i965: avoid memsetting all the BRW_WM_MAX_INSN arrays for every compile.

For an app that's blowing out the state cache, like sauerbraten, the
memset of the giant arrays ended up taking 11% of the CPU even when only a
"few" of the entries got used.  With this, the WM program compile drops back
down to 1% of CPU time.

Bug #24981 (bisected to BRW_WM_MAX_INSN increase).

diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c
index 964ee10..d897132 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_wm.c
+++ b/src/mesa/drivers/dri/i965/brw_wm.c
@@ -152,8 +152,22 @@ static void do_wm_prog( struct brw_context *brw,
           */
          return;
       }
+      c->instruction = _mesa_calloc(BRW_WM_MAX_INSN * sizeof(*c->instruction));
+      c->prog_instructions = _mesa_calloc(BRW_WM_MAX_INSN *
+                                         sizeof(*c->prog_instructions));
+      c->vreg = _mesa_calloc(BRW_WM_MAX_VREG * sizeof(*c->vreg));
+      c->refs = _mesa_calloc(BRW_WM_MAX_REF * sizeof(*c->refs));
+      c->vreg = _mesa_calloc(BRW_WM_MAX_VREG * sizeof(*c->vreg));
    } else {
+      void *instruction = c->instruction;
+      void *prog_instructions = c->prog_instructions;
+      void *vreg = c->vreg;
+      void *refs = c->refs;
       memset(c, 0, sizeof(*brw->wm.compile_data));
+      c->instruction = instruction;
+      c->prog_instructions = prog_instructions;
+      c->vreg = vreg;
+      c->refs = refs;
    }
    memcpy(&c->key, key, sizeof(*key));
 

diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h
index 1fa2f1c..7db212e 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_wm.h
+++ b/src/mesa/drivers/dri/i965/brw_wm.h
@@ -202,7 +202,7 @@ struct brw_wm_compile {
     * simplifying and adding instructions for interpolation and
     * framebuffer writes.
     */
-   struct prog_instruction prog_instructions[BRW_WM_MAX_INSN];
+   struct prog_instruction *prog_instructions;
    GLuint nr_fp_insns;
    GLuint fp_temp;
    GLuint fp_interp_emitted;
@@ -213,7 +213,7 @@ struct brw_wm_compile {
    struct prog_src_register pixel_w;
 
 
-   struct brw_wm_value vreg[BRW_WM_MAX_VREG];
+   struct brw_wm_value *vreg;
    GLuint nr_vreg;
 
    struct brw_wm_value creg[BRW_WM_MAX_PARAM];
@@ -230,10 +230,10 @@ struct brw_wm_compile {
    struct brw_wm_ref undef_ref;
    struct brw_wm_value undef_value;
 
-   struct brw_wm_ref refs[BRW_WM_MAX_REF];
+   struct brw_wm_ref *refs;
    GLuint nr_refs;
 
-   struct brw_wm_instruction instruction[BRW_WM_MAX_INSN];
+   struct brw_wm_instruction *instruction;
    GLuint nr_insns;
 
    struct brw_wm_constref constref[BRW_WM_MAX_CONST];

diff --git a/src/mesa/drivers/dri/i965/brw_wm_fp.c b/src/mesa/drivers/dri/i965/brw_wm_fp.c
index 549afd3..1c4f62b 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_wm_fp.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_fp.c
@@ -182,6 +182,8 @@ static void release_temp( struct brw_wm_compile *c, struct prog_dst_register tem
 static struct prog_instruction *get_fp_inst(struct brw_wm_compile *c)
 {
    assert(c->nr_fp_insns < BRW_WM_MAX_INSN);
+   memset(&c->prog_instructions[c->nr_fp_insns], 0,
+         sizeof(*c->prog_instructions));
    return &c->prog_instructions[c->nr_fp_insns++];
 }
 

diff --git a/src/mesa/drivers/dri/i965/brw_wm_pass0.c b/src/mesa/drivers/dri/i965/brw_wm_pass0.c
index 602b135..ff4c082 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_wm_pass0.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_pass0.c
@@ -42,12 +42,14 @@
 static struct brw_wm_ref *get_ref( struct brw_wm_compile *c )
 {
    assert(c->nr_refs < BRW_WM_MAX_REF);
+   memset(&c->refs[c->nr_refs], 0, sizeof(*c->refs));
    return &c->refs[c->nr_refs++];
 }
 
 static struct brw_wm_value *get_value( struct brw_wm_compile *c)
 {
    assert(c->nr_refs < BRW_WM_MAX_VREG);
+   memset(&c->vreg[c->nr_vreg], 0, sizeof(*c->vreg));
    return &c->vreg[c->nr_vreg++];
 }
 
@@ -55,6 +57,7 @@ static struct brw_wm_value *get_value( struct brw_wm_compile *c)
 static struct brw_wm_instruction *get_instruction( struct brw_wm_compile *c )
 {
    assert(c->nr_insns < BRW_WM_MAX_INSN);
+   memset(&c->instruction[c->nr_insns], 0, sizeof(*c->instruction));
    return &c->instruction[c->nr_insns++];
 }