intel/eu: Add some new helpers

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6244>

diff --git a/src/intel/compiler/brw_eu.h b/src/intel/compiler/brw_eu.h
index c4b44f3ced34fb5e99084c7ce6eac28ce4ed0f2c..afb7cd537a235d9fa521b7f24fe5324d328e43b4 100644 (file)
--- a/src/intel/compiler/brw_eu.h
+++ b/src/intel/compiler/brw_eu.h
@@ -188,6 +188,9 @@ const unsigned *brw_get_program( struct brw_codegen *p, unsigned *sz );
 bool brw_try_override_assembly(struct brw_codegen *p, int start_offset,
                                const char *identifier);
 
+void brw_realign(struct brw_codegen *p, unsigned align);
+int brw_append_data(struct brw_codegen *p, void *data,
+                    unsigned size, unsigned align);
 brw_inst *brw_next_insn(struct brw_codegen *p, unsigned opcode);
 void brw_set_dest(struct brw_codegen *p, brw_inst *insn, struct brw_reg dest);
 void brw_set_src0(struct brw_codegen *p, brw_inst *insn, struct brw_reg reg);

diff --git a/src/intel/compiler/brw_eu_emit.c b/src/intel/compiler/brw_eu_emit.c
index 08d64800f7a8070b85636e858871f97c136fba5d..e6c6dcb1607aaad1f418a668eacef3ac2ffe0893 100644 (file)
--- a/src/intel/compiler/brw_eu_emit.c
+++ b/src/intel/compiler/brw_eu_emit.c
@@ -640,20 +640,62 @@ brw_inst_set_state(const struct gen_device_info *devinfo,
       brw_inst_set_acc_wr_control(devinfo, insn, state->acc_wr_control);
 }
 
-#define next_insn brw_next_insn
-brw_inst *
-brw_next_insn(struct brw_codegen *p, unsigned opcode)
+static brw_inst *
+brw_append_insns(struct brw_codegen *p, unsigned nr_insn, unsigned align)
 {
-   const struct gen_device_info *devinfo = p->devinfo;
-   brw_inst *insn;
+   assert(util_is_power_of_two_or_zero(sizeof(brw_inst)));
+   assert(util_is_power_of_two_or_zero(align));
+   const unsigned align_insn = MAX2(align / sizeof(brw_inst), 1);
+   const unsigned start_insn = ALIGN(p->nr_insn, align_insn);
+   const unsigned new_nr_insn = start_insn + nr_insn;
 
-   if (p->nr_insn + 1 > p->store_size) {
-      p->store_size <<= 1;
+   if (p->store_size < new_nr_insn) {
+      p->store_size = util_next_power_of_two(new_nr_insn * sizeof(brw_inst));
       p->store = reralloc(p->mem_ctx, p->store, brw_inst, p->store_size);
    }
 
-   p->next_insn_offset += 16;
-   insn = &p->store[p->nr_insn++];
+   /* Memset any padding due to alignment to 0.  We don't want to be hashing
+    * or caching a bunch of random bits we got from a memory allocation.
+    */
+   if (p->nr_insn < start_insn) {
+      memset(&p->store[p->nr_insn], 0,
+             (start_insn - p->nr_insn) * sizeof(brw_inst));
+   }
+
+   assert(p->next_insn_offset == p->nr_insn * sizeof(brw_inst));
+   p->nr_insn = new_nr_insn;
+   p->next_insn_offset = new_nr_insn * sizeof(brw_inst);
+
+   return &p->store[start_insn];
+}
+
+void
+brw_realign(struct brw_codegen *p, unsigned align)
+{
+   brw_append_insns(p, 0, align);
+}
+
+int
+brw_append_data(struct brw_codegen *p, void *data,
+                unsigned size, unsigned align)
+{
+   unsigned nr_insn = DIV_ROUND_UP(size, sizeof(brw_inst));
+   void *dst = brw_append_insns(p, nr_insn, align);
+   memcpy(dst, data, size);
+
+   /* If it's not a whole number of instructions, memset the end */
+   if (size < nr_insn * sizeof(brw_inst))
+      memset(dst + size, 0, nr_insn * sizeof(brw_inst) - size);
+
+   return dst - (void *)p->store;
+}
+
+#define next_insn brw_next_insn
+brw_inst *
+brw_next_insn(struct brw_codegen *p, unsigned opcode)
+{
+   const struct gen_device_info *devinfo = p->devinfo;
+   brw_inst *insn = brw_append_insns(p, 1, sizeof(brw_inst));
 
    memset(insn, 0, sizeof(*insn));
    brw_inst_set_opcode(devinfo, insn, opcode);

diff --git a/src/intel/compiler/brw_fs_generator.cpp b/src/intel/compiler/brw_fs_generator.cpp
index 94eb84870f1200c4d4a26200f41425c6f1ee6956..df357440d495cad69e3cb4c52094635cf6089c81 100644 (file)
--- a/src/intel/compiler/brw_fs_generator.cpp
+++ b/src/intel/compiler/brw_fs_generator.cpp
@@ -1783,8 +1783,7 @@ fs_generator::generate_code(const cfg_t *cfg, int dispatch_width,
                             struct brw_compile_stats *stats)
 {
    /* align to 64 byte boundary. */
-   while (p->next_insn_offset % 64)
-      brw_NOP(p);
+   brw_realign(p, 64);
 
    this->dispatch_width = dispatch_width;