gallium: remove TGSI_OPCODE_SUB

It's redundant with the source modifier.

Reviewed-by: Nicolai Hähnle <[email protected]>

17 files changed, 61 insertions, 94 deletions

diff --git a/src/gallium/auxiliary/draw/draw_pipe_aaline.c b/src/gallium/auxiliary/draw/draw_pipe_aaline.c
index c236caa3edd..57ca12e2a2a 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_aaline.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_aaline.c
@@ -285,7 +285,7 @@ aa_transform_epilog(struct tgsi_transform_context *ctx)
                               TGSI_FILE_OUTPUT, aactx->colorOutput,
                               TGSI_WRITEMASK_W,
                               TGSI_FILE_TEMPORARY, aactx->colorTemp,
-                              TGSI_FILE_TEMPORARY, aactx->texTemp);
+                              TGSI_FILE_TEMPORARY, aactx->texTemp, false);
    }
 }
 
diff --git a/src/gallium/auxiliary/draw/draw_pipe_aapoint.c b/src/gallium/auxiliary/draw/draw_pipe_aapoint.c
index 33ef8ec1702..2b96b8ad446 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_aapoint.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_aapoint.c
@@ -213,13 +213,13 @@ aa_transform_prolog(struct tgsi_transform_context *ctx)
    tgsi_transform_op2_inst(ctx, TGSI_OPCODE_MUL,
                            TGSI_FILE_TEMPORARY, tmp0, TGSI_WRITEMASK_XY,
                            TGSI_FILE_INPUT, texInput,
-                           TGSI_FILE_INPUT, texInput);
+                           TGSI_FILE_INPUT, texInput, false);
 
    /* ADD t0.x, t0.x, t0.y;  # x^2 + y^2 */
    tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_ADD,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_WRITEMASK_X,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_X,
-                               TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_Y);
+                               TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_Y, false);
 
 #if NORMALIZE  /* OPTIONAL normalization of length */
    /* RSQ t0.x, t0.x; */
@@ -237,7 +237,7 @@ aa_transform_prolog(struct tgsi_transform_context *ctx)
    tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_SGT,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_WRITEMASK_Y,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_X,
-                               TGSI_FILE_INPUT, texInput, TGSI_SWIZZLE_W);
+                               TGSI_FILE_INPUT, texInput, TGSI_SWIZZLE_W, false);
 
    /* KILL_IF -tmp0.yyyy;   # if -tmp0.y < 0, KILL */
    tgsi_transform_kill_inst(ctx, TGSI_FILE_TEMPORARY, tmp0,
@@ -246,10 +246,10 @@ aa_transform_prolog(struct tgsi_transform_context *ctx)
    /* compute coverage factor = (1-d)/(1-k) */
 
    /* SUB t0.z, tex.w, tex.z;  # m = 1 - k */
-   tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_SUB,
+   tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_ADD,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_WRITEMASK_Z,
                                TGSI_FILE_INPUT, texInput, TGSI_SWIZZLE_W,
-                               TGSI_FILE_INPUT, texInput, TGSI_SWIZZLE_Z);
+                               TGSI_FILE_INPUT, texInput, TGSI_SWIZZLE_Z, true);
 
    /* RCP t0.z, t0.z;  # t0.z = 1 / m */
    newInst = tgsi_default_full_instruction();
@@ -265,22 +265,22 @@ aa_transform_prolog(struct tgsi_transform_context *ctx)
    ctx->emit_instruction(ctx, &newInst);
 
    /* SUB t0.y, 1, t0.x;  # d = 1 - d */
-   tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_SUB,
+   tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_ADD,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_WRITEMASK_Y,
                                TGSI_FILE_INPUT, texInput, TGSI_SWIZZLE_W,
-                               TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_X);
+                               TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_X, true);
 
    /* MUL t0.w, t0.y, t0.z;   # coverage = d * m */
    tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_MUL,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_WRITEMASK_W,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_Y,
-                               TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_Z);
+                               TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_Z, false);
 
    /* SLE t0.y, t0.x, tex.z;  # bool b = distance <= k */
    tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_SLE,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_WRITEMASK_Y,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_X,
-                               TGSI_FILE_INPUT, texInput, TGSI_SWIZZLE_Z);
+                               TGSI_FILE_INPUT, texInput, TGSI_SWIZZLE_Z, false);
 
    /* CMP t0.w, -t0.y, tex.w, t0.w;
     *  # if -t0.y < 0 then
@@ -318,7 +318,7 @@ aa_transform_epilog(struct tgsi_transform_context *ctx)
                            TGSI_FILE_OUTPUT, aactx->colorOutput,
                            TGSI_WRITEMASK_W,
                            TGSI_FILE_TEMPORARY, aactx->colorTemp,
-                           TGSI_FILE_TEMPORARY, aactx->tmp0);
+                           TGSI_FILE_TEMPORARY, aactx->tmp0, false);
 }
 
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c
index 7d939e8fe4d..91e959f3451 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c
@@ -368,8 +368,8 @@ exp_emit(
                                        TGSI_OPCODE_EX2, floor_x);
 
    /* src0.x - floor( src0.x ) */
-   emit_data->output[TGSI_CHAN_Y] = lp_build_emit_llvm_binary(bld_base,
-                   TGSI_OPCODE_SUB,  emit_data->args[0] /* src0.x */, floor_x);
+   emit_data->output[TGSI_CHAN_Y] =
+      lp_build_sub(&bld_base->base, emit_data->args[0] /* src0.x */, floor_x);
 
    /* 2 ^ src0.x */
    emit_data->output[TGSI_CHAN_Z] = lp_build_emit_llvm_unary(bld_base,
@@ -394,8 +394,8 @@ frc_emit(
    LLVMValueRef tmp;
    tmp = lp_build_emit_llvm_unary(bld_base, TGSI_OPCODE_FLR,
                                   emit_data->args[0]);
-   emit_data->output[emit_data->chan] = lp_build_emit_llvm_binary(bld_base,
-                                       TGSI_OPCODE_SUB, emit_data->args[0], tmp);
+   emit_data->output[emit_data->chan] =
+      lp_build_sub(&bld_base->base, emit_data->args[0], tmp);
 }
 
 /* TGSI_OPCODE_KILL_IF */
@@ -770,19 +770,6 @@ const struct lp_build_tgsi_action scs_action = {
    scs_emit	 /* emit */
 };
 
-/* TGSI_OPCODE_SUB */
-static void
-sub_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[emit_data->chan] =
-      LLVMBuildFSub(bld_base->base.gallivm->builder,
-                    emit_data->args[0],
-                    emit_data->args[1], "");
-}
-
 /* TGSI_OPCODE_F2U */
 static void
 f2u_emit(
@@ -956,7 +943,7 @@ xpd_helper(
    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_emit_llvm_binary(bld_base, TGSI_OPCODE_SUB, tmp0, tmp1);
+   return lp_build_sub(&bld_base->base, tmp0, tmp1);
 }
 
 static void
@@ -1352,7 +1339,6 @@ lp_set_default_actions(struct lp_build_tgsi_context * bld_base)
    bld_base->op_actions[TGSI_OPCODE_MUL].emit = mul_emit;
    bld_base->op_actions[TGSI_OPCODE_DIV].emit = fdiv_emit;
    bld_base->op_actions[TGSI_OPCODE_RCP].emit = rcp_emit;
-   bld_base->op_actions[TGSI_OPCODE_SUB].emit = sub_emit;
 
    bld_base->op_actions[TGSI_OPCODE_UARL].emit = mov_emit;
    bld_base->op_actions[TGSI_OPCODE_F2U].emit = f2u_emit;
@@ -2071,19 +2057,6 @@ ssg_emit_cpu(
                                                        emit_data->args[0]);
 }
 
-/* TGSI_OPCODE_SUB (CPU Only) */
-
-static void
-sub_emit_cpu(
-   const struct lp_build_tgsi_action * action,
-   struct lp_build_tgsi_context * bld_base,
-   struct lp_build_emit_data * emit_data)
-{
-   emit_data->output[emit_data->chan] = lp_build_sub(&bld_base->base,
-                                                        emit_data->args[0],
-                                                        emit_data->args[1]);
-}
-
 /* TGSI_OPCODE_TRUNC (CPU Only) */
 
 static void
@@ -2624,7 +2597,6 @@ lp_set_default_actions_cpu(
    bld_base->op_actions[TGSI_OPCODE_SLT].emit = slt_emit_cpu;
    bld_base->op_actions[TGSI_OPCODE_SNE].emit = sne_emit_cpu;
    bld_base->op_actions[TGSI_OPCODE_SSG].emit = ssg_emit_cpu;
-   bld_base->op_actions[TGSI_OPCODE_SUB].emit = sub_emit_cpu;
    bld_base->op_actions[TGSI_OPCODE_TRUNC].emit = trunc_emit_cpu;
 
    bld_base->rsq_action.emit = recip_sqrt_emit_cpu;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c
index a5e439feb09..6c177b02cb1 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c
@@ -591,12 +591,6 @@ lp_emit_instruction_aos(
       dst0 = lp_build_add(&bld->bld_base.base, tmp0, src2);
       break;
 
-   case TGSI_OPCODE_SUB:
-      src0 = lp_build_emit_fetch(&bld->bld_base, inst, 0, LP_CHAN_ALL);
-      src1 = lp_build_emit_fetch(&bld->bld_base, inst, 1, LP_CHAN_ALL);
-      dst0 = lp_build_sub(&bld->bld_base.base, src0, src1);
-      break;
-
    case TGSI_OPCODE_LRP:
       src0 = lp_build_emit_fetch(&bld->bld_base, inst, 0, LP_CHAN_ALL);
       src1 = lp_build_emit_fetch(&bld->bld_base, inst, 1, LP_CHAN_ALL);
diff --git a/src/gallium/auxiliary/nir/tgsi_to_nir.c b/src/gallium/auxiliary/nir/tgsi_to_nir.c
index af4a6e034b6..f3e87005122 100644
--- a/src/gallium/auxiliary/nir/tgsi_to_nir.c
+++ b/src/gallium/auxiliary/nir/tgsi_to_nir.c
@@ -1535,7 +1535,6 @@ static const nir_op op_trans[TGSI_OPCODE_LAST] = {
    [TGSI_OPCODE_SLT] = nir_op_slt,
    [TGSI_OPCODE_SGE] = nir_op_sge,
    [TGSI_OPCODE_MAD] = nir_op_ffma,
-   [TGSI_OPCODE_SUB] = nir_op_fsub,
    [TGSI_OPCODE_LRP] = 0,
    [TGSI_OPCODE_SQRT] = nir_op_fsqrt,
    [TGSI_OPCODE_DP2A] = 0,
diff --git a/src/gallium/auxiliary/tgsi/tgsi_aa_point.c b/src/gallium/auxiliary/tgsi/tgsi_aa_point.c
index 9016effd388..4b14a2fc98f 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_aa_point.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_aa_point.c
@@ -148,16 +148,16 @@ aa_prolog(struct tgsi_transform_context *ctx)
    tmp0 = ts->tmp;
 
    /* SUB t0.xy, texIn, (0.5, 0,5) */
-   tgsi_transform_op2_inst(ctx, TGSI_OPCODE_SUB,
+   tgsi_transform_op2_inst(ctx, TGSI_OPCODE_ADD,
                            TGSI_FILE_TEMPORARY, tmp0, TGSI_WRITEMASK_XY,
                            TGSI_FILE_INPUT, texIn,
-                           TGSI_FILE_IMMEDIATE, imm);
+                           TGSI_FILE_IMMEDIATE, imm, true);
 
    /* DP2 t0.x, t0.xy, t0.xy;  # t0.x = x^2 + y^2 */
    tgsi_transform_op2_inst(ctx, TGSI_OPCODE_DP2,
                            TGSI_FILE_TEMPORARY, tmp0, TGSI_WRITEMASK_X,
                            TGSI_FILE_TEMPORARY, tmp0,
-                           TGSI_FILE_TEMPORARY, tmp0);
+                           TGSI_FILE_TEMPORARY, tmp0, false);
 
    /* SQRT t0.x, t0.x */
    tgsi_transform_op1_inst(ctx, TGSI_OPCODE_SQRT,
@@ -167,22 +167,22 @@ aa_prolog(struct tgsi_transform_context *ctx)
    /* compute coverage factor = (0.5-d)/(0.5-k) */
 
    /* SUB t0.w, 0.5, texIn.z;  # t0.w = 0.5-k */
-   tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_SUB,
+   tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_ADD,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_WRITEMASK_W,
                                TGSI_FILE_IMMEDIATE, imm, TGSI_SWIZZLE_X,
-                               TGSI_FILE_INPUT, texIn, TGSI_SWIZZLE_Z);
+                               TGSI_FILE_INPUT, texIn, TGSI_SWIZZLE_Z, true);
 
    /* SUB t0.y, 0.5, t0.x;  # t0.y = 0.5-d */
-   tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_SUB,
+   tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_ADD,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_WRITEMASK_Y,
                                TGSI_FILE_IMMEDIATE, imm, TGSI_SWIZZLE_X,
-                               TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_X);
+                               TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_X, true);
 
    /* DIV t0.w, t0.y, t0.w;  # coverage = (0.5-d)/(0.5-k) */
    tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_DIV,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_WRITEMASK_W,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_Y,
-                               TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_W);
+                               TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_W, false);
 
    /* If the coverage value is negative, it means the fragment is outside
     * the point's circular boundary.  Kill it.
@@ -198,7 +198,7 @@ aa_prolog(struct tgsi_transform_context *ctx)
    tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_MIN,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_WRITEMASK_W,
                                TGSI_FILE_TEMPORARY, tmp0, TGSI_SWIZZLE_W,
-                               TGSI_FILE_IMMEDIATE, imm, TGSI_SWIZZLE_W);
+                               TGSI_FILE_IMMEDIATE, imm, TGSI_SWIZZLE_W, false);
 }
 
 /**
@@ -249,7 +249,7 @@ aa_epilog(struct tgsi_transform_context *ctx)
                            TGSI_FILE_OUTPUT, ts->color_out,
                            TGSI_WRITEMASK_W,
                            TGSI_FILE_TEMPORARY, ts->color_tmp,
-                           TGSI_FILE_TEMPORARY, ts->tmp);
+                           TGSI_FILE_TEMPORARY, ts->tmp, false);
 }
 
 /**
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 2f89de663a4..915cd10e683 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -5208,10 +5208,6 @@ exec_instruction(
       exec_vector_trinary(mach, inst, micro_mad, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
       break;
 
-   case TGSI_OPCODE_SUB:
-      exec_vector_binary(mach, inst, micro_sub, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
-      break;
-
    case TGSI_OPCODE_LRP:
       exec_vector_trinary(mach, inst, micro_lrp, 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 9b2431fe6b1..a339ec281c7 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_info.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_info.c
@@ -54,7 +54,7 @@ static const struct tgsi_opcode_info opcode_info[TGSI_OPCODE_LAST] =
    { 1, 2, 0, 0, 0, 0, 0, COMP, "SLT", TGSI_OPCODE_SLT },
    { 1, 2, 0, 0, 0, 0, 0, COMP, "SGE", TGSI_OPCODE_SGE },
    { 1, 3, 0, 0, 0, 0, 0, COMP, "MAD", TGSI_OPCODE_MAD },
-   { 1, 2, 0, 0, 0, 0, 0, COMP, "SUB", TGSI_OPCODE_SUB },
+   { 1, 2, 0, 0, 0, 0, 0, COMP, "", 17 }, /* removed */
    { 1, 3, 0, 0, 0, 0, 0, COMP, "LRP", TGSI_OPCODE_LRP },
    { 1, 3, 0, 0, 0, 0, 0, COMP, "FMA", TGSI_OPCODE_FMA },
    { 1, 1, 0, 0, 0, 0, 0, REPL, "SQRT", TGSI_OPCODE_SQRT },
diff --git a/src/gallium/auxiliary/tgsi/tgsi_lowering.c b/src/gallium/auxiliary/tgsi/tgsi_lowering.c
index b0a28f27107..bf6cbb361a9 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_lowering.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_lowering.c
@@ -468,12 +468,13 @@ transform_frc(struct tgsi_transform_context *tctx,
 
       /* SUB dst, src, tmpA */
       new_inst = tgsi_default_full_instruction();
-      new_inst.Instruction.Opcode = TGSI_OPCODE_SUB;
+      new_inst.Instruction.Opcode = TGSI_OPCODE_ADD;
       new_inst.Instruction.NumDstRegs = 1;
       reg_dst(&new_inst.Dst[0], dst, TGSI_WRITEMASK_XYZW);
       new_inst.Instruction.NumSrcRegs = 2;
       reg_src(&new_inst.Src[0], src, SWIZ(X, Y, Z, W));
       reg_src(&new_inst.Src[1], &ctx->tmp[A].src, SWIZ(X, Y, Z, W));
+      new_inst.Src[1].Register.Negate = 1;
       tctx->emit_instruction(tctx, &new_inst);
    }
 }
@@ -689,12 +690,13 @@ transform_exp(struct tgsi_transform_context *tctx,
 
          /* SUB tmpA.x, src.x, tmpA.x */
          new_inst = tgsi_default_full_instruction();
-         new_inst.Instruction.Opcode = TGSI_OPCODE_SUB;
+         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], src, SWIZ(X, _, _, _));
          reg_src(&new_inst.Src[1], &ctx->tmp[A].src, SWIZ(X, _, _, _));
+         new_inst.Src[1].Register.Negate = 1;
          tctx->emit_instruction(tctx, &new_inst);
      } else {
          /* FLR tmpA.x, src.x */
@@ -722,12 +724,13 @@ transform_exp(struct tgsi_transform_context *tctx,
    if (dst->Register.WriteMask & TGSI_WRITEMASK_Y) {
       /* SUB dst.y, src.x, tmpA.x */
       new_inst = tgsi_default_full_instruction();
-      new_inst.Instruction.Opcode = TGSI_OPCODE_SUB;
+      new_inst.Instruction.Opcode = TGSI_OPCODE_ADD;
       new_inst.Instruction.NumDstRegs = 1;
       reg_dst(&new_inst.Dst[0], dst, TGSI_WRITEMASK_Y);
       new_inst.Instruction.NumSrcRegs = 2;
       reg_src(&new_inst.Src[0], src, SWIZ(_, X, _, _));
       reg_src(&new_inst.Src[1], &ctx->tmp[A].src, SWIZ(_, X, _, _));
+      new_inst.Src[1].Register.Negate = 1;
       tctx->emit_instruction(tctx, &new_inst);
    }
 
@@ -822,12 +825,13 @@ transform_log(struct tgsi_transform_context *tctx,
 
          /* SUB tmpA.y, tmpA.x, tmpA.y */
          new_inst = tgsi_default_full_instruction();
-         new_inst.Instruction.Opcode = TGSI_OPCODE_SUB;
+         new_inst.Instruction.Opcode = TGSI_OPCODE_ADD;
          new_inst.Instruction.NumDstRegs = 1;
          reg_dst(&new_inst.Dst[0], &ctx->tmp[A].dst, TGSI_WRITEMASK_Y);
          new_inst.Instruction.NumSrcRegs = 2;
          reg_src(&new_inst.Src[0], &ctx->tmp[A].src, SWIZ(_, X, _, _));
          reg_src(&new_inst.Src[1], &ctx->tmp[A].src, SWIZ(_, Y, _, _));
+         new_inst.Src[1].Register.Negate = 1;
          tctx->emit_instruction(tctx, &new_inst);
       } else {
          /* FLR tmpA.y, tmpA.x */
@@ -1072,15 +1076,14 @@ transform_flr_ceil(struct tgsi_transform_context *tctx,
 
       /* FLR: SUB dst, src, tmpA  CEIL: ADD dst, src, tmpA */
       new_inst = tgsi_default_full_instruction();
-      if (opcode == TGSI_OPCODE_CEIL)
-         new_inst.Instruction.Opcode = TGSI_OPCODE_ADD;
-      else
-         new_inst.Instruction.Opcode = TGSI_OPCODE_SUB;
+      new_inst.Instruction.Opcode = TGSI_OPCODE_ADD;
       new_inst.Instruction.NumDstRegs = 1;
       reg_dst(&new_inst.Dst[0], dst, TGSI_WRITEMASK_XYZW);
       new_inst.Instruction.NumSrcRegs = 2;
       reg_src(&new_inst.Src[0], src0, SWIZ(X, Y, Z, W));
       reg_src(&new_inst.Src[1], &ctx->tmp[A].src, SWIZ(X, Y, Z, W));
+      if (opcode == TGSI_OPCODE_FLR)
+         new_inst.Src[1].Register.Negate = 1;
       tctx->emit_instruction(tctx, &new_inst);
    }
 }
@@ -1124,7 +1127,7 @@ transform_trunc(struct tgsi_transform_context *tctx,
          tctx->emit_instruction(tctx, &new_inst);
 
          new_inst = tgsi_default_full_instruction();
-         new_inst.Instruction.Opcode = TGSI_OPCODE_SUB;
+         new_inst.Instruction.Opcode = TGSI_OPCODE_ADD;
          new_inst.Instruction.NumDstRegs = 1;
          reg_dst(&new_inst.Dst[0], &ctx->tmp[A].dst, TGSI_WRITEMASK_XYZW);
          new_inst.Instruction.NumSrcRegs = 2;
@@ -1132,6 +1135,7 @@ transform_trunc(struct tgsi_transform_context *tctx,
          new_inst.Src[0].Register.Absolute = true;
          new_inst.Src[0].Register.Negate = false;
          reg_src(&new_inst.Src[1], &ctx->tmp[A].src, SWIZ(X, Y, Z, W));
+         new_inst.Src[1].Register.Negate = 1;
          tctx->emit_instruction(tctx, &new_inst);
       } else {
          new_inst = tgsi_default_full_instruction();
diff --git a/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h b/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
index d78dd6615b2..13c443fa842 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
@@ -73,7 +73,6 @@ OP12(MAX)
 OP12(SLT)
 OP12(SGE)
 OP13(MAD)
-OP12(SUB)
 OP13(LRP)
 OP11(SQRT)
 OP13(DP2A)
diff --git a/src/gallium/auxiliary/tgsi/tgsi_point_sprite.c b/src/gallium/auxiliary/tgsi/tgsi_point_sprite.c
index 713bd609dfd..f60a17c78a8 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_point_sprite.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_point_sprite.c
@@ -295,7 +295,7 @@ psprite_emit_vertex_inst(struct tgsi_transform_context *ctx,
    tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_MUL,
                   TGSI_FILE_TEMPORARY, ts->point_scale_tmp, TGSI_WRITEMASK_X,
                   TGSI_FILE_TEMPORARY, ts->point_size_tmp, TGSI_SWIZZLE_X,
-                  TGSI_FILE_TEMPORARY, ts->point_pos_tmp, TGSI_SWIZZLE_W);
+                  TGSI_FILE_TEMPORARY, ts->point_pos_tmp, TGSI_SWIZZLE_W, false);
 
    /* MUL point_scale.xy, point_scale.xx, inverseViewport.xy */
    inst = tgsi_default_full_instruction();
@@ -323,15 +323,15 @@ psprite_emit_vertex_inst(struct tgsi_transform_context *ctx,
                                   TGSI_FILE_IMMEDIATE, ts->point_imm,
                                   TGSI_SWIZZLE_Y,
                                   TGSI_FILE_TEMPORARY, ts->point_size_tmp,
-                                  TGSI_SWIZZLE_X);
+                                  TGSI_SWIZZLE_X, false);
 
-      tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_SUB,
+      tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_ADD,
                                   TGSI_FILE_TEMPORARY, ts->point_coord_k,
                                   TGSI_WRITEMASK_X,
                                   TGSI_FILE_IMMEDIATE, ts->point_imm,
                                   TGSI_SWIZZLE_Z,
                                   TGSI_FILE_TEMPORARY, ts->point_coord_k,
-                                  TGSI_SWIZZLE_X);
+                                  TGSI_SWIZZLE_X, true);
    }
 
 
@@ -442,13 +442,13 @@ psprite_inst(struct tgsi_transform_context *ctx,
       tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_MAX,
                  TGSI_FILE_TEMPORARY, ts->point_size_tmp, TGSI_WRITEMASK_X,
                  TGSI_FILE_TEMPORARY, ts->point_size_tmp, TGSI_SWIZZLE_X,
-                 TGSI_FILE_IMMEDIATE, ts->point_imm, TGSI_SWIZZLE_Y);
+                 TGSI_FILE_IMMEDIATE, ts->point_imm, TGSI_SWIZZLE_Y, false);
 
       /* MIN point_size_tmp.x, point_size_tmp.x, point_ivp.w */
       tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_MIN,
                  TGSI_FILE_TEMPORARY, ts->point_size_tmp, TGSI_WRITEMASK_X,
                  TGSI_FILE_TEMPORARY, ts->point_size_tmp, TGSI_SWIZZLE_X,
-                 TGSI_FILE_CONSTANT, ts->point_ivp, TGSI_SWIZZLE_W);
+                 TGSI_FILE_CONSTANT, ts->point_ivp, TGSI_SWIZZLE_W, false);
    }
    else if (inst->Dst[0].Register.File == TGSI_FILE_OUTPUT &&
             inst->Dst[0].Register.Index == ts->point_pos_out) {
diff --git a/src/gallium/auxiliary/tgsi/tgsi_transform.h b/src/gallium/auxiliary/tgsi/tgsi_transform.h
index c21ff959cbf..7ea82066fcf 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_transform.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_transform.h
@@ -281,7 +281,8 @@ tgsi_transform_op2_inst(struct tgsi_transform_context *ctx,
                         unsigned src0_file,
                         unsigned src0_index,
                         unsigned src1_file,
-                        unsigned src1_index)
+                        unsigned src1_index,
+                        bool src1_negate)
 {
    struct tgsi_full_instruction inst;
 
@@ -296,6 +297,7 @@ tgsi_transform_op2_inst(struct tgsi_transform_context *ctx,
    inst.Src[0].Register.Index = src0_index;
    inst.Src[1].Register.File = src1_file;
    inst.Src[1].Register.Index = src1_index;
+   inst.Src[1].Register.Negate = src1_negate;
 
    ctx->emit_instruction(ctx, &inst);
 }
@@ -388,7 +390,8 @@ tgsi_transform_op2_swz_inst(struct tgsi_transform_context *ctx,
                             unsigned src0_swizzle,
                             unsigned src1_file,
                             unsigned src1_index,
-                            unsigned src1_swizzle)
+                            unsigned src1_swizzle,
+                            bool src1_negate)
 {
    struct tgsi_full_instruction inst;
 
@@ -403,6 +406,7 @@ tgsi_transform_op2_swz_inst(struct tgsi_transform_context *ctx,
    inst.Src[0].Register.Index = src0_index;
    inst.Src[1].Register.File = src1_file;
    inst.Src[1].Register.Index = src1_index;
+   inst.Src[1].Register.Negate = src1_negate;
    switch (dst_writemask) {
    case TGSI_WRITEMASK_X:
       inst.Src[0].Register.SwizzleX = src0_swizzle;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_util.c b/src/gallium/auxiliary/tgsi/tgsi_util.c
index 4f02829e14d..4a6a2ae0484 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_util.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_util.c
@@ -192,7 +192,6 @@ tgsi_util_get_inst_usage_mask(const struct tgsi_full_instruction *inst,
    case TGSI_OPCODE_SLT:
    case TGSI_OPCODE_SGE:
    case TGSI_OPCODE_MAD:
-   case TGSI_OPCODE_SUB:
    case TGSI_OPCODE_LRP:
    case TGSI_OPCODE_FMA:
    case TGSI_OPCODE_FRC:
diff --git a/src/gallium/auxiliary/util/u_pstipple.c b/src/gallium/auxiliary/util/u_pstipple.c
index f6ea535c413..ae4cfa1d76a 100644
--- a/src/gallium/auxiliary/util/u_pstipple.c
+++ b/src/gallium/auxiliary/util/u_pstipple.c
@@ -344,7 +344,7 @@ pstip_transform_prolog(struct tgsi_transform_context *ctx)
                            TGSI_FILE_TEMPORARY, texTemp,
                            TGSI_WRITEMASK_XYZW,
                            pctx->wincoordFile, wincoordInput,
-                           TGSI_FILE_IMMEDIATE, pctx->numImmed);
+                           TGSI_FILE_IMMEDIATE, pctx->numImmed, false);
 
    /* TEX texTemp, texTemp, sampler, 2D; */
    tgsi_transform_tex_inst(ctx,
diff --git a/src/gallium/auxiliary/vl/vl_bicubic_filter.c b/src/gallium/auxiliary/vl/vl_bicubic_filter.c
index 0364d432687..774702c600e 100644
--- a/src/gallium/auxiliary/vl/vl_bicubic_filter.c
+++ b/src/gallium/auxiliary/vl/vl_bicubic_filter.c
@@ -186,8 +186,8 @@ create_frag_shader(struct vl_bicubic_filter *filter, unsigned video_width,
     * t = frac(temp)
     * vtex = floor(i_vtex)/i_size
     */
-   ureg_SUB(shader, ureg_writemask(t_array[21], TGSI_WRITEMASK_XY),
-            i_vtex, half_pixel);
+   ureg_ADD(shader, ureg_writemask(t_array[21], TGSI_WRITEMASK_XY),
+            i_vtex, ureg_negate(half_pixel));
    ureg_MUL(shader, ureg_writemask(t_array[22], TGSI_WRITEMASK_XY),
             ureg_src(t_array[21]), ureg_imm2f(shader, video_width, video_height));
    ureg_FRC(shader, ureg_writemask(t, TGSI_WRITEMASK_XY),
diff --git a/src/gallium/auxiliary/vl/vl_compositor.c b/src/gallium/auxiliary/vl/vl_compositor.c
index 297c3ab9dc7..1efcb752526 100644
--- a/src/gallium/auxiliary/vl/vl_compositor.c
+++ b/src/gallium/auxiliary/vl/vl_compositor.c
@@ -151,8 +151,8 @@ create_frag_shader_weave(struct ureg_program *shader, struct ureg_dst fragment)
     */
    for (i = 0; i < 2; ++i) {
       ureg_MOV(shader, ureg_writemask(t_tc[i], TGSI_WRITEMASK_X), i_tc[i]);
-      ureg_SUB(shader, ureg_writemask(t_tc[i], TGSI_WRITEMASK_YZ),
-               i_tc[i], ureg_imm1f(shader, 0.5f));
+      ureg_ADD(shader, ureg_writemask(t_tc[i], TGSI_WRITEMASK_YZ),
+               i_tc[i], ureg_imm1f(shader, -0.5f));
       ureg_ROUND(shader, ureg_writemask(t_tc[i], TGSI_WRITEMASK_YZ), ureg_src(t_tc[i]));
       ureg_MOV(shader, ureg_writemask(t_tc[i], TGSI_WRITEMASK_W),
                ureg_imm1f(shader, i ? 1.0f : 0.0f));
diff --git a/src/gallium/auxiliary/vl/vl_deint_filter.c b/src/gallium/auxiliary/vl/vl_deint_filter.c
index 3ca3b49543d..2eec5cbcd32 100644
--- a/src/gallium/auxiliary/vl/vl_deint_filter.c
+++ b/src/gallium/auxiliary/vl/vl_deint_filter.c
@@ -173,21 +173,21 @@ create_deint_frag_shader(struct vl_deint_filter *filter, unsigned field,
       // cur vs prev2
       ureg_TEX(shader, t_a, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_bot), sampler_cur);
       ureg_TEX(shader, t_b, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_bot), sampler_prevprev);
-      ureg_SUB(shader, ureg_writemask(t_diff, TGSI_WRITEMASK_X), ureg_src(t_a), ureg_src(t_b));
+      ureg_ADD(shader, ureg_writemask(t_diff, TGSI_WRITEMASK_X), ureg_src(t_a), ureg_negate(ureg_src(t_b)));
       // prev vs next
       ureg_TEX(shader, t_a, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_top), sampler_prev);
       ureg_TEX(shader, t_b, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_top), sampler_next);
-      ureg_SUB(shader, ureg_writemask(t_diff, TGSI_WRITEMASK_Y), ureg_src(t_a), ureg_src(t_b));
+      ureg_ADD(shader, ureg_writemask(t_diff, TGSI_WRITEMASK_Y), ureg_src(t_a), ureg_negate(ureg_src(t_b)));
    } else {
       /* interpolating bottom field -> current field is a top field */
       // cur vs prev2
       ureg_TEX(shader, t_a, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_top), sampler_cur);
       ureg_TEX(shader, t_b, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_top), sampler_prevprev);
-      ureg_SUB(shader, ureg_writemask(t_diff, TGSI_WRITEMASK_X), ureg_src(t_a), ureg_src(t_b));
+      ureg_ADD(shader, ureg_writemask(t_diff, TGSI_WRITEMASK_X), ureg_src(t_a), ureg_negate(ureg_src(t_b)));
       // prev vs next
       ureg_TEX(shader, t_a, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_bot), sampler_prev);
       ureg_TEX(shader, t_b, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_bot), sampler_next);
-      ureg_SUB(shader, ureg_writemask(t_diff, TGSI_WRITEMASK_Y), ureg_src(t_a), ureg_src(t_b));
+      ureg_ADD(shader, ureg_writemask(t_diff, TGSI_WRITEMASK_Y), ureg_src(t_a), ureg_negate(ureg_src(t_b)));
    }
 
    // absolute maximum of differences