i965/fs: Implement scratch reads and writes of 4 GRFs at a time.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c
index 6cbe673e94dc28d8662764e41c59b5478cd10ccf..9e544a9d18ab530a96df676450d4602024e8e619 100644 (file)
--- 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 4b4792f96e4a69dffc8db9146aa4bc6e6c07d648..7308fdd9986ae284db56066e2987f9997b94f67c 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
@@ -1061,6 +1061,7 @@ 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,
@@ -1073,6 +1074,7 @@ 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),
@@ -1082,6 +1084,8 @@ 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 e770a7f91cca6987e9aba07ebad4ada841e9499e..7fdedc084c5dc85ce1fcba508817b22a1add95bc 100644 (file)
--- 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