gallium: remove TGSI opcode XPD

use MUL+MAD+MOV instead.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>

24 files changed, 41 insertions, 539 deletions

diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c
index f6baca0ec4..52c9a86b2e 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c
@@ -858,61 +858,6 @@ static void fmin_emit(
                                    emit_data->args[1], emit_data->args[0], "");
 }
 
-/* TGSI_OPCODE_XPD */
-
-static void
-xpd_fetch_args(
-   struct lp_build_tgsi_context * bld_base,
-   struct lp_build_emit_data * emit_data)
-{
-   dp_fetch_args(bld_base, emit_data, 3);
-}
-
-/**
- * (a * b) - (c * d)
- */
-static LLVMValueRef
-xpd_helper(
-  struct lp_build_tgsi_context * bld_base,
-  LLVMValueRef a,
-  LLVMValueRef b,
-  LLVMValueRef c,
-  LLVMValueRef d)
-{
-   LLVMValueRef tmp0, tmp1;
-
-   tmp0 = lp_build_emit_llvm_binary(bld_base, TGSI_OPCODE_MUL, a, b);
-   tmp1 = lp_build_emit_llvm_binary(bld_base, TGSI_OPCODE_MUL, c, d);
-
-   return lp_build_sub(&bld_base->base, tmp0, tmp1);
-}
-
-static void
-xpd_emit(
-   const struct lp_build_tgsi_action * action,
-   struct lp_build_tgsi_context * bld_base,
-   struct lp_build_emit_data * emit_data)
-{
-   emit_data->output[TGSI_CHAN_X] = xpd_helper(bld_base,
-              emit_data->args[1] /* src0.y */, emit_data->args[5] /* src1.z */,
-              emit_data->args[4] /* src1.y */, emit_data->args[2] /* src0.z */);
-
-   emit_data->output[TGSI_CHAN_Y] = xpd_helper(bld_base,
-              emit_data->args[2] /* src0.z */, emit_data->args[3] /* src1.x */,
-              emit_data->args[5] /* src1.z */, emit_data->args[0] /* src0.x */);
-
-   emit_data->output[TGSI_CHAN_Z] = xpd_helper(bld_base,
-              emit_data->args[0] /* src0.x */, emit_data->args[4] /* src1.y */,
-              emit_data->args[3] /* src1.x */, emit_data->args[1] /* src0.y */);
-
-   emit_data->output[TGSI_CHAN_W] = bld_base->base.one;
-}
-
-const struct lp_build_tgsi_action xpd_action = {
-   xpd_fetch_args,	 /* fetch_args */
-   xpd_emit	 /* emit */
-};
-
 /* TGSI_OPCODE_D2F */
 static void
 d2f_emit(
@@ -1252,7 +1197,6 @@ lp_set_default_actions(struct lp_build_tgsi_context * bld_base)
    bld_base->op_actions[TGSI_OPCODE_POW] = pow_action;
    bld_base->op_actions[TGSI_OPCODE_SCS] = scs_action;
    bld_base->op_actions[TGSI_OPCODE_UP2H] = up2h_action;
-   bld_base->op_actions[TGSI_OPCODE_XPD] = xpd_action;
 
    bld_base->op_actions[TGSI_OPCODE_BREAKC].fetch_args = scalar_unary_fetch_args;
    bld_base->op_actions[TGSI_OPCODE_SWITCH].fetch_args = scalar_unary_fetch_args;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c
index 46abda0cde..567ed68927 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c
@@ -590,9 +590,6 @@ lp_emit_instruction_aos(
       dst0 = lp_build_pow(&bld->bld_base.base, src0, src1);
       break;
 
-   case TGSI_OPCODE_XPD:
-      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 ea1f064b83..55deb29ca5 100644
--- a/src/gallium/auxiliary/nir/tgsi_to_nir.c
+++ b/src/gallium/auxiliary/nir/tgsi_to_nir.c
@@ -986,21 +986,6 @@ ttn_sgt(nir_builder *b, nir_op op, nir_alu_dest dest, nir_ssa_def **src)
 }
 
 static void
-ttn_xpd(nir_builder *b, nir_op op, nir_alu_dest dest, nir_ssa_def **src)
-{
-   ttn_move_dest_masked(b, dest,
-                        nir_fsub(b,
-                                 nir_fmul(b,
-                                          ttn_swizzle(b, src[0], Y, Z, X, X),
-                                          ttn_swizzle(b, src[1], Z, X, Y, X)),
-                                 nir_fmul(b,
-                                          ttn_swizzle(b, src[1], Y, Z, X, X),
-                                          ttn_swizzle(b, src[0], Z, X, Y, X))),
-                        TGSI_WRITEMASK_XYZ);
-   ttn_move_dest_masked(b, dest, nir_imm_float(b, 1.0), TGSI_WRITEMASK_W);
-}
-
-static void
 ttn_dp2(nir_builder *b, nir_op op, nir_alu_dest dest, nir_ssa_def **src)
 {
    ttn_move_dest(b, dest, nir_fdot2(b, src[0], src[1]));
@@ -1526,7 +1511,6 @@ static const nir_op op_trans[TGSI_OPCODE_LAST] = {
    [TGSI_OPCODE_EX2] = nir_op_fexp2,
    [TGSI_OPCODE_LG2] = nir_op_flog2,
    [TGSI_OPCODE_POW] = nir_op_fpow,
-   [TGSI_OPCODE_XPD] = 0,
    [TGSI_OPCODE_COS] = nir_op_fcos,
    [TGSI_OPCODE_DDX] = nir_op_fddx,
    [TGSI_OPCODE_DDY] = nir_op_fddy,
@@ -1739,10 +1723,6 @@ ttn_emit_instruction(struct ttn_compile *c)
       ttn_lit(b, op_trans[tgsi_op], dest, src);
       break;
 
-   case TGSI_OPCODE_XPD:
-      ttn_xpd(b, op_trans[tgsi_op], dest, src);
-      break;
-
    case TGSI_OPCODE_DP2:
       ttn_dp2(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 c9265fbfda..bce158b42a 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -3313,51 +3313,6 @@ exec_scs(struct tgsi_exec_machine *mach,
 }
 
 static void
-exec_xpd(struct tgsi_exec_machine *mach,
-         const struct tgsi_full_instruction *inst)
-{
-   union tgsi_exec_channel r[6];
-   union tgsi_exec_channel d[3];
-
-   fetch_source(mach, &r[0], &inst->Src[0], TGSI_CHAN_Y, TGSI_EXEC_DATA_FLOAT);
-   fetch_source(mach, &r[1], &inst->Src[1], TGSI_CHAN_Z, TGSI_EXEC_DATA_FLOAT);
-
-   micro_mul(&r[2], &r[0], &r[1]);
-
-   fetch_source(mach, &r[3], &inst->Src[0], TGSI_CHAN_Z, TGSI_EXEC_DATA_FLOAT);
-   fetch_source(mach, &r[4], &inst->Src[1], TGSI_CHAN_Y, TGSI_EXEC_DATA_FLOAT);
-
-   micro_mul(&r[5], &r[3], &r[4] );
-   micro_sub(&d[TGSI_CHAN_X], &r[2], &r[5]);
-
-   fetch_source(mach, &r[2], &inst->Src[1], TGSI_CHAN_X, TGSI_EXEC_DATA_FLOAT);
-
-   micro_mul(&r[3], &r[3], &r[2]);
-
-   fetch_source(mach, &r[5], &inst->Src[0], TGSI_CHAN_X, TGSI_EXEC_DATA_FLOAT);
-
-   micro_mul(&r[1], &r[1], &r[5]);
-   micro_sub(&d[TGSI_CHAN_Y], &r[3], &r[1]);
-
-   micro_mul(&r[5], &r[5], &r[4]);
-   micro_mul(&r[0], &r[0], &r[2]);
-   micro_sub(&d[TGSI_CHAN_Z], &r[5], &r[0]);
-
-   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
-      store_dest(mach, &d[TGSI_CHAN_X], &inst->Dst[0], inst, TGSI_CHAN_X, TGSI_EXEC_DATA_FLOAT);
-   }
-   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
-      store_dest(mach, &d[TGSI_CHAN_Y], &inst->Dst[0], inst, TGSI_CHAN_Y, TGSI_EXEC_DATA_FLOAT);
-   }
-   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
-      store_dest(mach, &d[TGSI_CHAN_Z], &inst->Dst[0], inst, TGSI_CHAN_Z, TGSI_EXEC_DATA_FLOAT);
-   }
-   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
-      store_dest(mach, &OneVec, &inst->Dst[0], inst, TGSI_CHAN_W, TGSI_EXEC_DATA_FLOAT);
-   }
-}
-
-static void
 exec_dst(struct tgsi_exec_machine *mach,
          const struct tgsi_full_instruction *inst)
 {
@@ -5153,10 +5108,6 @@ exec_instruction(
       exec_scalar_binary(mach, inst, micro_pow, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
       break;
 
-   case TGSI_OPCODE_XPD:
-      exec_xpd(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 b2477940b9..17f56fd9b7 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_info.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_info.c
@@ -68,7 +68,7 @@ static const struct tgsi_opcode_info opcode_info[TGSI_OPCODE_LAST] =
    { 1, 1, 0, 0, 0, 0, 0, REPL, "EX2", TGSI_OPCODE_EX2 },
    { 1, 1, 0, 0, 0, 0, 0, REPL, "LG2", TGSI_OPCODE_LG2 },
    { 1, 2, 0, 0, 0, 0, 0, REPL, "POW", TGSI_OPCODE_POW },
-   { 1, 2, 0, 0, 0, 0, 0, COMP, "XPD", TGSI_OPCODE_XPD },
+   { 1, 2, 0, 0, 0, 0, 0, COMP, "", 31 }, /* removed */
    { 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 },
diff --git a/src/gallium/auxiliary/tgsi/tgsi_lowering.c b/src/gallium/auxiliary/tgsi/tgsi_lowering.c
index f3b5ade226..fa9d579f77 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_lowering.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_lowering.c
@@ -258,65 +258,6 @@ transform_dst(struct tgsi_transform_context *tctx,
    }
 }
 
-/* XPD - Cross Product
- *   dst.x = src0.y \times src1.z - src1.y \times src0.z
- *   dst.y = src0.z \times src1.x - src1.z \times src0.x
- *   dst.z = src0.x \times src1.y - src1.x \times src0.y
- *   dst.w = 1.0
- *
- * ; needs: 1 tmp, imm{1.0}
- * MUL tmpA.xyz, src1.yzx, src0.zxy
- * MAD dst.xyz, src0.yzx, src1.zxy, -tmpA.xyz
- * MOV dst.w, imm{1.0}
- */
-#define XPD_GROW (NINST(2) + NINST(3) + NINST(1) - OINST(2))
-#define XPD_TMP  1
-static void
-transform_xpd(struct tgsi_transform_context *tctx,
-              struct tgsi_full_instruction *inst)
-{
-   struct tgsi_lowering_context *ctx = tgsi_lowering_context(tctx);
-   struct tgsi_full_dst_register *dst  = &inst->Dst[0];
-   struct tgsi_full_src_register *src0 = &inst->Src[0];
-   struct tgsi_full_src_register *src1 = &inst->Src[1];
-   struct tgsi_full_instruction new_inst;
-
-   if (dst->Register.WriteMask & TGSI_WRITEMASK_XYZ) {
-      /* MUL tmpA.xyz, src1.yzx, src0.zxy */
-      new_inst = tgsi_default_full_instruction();
-      new_inst.Instruction.Opcode = TGSI_OPCODE_MUL;
-      new_inst.Instruction.NumDstRegs = 1;
-      reg_dst(&new_inst.Dst[0], &ctx->tmp[A].dst, TGSI_WRITEMASK_XYZ);
-      new_inst.Instruction.NumSrcRegs = 2;
-      reg_src(&new_inst.Src[0], src1, SWIZ(Y, Z, X, _));
-      reg_src(&new_inst.Src[1], src0, SWIZ(Z, X, Y, _));
-      tctx->emit_instruction(tctx, &new_inst);
-
-      /* MAD dst.xyz, src0.yzx, src1.zxy, -tmpA.xyz */
-      new_inst = tgsi_default_full_instruction();
-      new_inst.Instruction.Opcode = TGSI_OPCODE_MAD;
-      new_inst.Instruction.NumDstRegs = 1;
-      reg_dst(&new_inst.Dst[0], dst, TGSI_WRITEMASK_XYZ);
-      new_inst.Instruction.NumSrcRegs = 3;
-      reg_src(&new_inst.Src[0], src0, SWIZ(Y, Z, X, _));
-      reg_src(&new_inst.Src[1], src1, SWIZ(Z, X, Y, _));
-      reg_src(&new_inst.Src[2], &ctx->tmp[A].src, SWIZ(X, Y, Z, _));
-      new_inst.Src[2].Register.Negate = true;
-      tctx->emit_instruction(tctx, &new_inst);
-   }
-
-   if (dst->Register.WriteMask & TGSI_WRITEMASK_W) {
-      /* MOV dst.w, imm{1.0} */
-      new_inst = tgsi_default_full_instruction();
-      new_inst.Instruction.Opcode = TGSI_OPCODE_MOV;
-      new_inst.Instruction.NumDstRegs = 1;
-      reg_dst(&new_inst.Dst[0], dst, TGSI_WRITEMASK_W);
-      new_inst.Instruction.NumSrcRegs = 1;
-      reg_src(&new_inst.Src[0], &ctx->imm, SWIZ(_, _, _, Y));
-      tctx->emit_instruction(tctx, &new_inst);
-   }
-}
-
 /* SCS - Sine Cosine
  *   dst.x = \cos{src.x}
  *   dst.y = \sin{src.x}
@@ -1466,11 +1407,6 @@ transform_instr(struct tgsi_transform_context *tctx,
          goto skip;
       transform_dst(tctx, inst);
       break;
-   case TGSI_OPCODE_XPD:
-      if (!ctx->config->lower_XPD)
-         goto skip;
-      transform_xpd(tctx, inst);
-      break;
    case TGSI_OPCODE_SCS:
       if (!ctx->config->lower_SCS)
          goto skip;
@@ -1599,7 +1535,6 @@ tgsi_transform_lowering(const struct tgsi_lowering_config *config,
 #define OPCS(x) ((config->lower_ ## x) ? info->opcode_count[TGSI_OPCODE_ ## x] : 0)
    /* if there are no instructions to lower, then we are done: */
    if (!(OPCS(DST) ||
-         OPCS(XPD) ||
          OPCS(SCS) ||
          OPCS(LRP) ||
          OPCS(FRC) ||
@@ -1629,10 +1564,6 @@ tgsi_transform_lowering(const struct tgsi_lowering_config *config,
       newlen += DST_GROW * OPCS(DST);
       numtmp = MAX2(numtmp, DST_TMP);
    }
-   if (OPCS(XPD)) {
-      newlen += XPD_GROW * OPCS(XPD);
-      numtmp = MAX2(numtmp, XPD_TMP);
-   }
    if (OPCS(SCS)) {
       newlen += SCS_GROW * OPCS(SCS);
       numtmp = MAX2(numtmp, SCS_TMP);
diff --git a/src/gallium/auxiliary/tgsi/tgsi_lowering.h b/src/gallium/auxiliary/tgsi/tgsi_lowering.h
index f65d915c65..709a63a601 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_lowering.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_lowering.h
@@ -55,7 +55,6 @@ struct tgsi_lowering_config
     * enable lowering of TGSI_OPCODE_<opc>
     */
    unsigned lower_DST:1;
-   unsigned lower_XPD:1;
    unsigned lower_SCS:1;
    unsigned lower_LRP:1;
    unsigned lower_FRC:1;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h b/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
index 68b4af81b7..111edf3ca0 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
@@ -63,7 +63,6 @@ OP11(ROUND)
 OP11(EX2)
 OP11(LG2)
 OP12(POW)
-OP12(XPD)
 OP11(COS)
 OP11(DDX)
 OP11(DDY)
diff --git a/src/gallium/docs/source/tgsi.rst b/src/gallium/docs/source/tgsi.rst
index 54767a7189..3441907b6c 100644
--- a/src/gallium/docs/source/tgsi.rst
+++ b/src/gallium/docs/source/tgsi.rst
@@ -350,18 +350,6 @@ This instruction replicates its result.
 
   dst = src0.x^{src1.x}
 
-.. opcode:: XPD - Cross Product
-
-.. math::
-
-  dst.x = src0.y \times src1.z - src1.y \times src0.z
-
-  dst.y = src0.z \times src1.x - src1.z \times src0.x
-
-  dst.z = src0.x \times src1.y - src1.x \times src0.y
-
-  dst.w = 1
-
 
 .. opcode:: COS - Cosine
 
@@ -3663,7 +3651,7 @@ of the operands are equal to 0. That means that 0 * Inf = 0. This
 should be set the same way for an entire pipeline. Note that this
 applies not only to the literal MUL TGSI opcode, but all FP32
 multiplications implied by other operations, such as MAD, FMA, DP2,
-DP3, DP4, DST, LOG, LRP, XPD, and possibly others. If there is a
+DP3, DP4, DST, LOG, LRP, and possibly others. If there is a
 mismatch between shaders, then it is unspecified whether this behavior
 will be enabled.
 
diff --git a/src/gallium/drivers/etnaviv/etnaviv_compiler.c b/src/gallium/drivers/etnaviv/etnaviv_compiler.c
index 3ccb737381..4f09f71cc1 100644
--- a/src/gallium/drivers/etnaviv/etnaviv_compiler.c
+++ b/src/gallium/drivers/etnaviv/etnaviv_compiler.c
@@ -2317,7 +2317,6 @@ etna_compile_shader(struct etna_shader_variant *v)
       .lower_LOG = true,
       .lower_DP2 = true,
       .lower_TRUNC = true,
-      .lower_XPD = true
    };
 
    c = CALLOC_STRUCT(etna_compile);
diff --git a/src/gallium/drivers/i915/i915_fpc_optimize.c b/src/gallium/drivers/i915/i915_fpc_optimize.c
index da06e16bf0..fb97e9ea86 100644
--- a/src/gallium/drivers/i915/i915_fpc_optimize.c
+++ b/src/gallium/drivers/i915/i915_fpc_optimize.c
@@ -118,7 +118,6 @@ static const struct {
    [ TGSI_OPCODE_TRUNC   ] = { false,  false,                  0,  1,  1 },
    [ TGSI_OPCODE_TXB     ] = {  true,  false,                  0,  1,  2 },
    [ TGSI_OPCODE_TXP     ] = {  true,  false,                  0,  1,  2 },
-   [ TGSI_OPCODE_XPD     ] = { false,  false,                  0,  1,  2 },
 };
 
 static boolean op_has_dst(unsigned opcode)
diff --git a/src/gallium/drivers/i915/i915_fpc_translate.c b/src/gallium/drivers/i915/i915_fpc_translate.c
index 22a42eeeab..2faab338b7 100644
--- a/src/gallium/drivers/i915/i915_fpc_translate.c
+++ b/src/gallium/drivers/i915/i915_fpc_translate.c
@@ -1027,32 +1027,6 @@ i915_translate_instruction(struct i915_fp_compile *p,
       emit_tex(p, inst, T0_TEXLDP, fs);
       break;
 
-   case TGSI_OPCODE_XPD:
-      /* Cross product:
-       *      result.x = src0.y * src1.z - src0.z * src1.y;
-       *      result.y = src0.z * src1.x - src0.x * src1.z;
-       *      result.z = src0.x * src1.y - src0.y * src1.x;
-       *      result.w = undef;
-       */
-      src0 = src_vector(p, &inst->Src[0], fs);
-      src1 = src_vector(p, &inst->Src[1], fs);
-      tmp = i915_get_utemp(p);
-
-      i915_emit_arith(p,
-                      A0_MUL,
-                      tmp, A0_DEST_CHANNEL_ALL, 0,
-                      swizzle(src0, Z, X, Y, ONE),
-                      swizzle(src1, Y, Z, X, ONE), 0);
-
-      i915_emit_arith(p,
-                      A0_MAD,
-                      get_result_vector(p, &inst->Dst[0]),
-                      get_result_flags(inst), 0,
-                      swizzle(src0, Y, Z, X, ONE),
-                      swizzle(src1, Z, X, Y, ONE),
-                      negate(tmp, 1, 1, 1, 0));
-      break;
-
    default:
       i915_program_error(p, "bad opcode %d", inst->Instruction.Opcode);
       p->error = 1;
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 c06e74e2f8..a862f985fd 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
@@ -346,14 +346,6 @@ unsigned int Instruction::srcMask(unsigned int s) const
       return mask;
    case TGSI_OPCODE_TXQ:
       return 1;
-   case TGSI_OPCODE_XPD:
-   {
-      unsigned int x = 0;
-      if (mask & 1) x |= 0x6;
-      if (mask & 2) x |= 0x5;
-      if (mask & 4) x |= 0x3;
-      return x;
-   }
    case TGSI_OPCODE_D2I:
    case TGSI_OPCODE_D2U:
    case TGSI_OPCODE_D2F:
@@ -3347,25 +3339,6 @@ Converter::handleInstruction(const struct tgsi_full_instruction *insn)
    case TGSI_OPCODE_LIT:
       handleLIT(dst0);
       break;
-   case TGSI_OPCODE_XPD:
-      FOR_EACH_DST_ENABLED_CHANNEL(0, c, tgsi) {
-         if (c < 3) {
-            val0 = getSSA();
-            src0 = fetchSrc(1, (c + 1) % 3);
-            src1 = fetchSrc(0, (c + 2) % 3);
-            mkOp2(OP_MUL, TYPE_F32, val0, src0, src1)
-               ->dnz = info->io.mul_zero_wins;
-            mkOp1(OP_NEG, TYPE_F32, val0, val0);
-
-            src0 = fetchSrc(0, (c + 1) % 3);
-            src1 = fetchSrc(1, (c + 2) % 3);
-            mkOp3(OP_MAD, TYPE_F32, dst0[c], src0, src1, val0)
-               ->dnz = info->io.mul_zero_wins;
-         } else {
-            loadImm(dst0[c], 1.0f);
-         }
-      }
-      break;
    case TGSI_OPCODE_ISSG:
    case TGSI_OPCODE_SSG:
       FOR_EACH_DST_ENABLED_CHANNEL(0, c, tgsi) {
diff --git a/src/gallium/drivers/nouveau/nv30/nv30_vertprog.h b/src/gallium/drivers/nouveau/nv30/nv30_vertprog.h
index 5556e0c77b..89d5b935e9 100644
--- a/src/gallium/drivers/nouveau/nv30/nv30_vertprog.h
+++ b/src/gallium/drivers/nouveau/nv30/nv30_vertprog.h
@@ -10,7 +10,6 @@
  *   POW - EX2 + MUL + LG2
  *   SUB - ADD, second source negated
  *   SWZ - MOV
- *   XPD -
  *
  * Register access
  *   - Only one INPUT can be accessed per-instruction (move extras into TEMPs)
diff --git a/src/gallium/drivers/nouveau/nv30/nvfx_fragprog.c b/src/gallium/drivers/nouveau/nv30/nvfx_fragprog.c
index a3ed5c6526..7d006fb2dd 100644
--- a/src/gallium/drivers/nouveau/nv30/nvfx_fragprog.c
+++ b/src/gallium/drivers/nouveau/nv30/nvfx_fragprog.c
@@ -774,11 +774,6 @@ nvfx_fragprog_parse_instruction(struct nvfx_fpc *fpc,
         case TGSI_OPCODE_TXP:
                 nvfx_fp_emit(fpc, tex(sat, TXP, unit, dst, mask, src[0], none, none));
                 break;
-   case TGSI_OPCODE_XPD:
-      tmp = nvfx_src(temp(fpc));
-      nvfx_fp_emit(fpc, arith(0, MUL, tmp.reg, mask, swz(src[0], Z, X, Y, Y), swz(src[1], Y, Z, X, X), none));
-      nvfx_fp_emit(fpc, arith(sat, MAD, dst, (mask & ~NVFX_FP_MASK_W), swz(src[0], Y, Z, X, X), swz(src[1], Z, X, Y, Y), neg(tmp)));
-      break;
 
    case TGSI_OPCODE_IF:
       // MOVRC0 R31 (TR0.xyzw), R<src>:
diff --git a/src/gallium/drivers/nouveau/nv30/nvfx_shader.h b/src/gallium/drivers/nouveau/nv30/nvfx_shader.h
index e66d8af762..f196c4fc17 100644
--- a/src/gallium/drivers/nouveau/nv30/nvfx_shader.h
+++ b/src/gallium/drivers/nouveau/nv30/nvfx_shader.h
@@ -164,7 +164,6 @@
  *   RSQ - LG2 + EX2
  *   POW - LG2 + MUL + EX2
  *   SCS - COS + SIN
- *   XPD
  *
  * NV40 Looping
  *   Loops appear to be fairly expensive on NV40 at least, the proprietary
diff --git a/src/gallium/drivers/nouveau/nv30/nvfx_vertprog.c b/src/gallium/drivers/nouveau/nv30/nvfx_vertprog.c
index 8ba3d5a050..83823a148b 100644
--- a/src/gallium/drivers/nouveau/nv30/nvfx_vertprog.c
+++ b/src/gallium/drivers/nouveau/nv30/nvfx_vertprog.c
@@ -683,11 +683,6 @@ nvfx_vertprog_parse_instruction(struct nvfx_vpc *vpc,
       insn.cc_test = NVFX_COND_LT;
       nvfx_vp_emit(vpc, insn);
       break;
-   case TGSI_OPCODE_XPD:
-      tmp = nvfx_src(temp(vpc));
-      nvfx_vp_emit(vpc, arith(0, VEC, MUL, tmp.reg, mask, swz(src[0], Z, X, Y, Y), swz(src[1], Y, Z, X, X), none));
-      nvfx_vp_emit(vpc, arith(sat, VEC, MAD, dst, (mask & ~NVFX_VP_MASK_W), swz(src[0], Y, Z, X, X), swz(src[1], Z, X, Y, Y), neg(tmp)));
-      break;
    case TGSI_OPCODE_IF:
       insn = arith(0, VEC, MOV, none.reg, NVFX_VP_MASK_X, src[0], none, none);
       insn.cc_update = 1;
diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.c b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
index 78af1242b4..fa6c0b9104 100644
--- a/src/gallium/drivers/r300/r300_tgsi_to_rc.c
+++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
@@ -57,7 +57,6 @@ static unsigned translate_opcode(unsigned opcode)
         case TGSI_OPCODE_EX2: return RC_OPCODE_EX2;
         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_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 fd76c93de8..18d4bc4c60 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -7844,78 +7844,6 @@ static int tgsi_ucmp(struct r600_shader_ctx *ctx)
 	return 0;
 }
 
-static int tgsi_xpd(struct r600_shader_ctx *ctx)
-{
-	struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
-	static const unsigned int src0_swizzle[] = {2, 0, 1};
-	static const unsigned int src1_swizzle[] = {1, 2, 0};
-	struct r600_bytecode_alu alu;
-	uint32_t use_temp = 0;
-	int i, r;
-
-	if (inst->Dst[0].Register.WriteMask != 0xf)
-		use_temp = 1;
-
-	for (i = 0; i < 4; i++) {
-		memset(&alu, 0, sizeof(struct r600_bytecode_alu));
-		alu.op = ALU_OP2_MUL;
-		if (i < 3) {
-			r600_bytecode_src(&alu.src[0], &ctx->src[0], src0_swizzle[i]);
-			r600_bytecode_src(&alu.src[1], &ctx->src[1], src1_swizzle[i]);
-		} else {
-			alu.src[0].sel = V_SQ_ALU_SRC_0;
-			alu.src[0].chan = i;
-			alu.src[1].sel = V_SQ_ALU_SRC_0;
-			alu.src[1].chan = i;
-		}
-
-		alu.dst.sel = ctx->temp_reg;
-		alu.dst.chan = i;
-		alu.dst.write = 1;
-
-		if (i == 3)
-			alu.last = 1;
-		r = r600_bytecode_add_alu(ctx->bc, &alu);
-		if (r)
-			return r;
-	}
-
-	for (i = 0; i < 4; i++) {
-		memset(&alu, 0, sizeof(struct r600_bytecode_alu));
-		alu.op = ALU_OP3_MULADD;
-
-		if (i < 3) {
-			r600_bytecode_src(&alu.src[0], &ctx->src[0], src1_swizzle[i]);
-			r600_bytecode_src(&alu.src[1], &ctx->src[1], src0_swizzle[i]);
-		} else {
-			alu.src[0].sel = V_SQ_ALU_SRC_0;
-			alu.src[0].chan = i;
-			alu.src[1].sel = V_SQ_ALU_SRC_0;
-			alu.src[1].chan = i;
-		}
-
-		alu.src[2].sel = ctx->temp_reg;
-		alu.src[2].neg = 1;
-		alu.src[2].chan = i;
-
-		if (use_temp)
-			alu.dst.sel = ctx->temp_reg;
-		else
-			tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
-		alu.dst.chan = i;
-		alu.dst.write = 1;
-		alu.is_op3 = 1;
-		if (i == 3)
-			alu.last = 1;
-		r = r600_bytecode_add_alu(ctx->bc, &alu);
-		if (r)
-			return r;
-	}
-	if (use_temp)
-		return tgsi_helper_copy(ctx, inst);
-	return 0;
-}
-
 static int tgsi_exp(struct r600_shader_ctx *ctx)
 {
 	struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
@@ -9092,7 +9020,7 @@ static const struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction[]
 	[TGSI_OPCODE_EX2]	= { ALU_OP1_EXP_IEEE, tgsi_trans_srcx_replicate},
 	[TGSI_OPCODE_LG2]	= { ALU_OP1_LOG_IEEE, tgsi_trans_srcx_replicate},
 	[TGSI_OPCODE_POW]	= { ALU_OP0_NOP, tgsi_pow},
-	[TGSI_OPCODE_XPD]	= { ALU_OP0_NOP, tgsi_xpd},
+	[31]	= { ALU_OP0_NOP, tgsi_unsupported},
 	[32]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[33]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[34]			= { ALU_OP0_NOP, tgsi_unsupported},
@@ -9290,7 +9218,7 @@ static const struct r600_shader_tgsi_instruction eg_shader_tgsi_instruction[] =
 	[TGSI_OPCODE_EX2]	= { ALU_OP1_EXP_IEEE, tgsi_trans_srcx_replicate},
 	[TGSI_OPCODE_LG2]	= { ALU_OP1_LOG_IEEE, tgsi_trans_srcx_replicate},
 	[TGSI_OPCODE_POW]	= { ALU_OP0_NOP, tgsi_pow},
-	[TGSI_OPCODE_XPD]	= { ALU_OP0_NOP, tgsi_xpd},
+	[31]	= { ALU_OP0_NOP, tgsi_unsupported},
 	[32]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[33]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[34]			= { ALU_OP0_NOP, tgsi_unsupported},
@@ -9513,7 +9441,7 @@ static const struct r600_shader_tgsi_instruction cm_shader_tgsi_instruction[] =
 	[TGSI_OPCODE_EX2]	= { ALU_OP1_EXP_IEEE, cayman_emit_float_instr},
 	[TGSI_OPCODE_LG2]	= { ALU_OP1_LOG_IEEE, cayman_emit_float_instr},
 	[TGSI_OPCODE_POW]	= { ALU_OP0_NOP, cayman_pow},
-	[TGSI_OPCODE_XPD]	= { ALU_OP0_NOP, tgsi_xpd},
+	[31]	= { ALU_OP0_NOP, tgsi_unsupported},
 	[32]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[33]			= { ALU_OP0_NOP, tgsi_unsupported},
 	[34]			= { ALU_OP0_NOP, tgsi_unsupported},
diff --git a/src/gallium/drivers/svga/svga_tgsi_insn.c b/src/gallium/drivers/svga/svga_tgsi_insn.c
index 928330ce14..a325a567c1 100644
--- a/src/gallium/drivers/svga/svga_tgsi_insn.c
+++ b/src/gallium/drivers/svga/svga_tgsi_insn.c
@@ -2196,63 +2196,6 @@ emit_pow(struct svga_shader_emitter *emit,
 
 
 /**
- * Translate/emit TGSI XPD (vector cross product) instruction.
- */
-static boolean
-emit_xpd(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] );
-   const struct src_register src1 = translate_src_register(
-      emit, &insn->Src[1] );
-   boolean need_dst_tmp = FALSE;
-
-   /* XPD can only output to a temporary */
-   if (SVGA3dShaderGetRegType(dst.value) != SVGA3DREG_TEMP)
-      need_dst_tmp = TRUE;
-
-   /* The dst reg must not be the same as src0 or src1*/
-   if (alias_src_dst(src0, dst) ||
-       alias_src_dst(src1, dst))
-      need_dst_tmp = TRUE;
-
-   if (need_dst_tmp) {
-      SVGA3dShaderDestToken tmp = get_temp( emit );
-
-      /* Obey DX9 restrictions on mask:
-       */
-      tmp.mask = dst.mask & TGSI_WRITEMASK_XYZ;
-
-      if (!submit_op2(emit, inst_token( SVGA3DOP_CRS ), tmp, src0, src1))
-         return FALSE;
-
-      if (!submit_op1(emit, inst_token( SVGA3DOP_MOV ), dst, src( tmp )))
-         return FALSE;
-   }
-   else {
-      if (!submit_op2(emit, inst_token( SVGA3DOP_CRS ), dst, src0, src1))
-         return FALSE;
-   }
-
-   /* Need to emit 1.0 to dst.w?
-    */
-   if (dst.mask & TGSI_WRITEMASK_W) {
-      struct src_register one = get_one_immediate( emit );
-
-      if (!submit_op1(emit,
-                      inst_token( SVGA3DOP_MOV ),
-                      writemask(dst, TGSI_WRITEMASK_W),
-                      one))
-         return FALSE;
-   }
-
-   return TRUE;
-}
-
-
-/**
  * Emit a LRP (linear interpolation) instruction.
  */
 static boolean
@@ -2986,9 +2929,6 @@ svga_emit_instruction(struct svga_shader_emitter *emit,
    case TGSI_OPCODE_BRK:
       return emit_brk( emit, insn );
 
-   case TGSI_OPCODE_XPD:
-      return emit_xpd( emit, insn );
-
    case TGSI_OPCODE_KILL:
       return emit_kill( emit, insn );
 
@@ -3604,7 +3544,6 @@ needs_to_create_common_immediate(const struct svga_shader_emitter *emit)
        emit->info.opcode_count[TGSI_OPCODE_SEQ] >= 1 ||
        emit->info.opcode_count[TGSI_OPCODE_EXP] >= 1 ||
        emit->info.opcode_count[TGSI_OPCODE_LOG] >= 1 ||
-       emit->info.opcode_count[TGSI_OPCODE_XPD] >= 1 ||
        emit->info.opcode_count[TGSI_OPCODE_KILL] >= 1)
       return TRUE;
 
diff --git a/src/gallium/drivers/svga/svga_tgsi_vgpu10.c b/src/gallium/drivers/svga/svga_tgsi_vgpu10.c
index e57e78d11a..9d86f72ea0 100644
--- a/src/gallium/drivers/svga/svga_tgsi_vgpu10.c
+++ b/src/gallium/drivers/svga/svga_tgsi_vgpu10.c
@@ -5210,117 +5210,6 @@ emit_txp(struct svga_shader_emitter_v10 *emit,
 }
 
 
-/*
- * Emit code for TGSI_OPCODE_XPD instruction.
- */
-static boolean
-emit_xpd(struct svga_shader_emitter_v10 *emit,
-         const struct tgsi_full_instruction *inst)
-{
-   /* dst.x = src0.y * src1.z - src1.y * src0.z
-    * dst.y = src0.z * src1.x - src1.z * src0.x
-    * dst.z = src0.x * src1.y - src1.x * src0.y
-    * dst.w = 1
-    */
-   struct tgsi_full_src_register s0_xxxx =
-      scalar_src(&inst->Src[0], TGSI_SWIZZLE_X);
-   struct tgsi_full_src_register s0_yyyy =
-      scalar_src(&inst->Src[0], TGSI_SWIZZLE_Y);
-   struct tgsi_full_src_register s0_zzzz =
-      scalar_src(&inst->Src[0], TGSI_SWIZZLE_Z);
-
-   struct tgsi_full_src_register s1_xxxx =
-      scalar_src(&inst->Src[1], TGSI_SWIZZLE_X);
-   struct tgsi_full_src_register s1_yyyy =
-      scalar_src(&inst->Src[1], TGSI_SWIZZLE_Y);
-   struct tgsi_full_src_register s1_zzzz =
-      scalar_src(&inst->Src[1], TGSI_SWIZZLE_Z);
-
-   unsigned tmp1 = get_temp_index(emit);
-   struct tgsi_full_src_register tmp1_src = make_src_temp_reg(tmp1);
-   struct tgsi_full_dst_register tmp1_dst = make_dst_temp_reg(tmp1);
-
-   unsigned tmp2 = get_temp_index(emit);
-   struct tgsi_full_src_register tmp2_src = make_src_temp_reg(tmp2);
-   struct tgsi_full_dst_register tmp2_dst = make_dst_temp_reg(tmp2);
-   struct tgsi_full_src_register neg_tmp2_src = negate_src(&tmp2_src);
-
-   unsigned tmp3 = get_temp_index(emit);
-   struct tgsi_full_src_register tmp3_src = make_src_temp_reg(tmp3);
-   struct tgsi_full_dst_register tmp3_dst = make_dst_temp_reg(tmp3);
-   struct tgsi_full_dst_register tmp3_dst_x =
-      writemask_dst(&tmp3_dst, TGSI_WRITEMASK_X);
-   struct tgsi_full_dst_register tmp3_dst_y =
-      writemask_dst(&tmp3_dst, TGSI_WRITEMASK_Y);
-   struct tgsi_full_dst_register tmp3_dst_z =
-      writemask_dst(&tmp3_dst, TGSI_WRITEMASK_Z);
-   struct tgsi_full_dst_register tmp3_dst_w =
-      writemask_dst(&tmp3_dst, TGSI_WRITEMASK_W);
-
-   /* Note: we put all the intermediate computations into tmp3 in case
-    * the XPD dest register is that same as one of the src regs (in which
-    * case we could clobber a src reg before we're done with it) .
-    *
-    * Note: we could get by with just one temp register instead of three
-    * since we're doing scalar operations and there's enough room in one
-    * temp for everything.
-    */
-
-   /* MUL tmp1, src0.y, src1.z */
-   /* MUL tmp2, src1.y, src0.z */
-   /* ADD tmp3.x, tmp1, -tmp2 */
-   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
-      emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp1_dst,
-                           &s0_yyyy, &s1_zzzz, FALSE);
-      emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp2_dst,
-                           &s1_yyyy, &s0_zzzz, FALSE);
-      emit_instruction_op2(emit, VGPU10_OPCODE_ADD, &tmp3_dst_x,
-                           &tmp1_src, &neg_tmp2_src, FALSE);
-   }
-
-   /* MUL tmp1, src0.z, src1.x */
-   /* MUL tmp2, src1.z, src0.x */
-   /* ADD tmp3.y, tmp1, -tmp2 */
-   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
-      emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp1_dst, &s0_zzzz,
-                           &s1_xxxx, FALSE);
-      emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp2_dst, &s1_zzzz,
-                           &s0_xxxx, FALSE);
-      emit_instruction_op2(emit, VGPU10_OPCODE_ADD, &tmp3_dst_y,
-                           &tmp1_src, &neg_tmp2_src, FALSE);
-   }
-
-   /* MUL tmp1, src0.x, src1.y */
-   /* MUL tmp2, src1.x, src0.y */
-   /* ADD tmp3.z, tmp1, -tmp2 */
-   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
-      emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp1_dst, &s0_xxxx,
-                           &s1_yyyy, FALSE);
-      emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp2_dst, &s1_xxxx,
-                           &s0_yyyy, FALSE);
-      emit_instruction_op2(emit, VGPU10_OPCODE_ADD, &tmp3_dst_z,
-                           &tmp1_src, &neg_tmp2_src, FALSE);
-   }
-
-   /* MOV tmp3.w, 1.0 */
-   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
-      struct tgsi_full_src_register one =
-         make_immediate_reg_float(emit, 1.0f);
-
-      emit_instruction_op1(emit, VGPU10_OPCODE_MOV, &tmp3_dst_w, &one, FALSE);
-   }
-
-   /* MOV dst, tmp3 */
-   emit_instruction_op1(emit, VGPU10_OPCODE_MOV, &inst->Dst[0], &tmp3_src,
-                        inst->Instruction.Saturate);
-
-
-   free_temp_indexes(emit);
-
-   return TRUE;
-}
-
-
 /**
  * Emit code for TGSI_OPCODE_TXD (explicit derivatives)
  */
@@ -5742,8 +5631,6 @@ emit_vgpu10_instruction(struct svga_shader_emitter_v10 *emit,
       return emit_txq(emit, inst);
    case TGSI_OPCODE_UIF:
       return emit_if(emit, inst);
-   case TGSI_OPCODE_XPD:
-      return emit_xpd(emit, inst);
    case TGSI_OPCODE_UMUL_HI:
    case TGSI_OPCODE_IMUL_HI:
    case TGSI_OPCODE_UDIV:
diff --git a/src/gallium/include/pipe/p_shader_tokens.h b/src/gallium/include/pipe/p_shader_tokens.h
index af0d0252c6..c8f31be16b 100644
--- a/src/gallium/include/pipe/p_shader_tokens.h
+++ b/src/gallium/include/pipe/p_shader_tokens.h
@@ -368,7 +368,7 @@ struct tgsi_property_data {
 #define TGSI_OPCODE_EX2                 28
 #define TGSI_OPCODE_LG2                 29
 #define TGSI_OPCODE_POW                 30
-#define TGSI_OPCODE_XPD                 31
+/* gap */
 #define TGSI_OPCODE_U2I64               32
 #define TGSI_OPCODE_CLOCK               33
 #define TGSI_OPCODE_I2I64               34
diff --git a/src/gallium/state_trackers/nine/nine_shader.c b/src/gallium/state_trackers/nine/nine_shader.c
index f405090811..60d56af83f 100644
--- a/src/gallium/state_trackers/nine/nine_shader.c
+++ b/src/gallium/state_trackers/nine/nine_shader.c
@@ -1588,6 +1588,29 @@ DECL_SPECIAL(ABS)
     return D3D_OK;
 }
 
+DECL_SPECIAL(XPD)
+{
+    struct ureg_program *ureg = tx->ureg;
+    struct ureg_dst dst = tx_dst_param(tx, &tx->insn.dst[0]);
+    struct ureg_src src0 = tx_src_param(tx, &tx->insn.src[0]);
+    struct ureg_src src1 = tx_src_param(tx, &tx->insn.src[1]);
+
+    ureg_MUL(ureg, ureg_writemask(dst, TGSI_WRITEMASK_XYZ),
+             ureg_swizzle(src0, TGSI_SWIZZLE_Y, TGSI_SWIZZLE_Z,
+                          TGSI_SWIZZLE_X, 0),
+             ureg_swizzle(src1, TGSI_SWIZZLE_Z, TGSI_SWIZZLE_X,
+                          TGSI_SWIZZLE_Y, 0));
+    ureg_MAD(ureg, ureg_writemask(dst, TGSI_WRITEMASK_XYZ),
+             ureg_swizzle(src0, TGSI_SWIZZLE_Z, TGSI_SWIZZLE_X,
+                          TGSI_SWIZZLE_Y, 0),
+             ureg_negate(ureg_swizzle(src1, TGSI_SWIZZLE_Y,
+                                      TGSI_SWIZZLE_Z, TGSI_SWIZZLE_X, 0)),
+             ureg_src(dst));
+    ureg_MOV(ureg, ureg_writemask(dst, TGSI_WRITEMASK_W),
+             ureg_imm1f(ureg, 1));
+    return D3D_OK;
+}
+
 DECL_SPECIAL(M4x4)
 {
     return NineTranslateInstruction_Mkxn(tx, 4, 4);
@@ -2915,7 +2938,7 @@ struct sm1_op_info inst_table[] =
     _OPI(DCL, NOP, V(0,0), V(3,0), V(0,0), V(3,0), 0, 0, SPECIAL(DCL)),
 
     _OPI(POW, POW, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, SPECIAL(POW)),
-    _OPI(CRS, XPD, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, NULL), /* XXX: .w */
+    _OPI(CRS, NOP, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, SPECIAL(XPD)), /* XXX: .w */
     _OPI(SGN, SSG, V(2,0), V(3,0), V(0,0), V(0,0), 1, 3, SPECIAL(SGN)), /* ignore src1,2 */
     _OPI(ABS, NOP, V(0,0), V(3,0), V(0,0), V(3,0), 1, 1, SPECIAL(ABS)),
     _OPI(NRM, NOP, V(0,0), V(3,0), V(0,0), V(3,0), 1, 1, SPECIAL(NRM)), /* NRM doesn't fit */
diff --git a/src/mesa/state_tracker/st_mesa_to_tgsi.c b/src/mesa/state_tracker/st_mesa_to_tgsi.c
index 5f78026dda..d2ec23ee61 100644
--- a/src/mesa/state_tracker/st_mesa_to_tgsi.c
+++ b/src/mesa/state_tracker/st_mesa_to_tgsi.c
@@ -505,8 +505,6 @@ translate_opcode( unsigned op )
       return TGSI_OPCODE_TXB;
    case OPCODE_TXP:
       return TGSI_OPCODE_TXP;
-   case OPCODE_XPD:
-      return TGSI_OPCODE_XPD;
    case OPCODE_END:
       return TGSI_OPCODE_END;
    default:
@@ -568,11 +566,17 @@ compile_instruction(
       break;
 
    case OPCODE_XPD:
-      dst[0] = ureg_writemask(dst[0], TGSI_WRITEMASK_XYZ );
-      ureg_insn( ureg, 
-                 translate_opcode( inst->Opcode ), 
-                 dst, num_dst, 
-                 src, num_src, 0 );
+      ureg_MUL(ureg, ureg_writemask(dst[0], TGSI_WRITEMASK_XYZ),
+               ureg_swizzle(src[0], TGSI_SWIZZLE_Y, TGSI_SWIZZLE_Z,
+                            TGSI_SWIZZLE_X, 0),
+               ureg_swizzle(src[1], TGSI_SWIZZLE_Z, TGSI_SWIZZLE_X,
+                            TGSI_SWIZZLE_Y, 0));
+      ureg_MAD(ureg, ureg_writemask(dst[0], TGSI_WRITEMASK_XYZ),
+               ureg_swizzle(src[0], TGSI_SWIZZLE_Z, TGSI_SWIZZLE_X,
+                            TGSI_SWIZZLE_Y, 0),
+               ureg_negate(ureg_swizzle(src[1], TGSI_SWIZZLE_Y,
+                                        TGSI_SWIZZLE_Z, TGSI_SWIZZLE_X, 0)),
+               ureg_src(dst[0]));
       break;
 
    case OPCODE_RSQ: