gallium: remove TGSI opcode DPH

use DP4 or DP3 + ADD.

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

10 files changed, 3 insertions, 139 deletions

diff --git a/src/gallium/drivers/etnaviv/etnaviv_compiler.c b/src/gallium/drivers/etnaviv/etnaviv_compiler.c
index 88f204cacaa..3ccb7373817 100644
--- a/src/gallium/drivers/etnaviv/etnaviv_compiler.c
+++ b/src/gallium/drivers/etnaviv/etnaviv_compiler.c
@@ -1664,33 +1664,6 @@ trans_lg2(const struct instr_translater *t, struct etna_compile *c,
 }
 
 static void
-trans_dph(const struct instr_translater *t, struct etna_compile *c,
-          const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
-{
-   /*
-   DP3 tmp.xyzw, src0.xyzw, src1,xyzw, void
-   ADD dst.xyzw, tmp.xyzw, void, src1.wwww
-   */
-   struct etna_native_reg temp = etna_compile_get_inner_temp(c);
-   struct etna_inst ins[2] = { };
-
-   ins[0].opcode = INST_OPCODE_DP3;
-   ins[0].dst = etna_native_to_dst(temp, INST_COMPS_X | INST_COMPS_Y |
-                                         INST_COMPS_Z | INST_COMPS_W);
-   ins[0].src[0] = src[0];
-   ins[0].src[1] = src[1];
-
-   ins[1].opcode = INST_OPCODE_ADD;
-   ins[1].sat = inst->Instruction.Saturate;
-   ins[1].dst = convert_dst(c, &inst->Dst[0]);
-   ins[1].src[0] = etna_native_to_src(temp, INST_SWIZ_IDENTITY);
-   ins[1].src[2] = swizzle(src[1], SWIZZLE(W, W, W, W));
-
-   emit_inst(c, &ins[0]);
-   emit_inst(c, &ins[1]);
-}
-
-static void
 trans_sampler(const struct instr_translater *t, struct etna_compile *c,
               const struct tgsi_full_instruction *inst,
               struct etna_inst_src *src)
@@ -1833,7 +1806,6 @@ static const struct instr_translater translaters[TGSI_OPCODE_LAST] = {
    INSTR(LRP, trans_lrp),
    INSTR(LIT, trans_lit),
    INSTR(SSG, trans_ssg),
-   INSTR(DPH, trans_dph),
 
    INSTR(SIN, trans_trig),
    INSTR(COS, trans_trig),
diff --git a/src/gallium/drivers/i915/i915_fpc_optimize.c b/src/gallium/drivers/i915/i915_fpc_optimize.c
index 5f2a8760478..da06e16bf0e 100644
--- a/src/gallium/drivers/i915/i915_fpc_optimize.c
+++ b/src/gallium/drivers/i915/i915_fpc_optimize.c
@@ -85,7 +85,6 @@ static const struct {
    [ TGSI_OPCODE_DP2     ] = { false,   true,   TGSI_SWIZZLE_ONE,  1,  2 },
    [ TGSI_OPCODE_DP3     ] = { false,   true,   TGSI_SWIZZLE_ONE,  1,  2 },
    [ TGSI_OPCODE_DP4     ] = { false,   true,   TGSI_SWIZZLE_ONE,  1,  2 },
-   [ TGSI_OPCODE_DPH     ] = { false,  false,                  0,  1,  2 },
    [ TGSI_OPCODE_DST     ] = { false,  false,                  0,  1,  2 },
    [ TGSI_OPCODE_END     ] = { false,  false,                  0,  0,  0 },
    [ TGSI_OPCODE_EX2     ] = { false,  false,                  0,  1,  1 },
diff --git a/src/gallium/drivers/i915/i915_fpc_translate.c b/src/gallium/drivers/i915/i915_fpc_translate.c
index 241c92dd28a..22a42eeeab4 100644
--- a/src/gallium/drivers/i915/i915_fpc_translate.c
+++ b/src/gallium/drivers/i915/i915_fpc_translate.c
@@ -604,17 +604,6 @@ i915_translate_instruction(struct i915_fp_compile *p,
       emit_simple_arith(p, inst, A0_DP4, 2, fs);
       break;
 
-   case TGSI_OPCODE_DPH:
-      src0 = src_vector(p, &inst->Src[0], fs);
-      src1 = src_vector(p, &inst->Src[1], fs);
-
-      i915_emit_arith(p,
-                      A0_DP4,
-                      get_result_vector(p, &inst->Dst[0]),
-                      get_result_flags(inst), 0,
-                      swizzle(src0, X, Y, Z, ONE), src1, 0);
-      break;
-
    case TGSI_OPCODE_DST:
       src0 = src_vector(p, &inst->Src[0], fs);
       src1 = src_vector(p, &inst->Src[1], fs);
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
index 23c7d106e39..c06e74e2f81 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
@@ -277,7 +277,6 @@ unsigned int Instruction::srcMask(unsigned int s) const
    case TGSI_OPCODE_DP3:
       return 0x7;
    case TGSI_OPCODE_DP4:
-   case TGSI_OPCODE_DPH:
    case TGSI_OPCODE_KILL_IF: /* WriteMask ignored */
       return 0xf;
    case TGSI_OPCODE_DST:
@@ -3321,13 +3320,6 @@ Converter::handleInstruction(const struct tgsi_full_instruction *insn)
       FOR_EACH_DST_ENABLED_CHANNEL(0, c, tgsi)
          mkMov(dst0[c], val0);
       break;
-   case TGSI_OPCODE_DPH:
-      val0 = buildDot(3);
-      src1 = fetchSrc(1, 3);
-      mkOp2(OP_ADD, TYPE_F32, val0, val0, src1);
-      FOR_EACH_DST_ENABLED_CHANNEL(0, c, tgsi)
-         mkMov(dst0[c], val0);
-      break;
    case TGSI_OPCODE_DST:
       if (dst0[0])
          loadImm(dst0[0], 1.0f);
diff --git a/src/gallium/drivers/nouveau/nv30/nvfx_fragprog.c b/src/gallium/drivers/nouveau/nv30/nvfx_fragprog.c
index 278a8a4a430..a3ed5c6526b 100644
--- a/src/gallium/drivers/nouveau/nv30/nvfx_fragprog.c
+++ b/src/gallium/drivers/nouveau/nv30/nvfx_fragprog.c
@@ -591,11 +591,6 @@ nvfx_fragprog_parse_instruction(struct nvfx_fpc *fpc,
    case TGSI_OPCODE_DP4:
       nvfx_fp_emit(fpc, arith(sat, DP4, dst, mask, src[0], src[1], none));
       break;
-   case TGSI_OPCODE_DPH:
-      tmp = nvfx_src(temp(fpc));
-      nvfx_fp_emit(fpc, arith(0, DP3, tmp.reg, NVFX_FP_MASK_X, src[0], src[1], none));
-      nvfx_fp_emit(fpc, arith(sat, ADD, dst, mask, swz(tmp, X, X, X, X), swz(src[1], W, W, W, W), none));
-      break;
    case TGSI_OPCODE_DST:
       nvfx_fp_emit(fpc, arith(sat, DST, dst, mask, src[0], src[1], none));
       break;
diff --git a/src/gallium/drivers/nouveau/nv30/nvfx_vertprog.c b/src/gallium/drivers/nouveau/nv30/nvfx_vertprog.c
index bec9975de9d..8ba3d5a0500 100644
--- a/src/gallium/drivers/nouveau/nv30/nvfx_vertprog.c
+++ b/src/gallium/drivers/nouveau/nv30/nvfx_vertprog.c
@@ -588,9 +588,6 @@ nvfx_vertprog_parse_instruction(struct nvfx_vpc *vpc,
    case TGSI_OPCODE_DP4:
       nvfx_vp_emit(vpc, arith(sat, VEC, DP4, dst, mask, src[0], src[1], none));
       break;
-   case TGSI_OPCODE_DPH:
-      nvfx_vp_emit(vpc, arith(sat, VEC, DPH, dst, mask, src[0], src[1], none));
-      break;
    case TGSI_OPCODE_DST:
       nvfx_vp_emit(vpc, arith(sat, VEC, DST, dst, mask, src[0], src[1], none));
       break;
diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.c b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
index cc0ac4810e8..78af1242b4f 100644
--- a/src/gallium/drivers/r300/r300_tgsi_to_rc.c
+++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
@@ -58,7 +58,6 @@ static unsigned translate_opcode(unsigned opcode)
         case TGSI_OPCODE_LG2: return RC_OPCODE_LG2;
         case TGSI_OPCODE_POW: return RC_OPCODE_POW;
         case TGSI_OPCODE_XPD: return RC_OPCODE_XPD;
-        case TGSI_OPCODE_DPH: return RC_OPCODE_DPH;
         case TGSI_OPCODE_COS: return RC_OPCODE_COS;
         case TGSI_OPCODE_DDX: return RC_OPCODE_DDX;
         case TGSI_OPCODE_DDY: return RC_OPCODE_DDY;
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index d3728fb61de..fd76c93de88 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -6617,13 +6617,6 @@ static int tgsi_dp(struct r600_shader_ctx *ctx)
 				alu.src[0].chan = alu.src[1].chan = 0;
 			}
 			break;
-		case TGSI_OPCODE_DPH:
-			if (i == 3) {
-				alu.src[0].sel = V_SQ_ALU_SRC_1;
-				alu.src[0].chan = 0;
-				alu.src[0].neg = 0;
-			}
-			break;
 		default:
 			break;
 		}
@@ -9103,7 +9096,7 @@ static const struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction[]
 	[32]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[33]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[34]			= { ALU_OP0_NOP, tgsi_unsupported},
-	[TGSI_OPCODE_DPH]	= { ALU_OP2_DOT4_IEEE, tgsi_dp},
+	[35]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[TGSI_OPCODE_COS]	= { ALU_OP1_COS, tgsi_trig},
 	[TGSI_OPCODE_DDX]	= { FETCH_OP_GET_GRADIENTS_H, tgsi_tex},
 	[TGSI_OPCODE_DDY]	= { FETCH_OP_GET_GRADIENTS_V, tgsi_tex},
@@ -9301,7 +9294,7 @@ static const struct r600_shader_tgsi_instruction eg_shader_tgsi_instruction[] =
 	[32]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[33]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[34]			= { ALU_OP0_NOP, tgsi_unsupported},
-	[TGSI_OPCODE_DPH]	= { ALU_OP2_DOT4_IEEE, tgsi_dp},
+	[35]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[TGSI_OPCODE_COS]	= { ALU_OP1_COS, tgsi_trig},
 	[TGSI_OPCODE_DDX]	= { FETCH_OP_GET_GRADIENTS_H, tgsi_tex},
 	[TGSI_OPCODE_DDY]	= { FETCH_OP_GET_GRADIENTS_V, tgsi_tex},
@@ -9524,7 +9517,7 @@ static const struct r600_shader_tgsi_instruction cm_shader_tgsi_instruction[] =
 	[32]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[33]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[34]			= { ALU_OP0_NOP, tgsi_unsupported},
-	[TGSI_OPCODE_DPH]	= { ALU_OP2_DOT4_IEEE, tgsi_dp},
+	[35]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[TGSI_OPCODE_COS]	= { ALU_OP1_COS, cayman_trig},
 	[TGSI_OPCODE_DDX]	= { FETCH_OP_GET_GRADIENTS_H, tgsi_tex},
 	[TGSI_OPCODE_DDY]	= { FETCH_OP_GET_GRADIENTS_V, tgsi_tex},
diff --git a/src/gallium/drivers/svga/svga_tgsi_insn.c b/src/gallium/drivers/svga/svga_tgsi_insn.c
index dd29f74a2a7..928330ce147 100644
--- a/src/gallium/drivers/svga/svga_tgsi_insn.c
+++ b/src/gallium/drivers/svga/svga_tgsi_insn.c
@@ -1237,39 +1237,6 @@ emit_dp2(struct svga_shader_emitter *emit,
 
 
 /**
- * Translate the following TGSI DPH instruction.
- *    DPH  DST, SRC1, SRC2
- * To the following SVGA3D instruction sequence.
- *    DP3  TMP, SRC1, SRC2
- *    ADD  DST, TMP, SRC2.wwww
- */
-static boolean
-emit_dph(struct svga_shader_emitter *emit,
-         const struct tgsi_full_instruction *insn )
-{
-   SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
-   const struct src_register src0 = translate_src_register(
-      emit, &insn->Src[0] );
-   struct src_register src1 =
-      translate_src_register(emit, &insn->Src[1]);
-   SVGA3dShaderDestToken temp = get_temp( emit );
-
-   /* DP3  TMP, SRC1, SRC2 */
-   if (!submit_op2( emit, inst_token( SVGA3DOP_DP3 ), temp, src0, src1 ))
-      return FALSE;
-
-   src1 = scalar(src1, TGSI_SWIZZLE_W);
-
-   /* ADD  DST, TMP, SRC2.wwww */
-   if (!submit_op2( emit, inst_token( SVGA3DOP_ADD ), dst,
-                    src( temp ), src1 ))
-      return FALSE;
-
-   return TRUE;
-}
-
-
-/**
  * Sine / Cosine helper function.
  */
 static boolean
@@ -2924,9 +2891,6 @@ svga_emit_instruction(struct svga_shader_emitter *emit,
    case TGSI_OPCODE_DP2:
       return emit_dp2( emit, insn );
 
-   case TGSI_OPCODE_DPH:
-      return emit_dph( emit, insn );
-
    case TGSI_OPCODE_COS:
       return emit_cos( emit, insn );
 
diff --git a/src/gallium/drivers/svga/svga_tgsi_vgpu10.c b/src/gallium/drivers/svga/svga_tgsi_vgpu10.c
index 56afd497d74..e57e78d11a5 100644
--- a/src/gallium/drivers/svga/svga_tgsi_vgpu10.c
+++ b/src/gallium/drivers/svga/svga_tgsi_vgpu10.c
@@ -3578,40 +3578,6 @@ emit_cmp(struct svga_shader_emitter_v10 *emit,
 
 
 /**
- * Emit code for TGSI_OPCODE_DPH instruction.
- */
-static boolean
-emit_dph(struct svga_shader_emitter_v10 *emit,
-         const struct tgsi_full_instruction *inst)
-{
-   /*
-    * DP3 tmp, s0, s1
-    * ADD dst, tmp, s1.wwww
-    */
-
-   struct tgsi_full_src_register s1_wwww =
-      swizzle_src(&inst->Src[1], TGSI_SWIZZLE_W, TGSI_SWIZZLE_W,
-                  TGSI_SWIZZLE_W, TGSI_SWIZZLE_W);
-
-   unsigned tmp = get_temp_index(emit);
-   struct tgsi_full_src_register tmp_src = make_src_temp_reg(tmp);
-   struct tgsi_full_dst_register tmp_dst = make_dst_temp_reg(tmp);
-
-   /* DP3 tmp, s0, s1 */
-   emit_instruction_op2(emit, VGPU10_OPCODE_DP3, &tmp_dst, &inst->Src[0],
-                        &inst->Src[1], FALSE);
-
-   /* ADD dst, tmp, s1.wwww */
-   emit_instruction_op2(emit, VGPU10_OPCODE_ADD, &inst->Dst[0], &tmp_src,
-                        &s1_wwww, inst->Instruction.Saturate);
-
-   free_temp_indexes(emit);
-
-   return TRUE;
-}
-
-
-/**
  * Emit code for TGSI_OPCODE_DST instruction.
  */
 static boolean
@@ -5712,8 +5678,6 @@ emit_vgpu10_instruction(struct svga_shader_emitter_v10 *emit,
       return emit_cmp(emit, inst);
    case TGSI_OPCODE_COS:
       return emit_sincos(emit, inst);
-   case TGSI_OPCODE_DPH:
-      return emit_dph(emit, inst);
    case TGSI_OPCODE_DST:
       return emit_dst(emit, inst);
    case TGSI_OPCODE_EX2: