Mesa (master): gallium/drivers: handle TGSI_OPCODE_CEIL

[Christoph Bumiller]

Module: Mesa
Branch: master
Commit: 6ae12bac596ce3a6aa8e09f638ad2cb4a7c18e5c
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=6ae12bac596ce3a6aa8e09f638ad2cb4a7c18e5c

Author: Christoph Bumiller <e0425955 at student.tuwien.ac.at>
Date:   Sat May  5 14:36:40 2012 +0200

gallium/drivers: handle TGSI_OPCODE_CEIL

---

 src/gallium/drivers/i915/i915_fpc_translate.c |   16 ++++++++++++++++
 src/gallium/drivers/nv30/nvfx_fragprog.c      |    5 +++++
 src/gallium/drivers/nv30/nvfx_vertprog.c      |    5 +++++
 src/gallium/drivers/r600/r600_asm.c           |    2 ++
 4 files changed, 28 insertions(+), 0 deletions(-)

diff --git a/src/gallium/drivers/i915/i915_fpc_translate.c b/src/gallium/drivers/i915/i915_fpc_translate.c
index 94c0c69..def9a03 100644
--- a/src/gallium/drivers/i915/i915_fpc_translate.c
+++ b/src/gallium/drivers/i915/i915_fpc_translate.c
@@ -512,6 +512,22 @@ i915_translate_instruction(struct i915_fp_compile *p,
       emit_simple_arith(p, inst, A0_ADD, 2, fs);
       break;
 
+   case TGSI_OPCODE_CEIL:
+      src0 = src_vector(p, &inst->Src[0], fs);
+      tmp = i915_get_utemp(p);
+      flags = get_result_flags(inst);
+      i915_emit_arith(p,
+                      A0_FLR,
+                      tmp,
+                      flags & A0_DEST_CHANNEL_ALL, 0,
+                      negate(src0, 1, 1, 1, 1), 0, 0);
+      i915_emit_arith(p,
+                      A0_MOV,
+                      get_result_vector(p, &inst->Dst[0]),
+                      flags, 0,
+                      negate(tmp, 1, 1, 1, 1), 0, 0);
+      break;
+
    case TGSI_OPCODE_CMP:
       src0 = src_vector(p, &inst->Src[0], fs);
       src1 = src_vector(p, &inst->Src[1], fs);
diff --git a/src/gallium/drivers/nv30/nvfx_fragprog.c b/src/gallium/drivers/nv30/nvfx_fragprog.c
index e562b45..592ad21 100644
--- a/src/gallium/drivers/nv30/nvfx_fragprog.c
+++ b/src/gallium/drivers/nv30/nvfx_fragprog.c
@@ -535,6 +535,11 @@ nvfx_fragprog_parse_instruction(struct nv30_context* nvfx, struct nvfx_fpc *fpc,
    case TGSI_OPCODE_ADD:
       nvfx_fp_emit(fpc, arith(sat, ADD, dst, mask, src[0], src[1], none));
       break;
+   case TGSI_OPCODE_CEIL:
+      tmp = nvfx_src(temp(fpc));
+      nvfx_fp_emit(fpc, arith(0, FLR, tmp.reg, mask, neg(src[0]), none, none));
+      nvfx_fp_emit(fpc, arith(sat, MOV, dst, mask, neg(tmp), none, none));
+      break;
    case TGSI_OPCODE_CMP:
       insn = arith(0, MOV, none.reg, mask, src[0], none, none);
       insn.cc_update = 1;
diff --git a/src/gallium/drivers/nv30/nvfx_vertprog.c b/src/gallium/drivers/nv30/nvfx_vertprog.c
index d7eb9fb..82972b3 100644
--- a/src/gallium/drivers/nv30/nvfx_vertprog.c
+++ b/src/gallium/drivers/nv30/nvfx_vertprog.c
@@ -550,6 +550,11 @@ nvfx_vertprog_parse_instruction(struct nv30_context *nv30, struct nvfx_vpc *vpc,
    case TGSI_OPCODE_ARL:
       nvfx_vp_emit(vpc, arith(0, VEC, ARL, dst, mask, src[0], none, none));
       break;
+   case TGSI_OPCODE_CEIL:
+      tmp = nvfx_src(temp(vpc));
+      nvfx_vp_emit(vpc, arith(0, VEC, FLR, tmp.reg, mask, neg(src[0]), none, none));
+      nvfx_vp_emit(vpc, arith(sat, VEC, MOV, dst, mask, neg(tmp), none, none));
+      break;
    case TGSI_OPCODE_CMP:
       insn = arith(0, VEC, MOV, none.reg, mask, src[0], none, none);
       insn.cc_update = 1;
diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c
index 0d67fbd..5a10bd9 100644
--- a/src/gallium/drivers/r600/r600_asm.c
+++ b/src/gallium/drivers/r600/r600_asm.c
@@ -94,6 +94,7 @@ static inline unsigned int r600_bytecode_get_num_operands(struct r600_bytecode *
 		case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_GPR_INT:
 		case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_INT:
 		case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FRACT:
+		case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_CEIL:
 		case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR:
 		case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_TRUNC:
 		case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE:
@@ -172,6 +173,7 @@ static inline unsigned int r600_bytecode_get_num_operands(struct r600_bytecode *
 		case EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV:
 		case EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_INT:
 		case EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FRACT:
+		case EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_CEIL:
 		case EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR:
 		case EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_TRUNC:
 		case EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE: