[Mesa-dev] [PATCH 16/21] i965/fs: Implement scratch reads and writes of 4 GRFs at a time.

[Francisco Jerez]

---
 src/mesa/drivers/dri/i965/brw_eu_emit.c           | 31 ++++++++++-------------
 src/mesa/drivers/dri/i965/brw_fs_generator.cpp    |  4 +++
 src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp |  4 +--
 3 files changed, 18 insertions(+), 21 deletions(-)

diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c
index 846fda0..af7caed 100644
--- a/src/mesa/drivers/dri/i965/brw_eu_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c
@@ -2065,21 +2065,19 @@ void brw_oword_block_write_scratch(struct brw_codegen *p,
 				   unsigned offset)
 {
    const struct brw_device_info *devinfo = p->devinfo;
-   uint32_t msg_control, msg_type;
-   int mlen;
+   uint32_t msg_type;
 
    if (devinfo->gen >= 6)
       offset /= 16;
 
    mrf = retype(mrf, BRW_REGISTER_TYPE_UD);
 
-   if (num_regs == 1) {
-      msg_control = BRW_DATAPORT_OWORD_BLOCK_2_OWORDS;
-      mlen = 2;
-   } else {
-      msg_control = BRW_DATAPORT_OWORD_BLOCK_4_OWORDS;
-      mlen = 3;
-   }
+   const unsigned mlen = 1 + num_regs;
+   const unsigned msg_control =
+      (num_regs == 1 ? BRW_DATAPORT_OWORD_BLOCK_2_OWORDS :
+       num_regs == 2 ? BRW_DATAPORT_OWORD_BLOCK_4_OWORDS :
+       num_regs == 4 ? BRW_DATAPORT_OWORD_BLOCK_8_OWORDS : 0);
+   assert(msg_control);
 
    /* Set up the message header.  This is g0, with g0.2 filled with
     * the offset.  We don't want to leave our offset around in g0 or
@@ -2180,8 +2178,6 @@ brw_oword_block_read_scratch(struct brw_codegen *p,
 			     unsigned offset)
 {
    const struct brw_device_info *devinfo = p->devinfo;
-   uint32_t msg_control;
-   int rlen;
 
    if (devinfo->gen >= 6)
       offset /= 16;
@@ -2200,13 +2196,12 @@ brw_oword_block_read_scratch(struct brw_codegen *p,
    }
    dest = retype(dest, BRW_REGISTER_TYPE_UW);
 
-   if (num_regs == 1) {
-      msg_control = BRW_DATAPORT_OWORD_BLOCK_2_OWORDS;
-      rlen = 1;
-   } else {
-      msg_control = BRW_DATAPORT_OWORD_BLOCK_4_OWORDS;
-      rlen = 2;
-   }
+   const unsigned rlen = num_regs;
+   const unsigned msg_control =
+      (num_regs == 1 ? BRW_DATAPORT_OWORD_BLOCK_2_OWORDS :
+       num_regs == 2 ? BRW_DATAPORT_OWORD_BLOCK_4_OWORDS :
+       num_regs == 4 ? BRW_DATAPORT_OWORD_BLOCK_8_OWORDS : 0);
+   assert(msg_control);
 
    {
       brw_push_insn_state(p);
diff --git a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
index 9751926..196ff07 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
@@ -1056,6 +1056,7 @@ fs_generator::generate_discard_jump(fs_inst *inst)
 void
 fs_generator::generate_scratch_write(fs_inst *inst, struct brw_reg src)
 {
+   assert(inst->exec_size <= 16 || inst->force_writemask_all);
    assert(inst->mlen != 0);
 
    brw_MOV(p,
@@ -1068,6 +1069,7 @@ fs_generator::generate_scratch_write(fs_inst *inst, struct brw_reg src)
 void
 fs_generator::generate_scratch_read(fs_inst *inst, struct brw_reg dst)
 {
+   assert(inst->exec_size <= 16 || inst->force_writemask_all);
    assert(inst->mlen != 0);
 
    brw_oword_block_read_scratch(p, dst, brw_message_reg(inst->base_mrf),
@@ -1077,6 +1079,8 @@ fs_generator::generate_scratch_read(fs_inst *inst, struct brw_reg dst)
 void
 fs_generator::generate_scratch_read_gen7(fs_inst *inst, struct brw_reg dst)
 {
+   assert(inst->exec_size <= 16 || inst->force_writemask_all);
+
    gen7_block_read_scratch(p, dst, inst->exec_size / 8, inst->offset);
 }
 
diff --git a/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp b/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp
index e770a7f..7fdedc0 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp
@@ -928,11 +928,9 @@ fs_visitor::spill_reg(int spill_reg)
             /* We read the largest power-of-two divisor of the register count
              * (because only POT scratch read blocks are allowed by the
              * hardware) up to the maximum supported block size.
-             * XXX - Bump the limit when the generator code is ready for
-             *       32-wide spills.
              */
             const unsigned width =
-               MIN2(16, 1u << (ffs(MAX2(1, regs_read) * 8) - 1));
+               MIN2(32, 1u << (ffs(MAX2(1, regs_read) * 8) - 1));
 
             /* Set exec_all() on unspill messages under the (rather
              * pessimistic) assumption that there is no one-to-one
-- 
2.7.3