gallium: remove TGSI opcode DPH

use DP4 or DP3 + ADD.

Reviewed-by: Roland Scheidegger <[email protected]>

9 files changed, 6 insertions, 105 deletions

diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c
index 0319b8885f3..f6baca0ec44 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c
@@ -234,22 +234,6 @@ static struct lp_build_tgsi_action dp4_action = {
    dp4_emit	 /* emit */
 };
 
-/* TGSI_OPCODE_DPH */
-static void
-dph_fetch_args(
-   struct lp_build_tgsi_context * bld_base,
-   struct lp_build_emit_data * emit_data)
-{
-   dp_fetch_args(bld_base, emit_data, 4);
-   /* src0.w */
-   emit_data->args[3] = bld_base->base.one;
-}
-
-const struct lp_build_tgsi_action dph_action = {
-   dph_fetch_args,	 /* fetch_args */
-   dp4_emit	 /* emit */
-};
-
 /* TGSI_OPCODE_DST */
 static void
 dst_fetch_args(
@@ -1258,7 +1242,6 @@ lp_set_default_actions(struct lp_build_tgsi_context * bld_base)
    bld_base->op_actions[TGSI_OPCODE_DP2] = dp2_action;
    bld_base->op_actions[TGSI_OPCODE_DP3] = dp3_action;
    bld_base->op_actions[TGSI_OPCODE_DP4] = dp4_action;
-   bld_base->op_actions[TGSI_OPCODE_DPH] = dph_action;
    bld_base->op_actions[TGSI_OPCODE_DST] = dst_action;
    bld_base->op_actions[TGSI_OPCODE_EXP] = exp_action;
    bld_base->op_actions[TGSI_OPCODE_LIT] = lit_action;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c
index 3cc50796bcd..46abda0cdec 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c
@@ -593,9 +593,6 @@ lp_emit_instruction_aos(
    case TGSI_OPCODE_XPD:
       return FALSE;
 
-   case TGSI_OPCODE_DPH:
-      return FALSE;
-
    case TGSI_OPCODE_COS:
       src0 = lp_build_emit_fetch(&bld->bld_base, inst, 0, LP_CHAN_ALL);
       tmp0 = swizzle_scalar_aos(bld, src0, TGSI_SWIZZLE_X);
diff --git a/src/gallium/auxiliary/nir/tgsi_to_nir.c b/src/gallium/auxiliary/nir/tgsi_to_nir.c
index 3daa896062f..ea1f064b83d 100644
--- a/src/gallium/auxiliary/nir/tgsi_to_nir.c
+++ b/src/gallium/auxiliary/nir/tgsi_to_nir.c
@@ -1019,13 +1019,6 @@ ttn_dp4(nir_builder *b, nir_op op, nir_alu_dest dest, nir_ssa_def **src)
 }
 
 static void
-ttn_dph(nir_builder *b, nir_op op, nir_alu_dest dest, nir_ssa_def **src)
-{
-   ttn_move_dest(b, dest, nir_fadd(b, nir_fdot3(b, src[0], src[1]),
-                                   ttn_channel(b, src[1], W)));
-}
-
-static void
 ttn_umad(nir_builder *b, nir_op op, nir_alu_dest dest, nir_ssa_def **src)
 {
    ttn_move_dest(b, dest, nir_iadd(b, nir_imul(b, src[0], src[1]), src[2]));
@@ -1534,7 +1527,6 @@ static const nir_op op_trans[TGSI_OPCODE_LAST] = {
    [TGSI_OPCODE_LG2] = nir_op_flog2,
    [TGSI_OPCODE_POW] = nir_op_fpow,
    [TGSI_OPCODE_XPD] = 0,
-   [TGSI_OPCODE_DPH] = 0,
    [TGSI_OPCODE_COS] = nir_op_fcos,
    [TGSI_OPCODE_DDX] = nir_op_fddx,
    [TGSI_OPCODE_DDY] = nir_op_fddy,
@@ -1763,10 +1755,6 @@ ttn_emit_instruction(struct ttn_compile *c)
       ttn_dp4(b, op_trans[tgsi_op], dest, src);
       break;
 
-   case TGSI_OPCODE_DPH:
-      ttn_dph(b, op_trans[tgsi_op], dest, src);
-      break;
-
    case TGSI_OPCODE_UMAD:
       ttn_umad(b, op_trans[tgsi_op], dest, src);
       break;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 34a4af6f55b..c9265fbfdae 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -3185,35 +3185,6 @@ exec_dp4(struct tgsi_exec_machine *mach,
 }
 
 static void
-exec_dph(struct tgsi_exec_machine *mach,
-         const struct tgsi_full_instruction *inst)
-{
-   unsigned int chan;
-   union tgsi_exec_channel arg[3];
-
-   fetch_source(mach, &arg[0], &inst->Src[0], TGSI_CHAN_X, TGSI_EXEC_DATA_FLOAT);
-   fetch_source(mach, &arg[1], &inst->Src[1], TGSI_CHAN_X, TGSI_EXEC_DATA_FLOAT);
-   micro_mul(&arg[2], &arg[0], &arg[1]);
-
-   fetch_source(mach, &arg[0], &inst->Src[0], TGSI_CHAN_Y, TGSI_EXEC_DATA_FLOAT);
-   fetch_source(mach, &arg[1], &inst->Src[1], TGSI_CHAN_Y, TGSI_EXEC_DATA_FLOAT);
-   micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]);
-
-   fetch_source(mach, &arg[0], &inst->Src[0], TGSI_CHAN_Z, TGSI_EXEC_DATA_FLOAT);
-   fetch_source(mach, &arg[1], &inst->Src[1], TGSI_CHAN_Z, TGSI_EXEC_DATA_FLOAT);
-   micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]);
-
-   fetch_source(mach, &arg[1], &inst->Src[1], TGSI_CHAN_W, TGSI_EXEC_DATA_FLOAT);
-   micro_add(&arg[0], &arg[0], &arg[1]);
-
-   for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
-      if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
-         store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
-      }
-   }
-}
-
-static void
 exec_dp2(struct tgsi_exec_machine *mach,
          const struct tgsi_full_instruction *inst)
 {
@@ -5186,10 +5157,6 @@ exec_instruction(
       exec_xpd(mach, inst);
       break;
 
-   case TGSI_OPCODE_DPH:
-      exec_dph(mach, inst);
-      break;
-
    case TGSI_OPCODE_COS:
       exec_scalar_unary(mach, inst, micro_cos, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
       break;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_info.c b/src/gallium/auxiliary/tgsi/tgsi_info.c
index c31705a0ee9..b2477940b9a 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_info.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_info.c
@@ -72,7 +72,7 @@ static const struct tgsi_opcode_info opcode_info[TGSI_OPCODE_LAST] =
    { 1, 1, 0, 0, 0, 0, 0, COMP, "U2I64", TGSI_OPCODE_U2I64 },
    { 1, 0, 0, 0, 0, 0, 0, OTHR, "CLOCK", TGSI_OPCODE_CLOCK },
    { 1, 1, 0, 0, 0, 0, 0, COMP, "I2I64", TGSI_OPCODE_I2I64 },
-   { 1, 2, 0, 0, 0, 0, 0, REPL, "DPH", TGSI_OPCODE_DPH },
+   { 1, 2, 0, 0, 0, 0, 0, REPL, "", 35 }, /* removed */
    { 1, 1, 0, 0, 0, 0, 0, REPL, "COS", TGSI_OPCODE_COS },
    { 1, 1, 0, 0, 0, 0, 0, COMP, "DDX", TGSI_OPCODE_DDX },
    { 1, 1, 0, 0, 0, 0, 0, COMP, "DDY", TGSI_OPCODE_DDY },
diff --git a/src/gallium/auxiliary/tgsi/tgsi_lowering.c b/src/gallium/auxiliary/tgsi/tgsi_lowering.c
index 3013a41a8ef..f3b5ade2269 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_lowering.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_lowering.c
@@ -914,9 +914,6 @@ transform_log(struct tgsi_transform_context *tctx,
  * DP3 - 3-component Dot Product
  *   dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z
  *
- * DPH - Homogeneous Dot Product
- *   dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z + src1.w
- *
  * DP2 - 2-component Dot Product
  *   dst = src0.x \times src1.x + src0.y \times src1.y
  *
@@ -924,16 +921,14 @@ transform_log(struct tgsi_transform_context *tctx,
  * operations, which is what you'd prefer for a ISA that is natively
  * scalar.  Probably a native vector ISA would at least already have
  * DP4/DP3 instructions, but perhaps there is room for an alternative
- * translation for DPH/DP2 using vector instructions.
+ * translation for DP2 using vector instructions.
  *
  * ; needs: 1 tmp
  * MUL tmpA.x, src0.x, src1.x
  * MAD tmpA.x, src0.y, src1.y, tmpA.x
- * if (DPH || DP3 || DP4) {
+ * if (DP3 || DP4) {
  *   MAD tmpA.x, src0.z, src1.z, tmpA.x
- *   if (DPH) {
- *     ADD tmpA.x, src1.w, tmpA.x
- *   } else if (DP4) {
+ *   if (DP4) {
  *     MAD tmpA.x, src0.w, src1.w, tmpA.x
  *   }
  * }
@@ -941,7 +936,6 @@ transform_log(struct tgsi_transform_context *tctx,
  */
 #define DP4_GROW  (NINST(2) + NINST(3) + NINST(3) + NINST(3) - OINST(2))
 #define DP3_GROW  (NINST(2) + NINST(3) + NINST(3) - OINST(2))
-#define DPH_GROW  (NINST(2) + NINST(3) + NINST(3) + NINST(2) - OINST(2))
 #define DP2_GROW  (NINST(2) + NINST(3) - OINST(2))
 #define DOTP_TMP  1
 static void
@@ -980,8 +974,7 @@ transform_dotp(struct tgsi_transform_context *tctx,
       reg_src(&new_inst.Src[1], src1, SWIZ(Y, Y, Y, Y));
       reg_src(&new_inst.Src[2], &ctx->tmp[A].src, SWIZ(X, X, X, X));
 
-      if ((opcode == TGSI_OPCODE_DPH) ||
-          (opcode == TGSI_OPCODE_DP3) ||
+      if ((opcode == TGSI_OPCODE_DP3) ||
           (opcode == TGSI_OPCODE_DP4)) {
          tctx->emit_instruction(tctx, &new_inst);
 
@@ -995,18 +988,7 @@ transform_dotp(struct tgsi_transform_context *tctx,
          reg_src(&new_inst.Src[1], src1, SWIZ(Z, Z, Z, Z));
          reg_src(&new_inst.Src[2], &ctx->tmp[A].src, SWIZ(X, X, X, X));
 
-         if (opcode == TGSI_OPCODE_DPH) {
-            tctx->emit_instruction(tctx, &new_inst);
-
-            /* ADD tmpA.x, src1.w, tmpA.x */
-            new_inst = tgsi_default_full_instruction();
-            new_inst.Instruction.Opcode = TGSI_OPCODE_ADD;
-            new_inst.Instruction.NumDstRegs = 1;
-            reg_dst(&new_inst.Dst[0], &ctx->tmp[A].dst, TGSI_WRITEMASK_X);
-            new_inst.Instruction.NumSrcRegs = 2;
-            reg_src(&new_inst.Src[0], src1, SWIZ(W, W, W, W));
-            reg_src(&new_inst.Src[1], &ctx->tmp[A].src, SWIZ(X, X, X, X));
-         } else if (opcode == TGSI_OPCODE_DP4) {
+         if (opcode == TGSI_OPCODE_DP4) {
             tctx->emit_instruction(tctx, &new_inst);
 
             /* MAD tmpA.x, src0.w, src1.w, tmpA.x */
@@ -1534,11 +1516,6 @@ transform_instr(struct tgsi_transform_context *tctx,
          goto skip;
       transform_dotp(tctx, inst);
       break;
-   case TGSI_OPCODE_DPH:
-      if (!ctx->config->lower_DPH)
-         goto skip;
-      transform_dotp(tctx, inst);
-      break;
    case TGSI_OPCODE_DP2:
       if (!ctx->config->lower_DP2)
          goto skip;
@@ -1632,7 +1609,6 @@ tgsi_transform_lowering(const struct tgsi_lowering_config *config,
          OPCS(LOG) ||
          OPCS(DP4) ||
          OPCS(DP3) ||
-         OPCS(DPH) ||
          OPCS(DP2) ||
          OPCS(FLR) ||
          OPCS(CEIL) ||
@@ -1693,10 +1669,6 @@ tgsi_transform_lowering(const struct tgsi_lowering_config *config,
       newlen += DP3_GROW * OPCS(DP3);
       numtmp = MAX2(numtmp, DOTP_TMP);
    }
-   if (OPCS(DPH)) {
-      newlen += DPH_GROW * OPCS(DPH);
-      numtmp = MAX2(numtmp, DOTP_TMP);
-   }
    if (OPCS(DP2)) {
       newlen += DP2_GROW * OPCS(DP2);
       numtmp = MAX2(numtmp, DOTP_TMP);
diff --git a/src/gallium/auxiliary/tgsi/tgsi_lowering.h b/src/gallium/auxiliary/tgsi/tgsi_lowering.h
index 85e4b8e9833..f65d915c654 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_lowering.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_lowering.h
@@ -65,7 +65,6 @@ struct tgsi_lowering_config
    unsigned lower_LOG:1;
    unsigned lower_DP4:1;
    unsigned lower_DP3:1;
-   unsigned lower_DPH:1;
    unsigned lower_DP2:1;
    unsigned lower_FLR:1;
    unsigned lower_CEIL:1;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h b/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
index e4cbdae2718..68b4af81b71 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
@@ -64,7 +64,6 @@ OP11(EX2)
 OP11(LG2)
 OP12(POW)
 OP12(XPD)
-OP12(DPH)
 OP11(COS)
 OP11(DDX)
 OP11(DDY)
diff --git a/src/gallium/auxiliary/tgsi/tgsi_util.c b/src/gallium/auxiliary/tgsi/tgsi_util.c
index fc61351f6ed..6756bca23df 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_util.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_util.c
@@ -274,10 +274,6 @@ tgsi_util_get_inst_usage_mask(const struct tgsi_full_instruction *inst,
       read_mask = TGSI_WRITEMASK_XYZW;
       break;
 
-   case TGSI_OPCODE_DPH:
-      read_mask = src_idx == 0 ? TGSI_WRITEMASK_XYZ : TGSI_WRITEMASK_XYZW;
-      break;
-
    case TGSI_OPCODE_TEX:
    case TGSI_OPCODE_TXD:
    case TGSI_OPCODE_TXB: