr_conflict(block, dst, &src[0], &src[1],

diff --git a/src/gallium/drivers/vc4/vc4_register_allocate.c b/src/gallium/drivers/vc4/vc4_register_allocate.c
index 2992a6be2f2ce3a2193c4c358f01561b98e10a63..e48b8ee599230314fd52bc869394fbe01ab9dfe1 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_register_allocate.c
+++ b/src/gallium/drivers/vc4/vc4_register_allocate.c
@@ -306,7 +306,7 @@ vc4_register_allocate(struct vc4_context *vc4, struct vc4_compile *c)
                  * can only be done from regfile A, while float unpacks can be
                  * either A or R4.
                  */
-                for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+                for (int i = 0; i < qir_get_nsrc(inst); i++) {
                         if (inst->src[i].file == QFILE_TEMP &&
                             inst->src[i].pack) {
                                 if (qir_is_float_input(inst)) {

diff --git a/src/gallium/drivers/vc4/vc4_reorder_uniforms.c b/src/gallium/drivers/vc4/vc4_reorder_uniforms.c
index 7d5076f429e6f66fd74fbcab33ce3dfb14a523cf..37acefdc0ba9388e2605aa11eebc4012fa8637e4 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_reorder_uniforms.c
+++ b/src/gallium/drivers/vc4/vc4_reorder_uniforms.c
@@ -46,7 +46,7 @@ qir_reorder_uniforms(struct vc4_compile *c)
         qir_for_each_inst_inorder(inst, c) {
                 uint32_t new = ~0;
 
-                for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+                for (int i = 0; i < qir_get_nsrc(inst); i++) {
                         if (inst->src[i].file != QFILE_UNIF)
                                 continue;