i965: Introduce the BROADCAST pseudo-opcode.

The BROADCAST instruction picks the channel from its first source
given by an index passed in as second source.  This will be used in
situations where all channels from the same SIMD thread have to agree
on the value of something, e.g. a surface binding table index.

This is in particular the case for UBO, sampler and image arrays,
which can be indexed dynamically with the restriction that all active
SIMD channels access the same index, provided to the shared unit as
part of a single scalar field of the message descriptor.  Simply
taking the index value from the first channel as we were doing until
now is incorrect, because it might contain an uninitialized value if
the channel had previously been disabled by non-uniform control flow.

v2: Minor style fixes.  Improve commit message.

Reviewed-by: Matt Turner <mattst88@gmail.com>

6 files changed, 98 insertions, 0 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_defines.h b/src/mesa/drivers/dri/i965/brw_defines.h
index 634885b..5072e68 100644
--- a/src/mesa/drivers/dri/i965/brw_defines.h
+++ b/src/mesa/drivers/dri/i965/brw_defines.h
@@ -918,6 +918,12 @@ enum opcode {
 
    SHADER_OPCODE_URB_WRITE_SIMD8,
 
+   /**
+    * Pick the channel from its first source register given by the index
+    * specified as second source.  Useful for variable indexing of surfaces.
+    */
+   SHADER_OPCODE_BROADCAST,
+
    VEC4_OPCODE_MOV_BYTES,
    VEC4_OPCODE_PACK_BYTES,
    VEC4_OPCODE_UNPACK_UNIFORM,
diff --git a/src/mesa/drivers/dri/i965/brw_eu.h b/src/mesa/drivers/dri/i965/brw_eu.h
index 84c1e57..a0c938a 100644
--- a/src/mesa/drivers/dri/i965/brw_eu.h
+++ b/src/mesa/drivers/dri/i965/brw_eu.h
@@ -461,6 +461,12 @@ brw_pixel_interpolator_query(struct brw_codegen *p,
                              unsigned msg_length,
                              unsigned response_length);
 
+void
+brw_broadcast(struct brw_codegen *p,
+              struct brw_reg dst,
+              struct brw_reg src,
+              struct brw_reg idx);
+
 /***********************************************************************
  * brw_eu_util.c:
  */
diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c
index 3039662..73bed49 100644
--- a/src/mesa/drivers/dri/i965/brw_eu_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c
@@ -3212,6 +3212,81 @@ brw_pixel_interpolator_query(struct brw_codegen *p,
    brw_inst_set_pi_message_data(devinfo, insn, data);
 }
 
+void
+brw_broadcast(struct brw_codegen *p,
+              struct brw_reg dst,
+              struct brw_reg src,
+              struct brw_reg idx)
+{
+   const struct brw_device_info *devinfo = p->devinfo;
+   const bool align1 = brw_inst_access_mode(devinfo, p->current) == BRW_ALIGN_1;
+   brw_inst *inst;
+
+   assert(src.file == BRW_GENERAL_REGISTER_FILE &&
+          src.address_mode == BRW_ADDRESS_DIRECT);
+
+   if ((src.vstride == 0 && (src.hstride == 0 || !align1)) ||
+       idx.file == BRW_IMMEDIATE_VALUE) {
+      /* Trivial, the source is already uniform or the index is a constant.
+       * We will typically not get here if the optimizer is doing its job, but
+       * asserting would be mean.
+       */
+      const unsigned i = idx.file == BRW_IMMEDIATE_VALUE ? idx.dw1.ud : 0;
+      brw_MOV(p, dst,
+              (align1 ? stride(suboffset(src, i), 0, 1, 0) :
+               stride(suboffset(src, 4 * i), 0, 4, 1)));
+   } else {
+      if (align1) {
+         const struct brw_reg addr =
+            retype(brw_address_reg(0), BRW_REGISTER_TYPE_UD);
+         const unsigned offset = src.nr * REG_SIZE + src.subnr;
+         /* Limit in bytes of the signed indirect addressing immediate. */
+         const unsigned limit = 512;
+
+         brw_push_insn_state(p);
+         brw_set_default_mask_control(p, BRW_MASK_DISABLE);
+         brw_set_default_predicate_control(p, BRW_PREDICATE_NONE);
+
+         /* Take into account the component size and horizontal stride. */
+         assert(src.vstride == src.hstride + src.width);
+         brw_SHL(p, addr, vec1(idx),
+                 brw_imm_ud(_mesa_logbase2(type_sz(src.type)) +
+                            src.hstride - 1));
+
+         /* We can only address up to limit bytes using the indirect
+          * addressing immediate, account for the difference if the source
+          * register is above this limit.
+          */
+         if (offset >= limit)
+            brw_ADD(p, addr, addr, brw_imm_ud(offset - offset % limit));
+
+         brw_pop_insn_state(p);
+
+         /* Use indirect addressing to fetch the specified component. */
+         brw_MOV(p, dst,
+                 retype(brw_vec1_indirect(addr.subnr, offset % limit),
+                        src.type));
+      } else {
+         /* In SIMD4x2 mode the index can be either zero or one, replicate it
+          * to all bits of a flag register,
+          */
+         inst = brw_MOV(p,
+                        brw_null_reg(),
+                        stride(brw_swizzle1(idx, 0), 0, 4, 1));
+         brw_inst_set_pred_control(devinfo, inst, BRW_PREDICATE_NONE);
+         brw_inst_set_cond_modifier(devinfo, inst, BRW_CONDITIONAL_NZ);
+         brw_inst_set_flag_reg_nr(devinfo, inst, 1);
+
+         /* and use predicated SEL to pick the right channel. */
+         inst = brw_SEL(p, dst,
+                        stride(suboffset(src, 4), 0, 4, 1),
+                        stride(src, 0, 4, 1));
+         brw_inst_set_pred_control(devinfo, inst, BRW_PREDICATE_NORMAL);
+         brw_inst_set_flag_reg_nr(devinfo, inst, 1);
+      }
+   }
+}
+
 /**
  * This instruction is generated as a single-channel align1 instruction by
  * both the VS and FS stages when using INTEL_DEBUG=shader_time.
diff --git a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
index b81978c..2c6a12e 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
@@ -2061,6 +2061,10 @@ fs_generator::generate_code(const cfg_t *cfg, int dispatch_width)
          generate_set_simd4x2_offset(inst, dst, src[0]);
          break;
 
+      case SHADER_OPCODE_BROADCAST:
+         brw_broadcast(p, dst, src[0], src[1]);
+         break;
+
       case FS_OPCODE_SET_OMASK:
          generate_set_omask(inst, dst, src[0]);
          break;
diff --git a/src/mesa/drivers/dri/i965/brw_shader.cpp b/src/mesa/drivers/dri/i965/brw_shader.cpp
index 20588e9..1944c26 100644
--- a/src/mesa/drivers/dri/i965/brw_shader.cpp
+++ b/src/mesa/drivers/dri/i965/brw_shader.cpp
@@ -517,6 +517,9 @@ brw_instruction_name(enum opcode op)
    case SHADER_OPCODE_URB_WRITE_SIMD8:
       return "gen8_urb_write_simd8";
 
+   case SHADER_OPCODE_BROADCAST:
+      return "broadcast";
+
    case VEC4_OPCODE_MOV_BYTES:
       return "mov_bytes";
    case VEC4_OPCODE_PACK_BYTES:
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp b/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
index c15fa16..3dc808c 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
@@ -1512,6 +1512,10 @@ vec4_generator::generate_code(const cfg_t *cfg)
          brw_memory_fence(p, dst);
          break;
 
+      case SHADER_OPCODE_BROADCAST:
+         brw_broadcast(p, dst, src[0], src[1]);
+         break;
+
       case VS_OPCODE_UNPACK_FLAGS_SIMD4X2:
          generate_unpack_flags(dst);
          break;