i965/fs: Expose arbitrary pull constant load sizes to the IR.

Change the FS generator to ask the dataport for enough owords worth of
constants to fill the execution size of the instruction -- Which means
that the visitor now needs to set the execution size correctly for
uniform pull constant load instructions, which we were kind of
neglecting until now.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

4 files changed, 26 insertions, 27 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c
index 6141bfb99e..8536a13072 100644
--- a/src/mesa/drivers/dri/i965/brw_eu_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c
@@ -2256,7 +2256,7 @@ gen7_block_read_scratch(struct brw_codegen *p,
 }
 
 /**
- * Read a float[4] vector from the data port constant cache.
+ * Read float[4] vectors from the data port constant cache.
  * Location (in buffer) should be a multiple of 16.
  * Used for fetching shader constants.
  */
@@ -2270,6 +2270,7 @@ void brw_oword_block_read(struct brw_codegen *p,
    const unsigned target_cache =
       (devinfo->gen >= 6 ? GEN6_SFID_DATAPORT_CONSTANT_CACHE :
        BRW_DATAPORT_READ_TARGET_DATA_CACHE);
+   const unsigned exec_size = 1 << brw_inst_exec_size(devinfo, p->current);
 
    /* On newer hardware, offset is in units of owords. */
    if (devinfo->gen >= 6)
@@ -2278,11 +2279,12 @@ void brw_oword_block_read(struct brw_codegen *p,
    mrf = retype(mrf, BRW_REGISTER_TYPE_UD);
 
    brw_push_insn_state(p);
-   brw_set_default_exec_size(p, BRW_EXECUTE_8);
    brw_set_default_predicate_control(p, BRW_PREDICATE_NONE);
    brw_set_default_compression_control(p, BRW_COMPRESSION_NONE);
    brw_set_default_mask_control(p, BRW_MASK_DISABLE);
 
+   brw_push_insn_state(p);
+   brw_set_default_exec_size(p, BRW_EXECUTE_8);
    brw_MOV(p, mrf, retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD));
 
    /* set message header global offset field (reg 0, element 2) */
@@ -2291,6 +2293,7 @@ void brw_oword_block_read(struct brw_codegen *p,
 			       mrf.nr,
 			       2), BRW_REGISTER_TYPE_UD),
 	   brw_imm_ud(offset));
+   brw_pop_insn_state(p);
 
    brw_inst *insn = next_insn(p, BRW_OPCODE_SEND);
 
@@ -2305,15 +2308,13 @@ void brw_oword_block_read(struct brw_codegen *p,
       brw_inst_set_base_mrf(devinfo, insn, mrf.nr);
    }
 
-   brw_set_dp_read_message(p,
-			   insn,
-			   bind_table_index,
-			   BRW_DATAPORT_OWORD_BLOCK_1_OWORDLOW,
+   brw_set_dp_read_message(p, insn, bind_table_index,
+                           BRW_DATAPORT_OWORD_BLOCK_DWORDS(exec_size),
 			   BRW_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ,
 			   target_cache,
 			   1, /* msg_length */
                            true, /* header_present */
-			   1); /* response_length (1 reg, 2 owords!) */
+			   DIV_ROUND_UP(exec_size, 8)); /* response_length */
 
    brw_pop_insn_state(p);
 }
diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp
index b22dc9a1a7..977fd8c35f 100644
--- a/src/mesa/drivers/dri/i965/brw_fs.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
@@ -2121,7 +2121,7 @@ fs_visitor::lower_constant_loads()
 
          assert(inst->src[i].stride == 0);
 
-         const fs_builder ubld = ibld.exec_all().group(8, 0);
+         const fs_builder ubld = ibld.exec_all().group(4, 0);
          struct brw_reg offset = brw_imm_ud((unsigned)(pull_index * 4) & ~15);
          ubld.emit(FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD,
                    dst, brw_imm_ud(index), offset);
diff --git a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
index 8b9fa8e504..93f4c4199b 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
@@ -1127,6 +1127,7 @@ fs_generator::generate_uniform_pull_constant_load(fs_inst *inst,
                                                   struct brw_reg index,
                                                   struct brw_reg offset)
 {
+   assert(type_sz(dst.type) == 4);
    assert(inst->mlen != 0);
 
    assert(index.file == BRW_IMMEDIATE_VALUE &&
@@ -1149,27 +1150,25 @@ fs_generator::generate_uniform_pull_constant_load_gen7(fs_inst *inst,
 {
    assert(index.type == BRW_REGISTER_TYPE_UD);
    assert(payload.file == BRW_GENERAL_REGISTER_FILE);
+   assert(type_sz(dst.type) == 4);
 
    if (index.file == BRW_IMMEDIATE_VALUE) {
       const uint32_t surf_index = index.ud;
 
       brw_push_insn_state(p);
-      brw_set_default_compression_control(p, BRW_COMPRESSION_NONE);
       brw_set_default_mask_control(p, BRW_MASK_DISABLE);
       brw_inst *send = brw_next_insn(p, BRW_OPCODE_SEND);
-      brw_inst_set_exec_size(devinfo, send, BRW_EXECUTE_4);
       brw_pop_insn_state(p);
 
-      brw_set_dest(p, send, vec4(retype(dst, BRW_REGISTER_TYPE_UD)));
-      brw_set_src0(p, send, vec4(retype(payload, BRW_REGISTER_TYPE_UD)));
-      brw_set_dp_read_message(p, send,
-                              surf_index,
-                              BRW_DATAPORT_OWORD_BLOCK_1_OWORDLOW,
+      brw_set_dest(p, send, retype(dst, BRW_REGISTER_TYPE_UD));
+      brw_set_src0(p, send, retype(payload, BRW_REGISTER_TYPE_UD));
+      brw_set_dp_read_message(p, send, surf_index,
+                              BRW_DATAPORT_OWORD_BLOCK_DWORDS(inst->exec_size),
                               GEN7_DATAPORT_DC_OWORD_BLOCK_READ,
                               GEN6_SFID_DATAPORT_CONSTANT_CACHE,
                               1, /* mlen */
                               true, /* header */
-                              1); /* rlen */
+                              DIV_ROUND_UP(inst->size_written, REG_SIZE));
 
    } else {
       struct brw_reg addr = vec1(retype(brw_address_reg(0), BRW_REGISTER_TYPE_UD));
@@ -1188,17 +1187,15 @@ fs_generator::generate_uniform_pull_constant_load_gen7(fs_inst *inst,
       /* dst = send(payload, a0.0 | <descriptor>) */
       brw_inst *insn = brw_send_indirect_message(
          p, GEN6_SFID_DATAPORT_CONSTANT_CACHE,
-         vec4(retype(dst, BRW_REGISTER_TYPE_UD)),
-         vec4(retype(payload, BRW_REGISTER_TYPE_UD)), addr);
-      brw_inst_set_exec_size(p->devinfo, insn, BRW_EXECUTE_4);
-      brw_set_dp_read_message(p, insn,
-                              0, /* surface */
-                              BRW_DATAPORT_OWORD_BLOCK_1_OWORDLOW,
+         retype(dst, BRW_REGISTER_TYPE_UD),
+         retype(payload, BRW_REGISTER_TYPE_UD), addr);
+      brw_set_dp_read_message(p, insn, 0 /* surface */,
+                              BRW_DATAPORT_OWORD_BLOCK_DWORDS(inst->exec_size),
                               GEN7_DATAPORT_DC_OWORD_BLOCK_READ,
                               GEN6_SFID_DATAPORT_CONSTANT_CACHE,
                               1, /* mlen */
                               true, /* header */
-                              1); /* rlen */
+                              DIV_ROUND_UP(inst->size_written, REG_SIZE));
 
       brw_pop_insn_state(p);
    }
diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
index bfb286bb34..7df7423245 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
@@ -4059,7 +4059,9 @@ fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr
           * and we have to split it if necessary.
           */
          const unsigned type_size = type_sz(dest.type);
-         const fs_reg packed_consts = bld.vgrf(BRW_REGISTER_TYPE_F);
+         const fs_builder ubld = bld.exec_all().group(4, 0);
+         const fs_reg packed_consts = ubld.vgrf(BRW_REGISTER_TYPE_F);
+
          for (unsigned c = 0; c < instr->num_components;) {
             const unsigned base = const_offset->u32[0] + c * type_size;
 
@@ -4067,9 +4069,8 @@ fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr
             const unsigned count = MIN2(instr->num_components - c,
                                         (16 - base % 16) / type_size);
 
-            bld.exec_all()
-               .emit(FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD,
-                     packed_consts, surf_index, brw_imm_ud(base & ~15));
+            ubld.emit(FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD,
+                      packed_consts, surf_index, brw_imm_ud(base & ~15));
 
             const fs_reg consts =
                retype(byte_offset(packed_consts, base & 15), dest.type);