r300/compiler: lower CEIL

diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.c b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
index 21a1c45..fcc0f9a 100644 (file)
--- a/src/gallium/drivers/r300/r300_tgsi_to_rc.c
+++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
@@ -115,7 +115,7 @@ static unsigned translate_opcode(unsigned opcode)
      /* case TGSI_OPCODE_ENDREP: return RC_OPCODE_ENDREP; */
      /* case TGSI_OPCODE_PUSHA: return RC_OPCODE_PUSHA; */
      /* case TGSI_OPCODE_POPA: return RC_OPCODE_POPA; */
-     /* case TGSI_OPCODE_CEIL: return RC_OPCODE_CEIL; */
+        case TGSI_OPCODE_CEIL: return RC_OPCODE_CEIL;
      /* case TGSI_OPCODE_I2F: return RC_OPCODE_I2F; */
      /* case TGSI_OPCODE_NOT: return RC_OPCODE_NOT; */
      /* case TGSI_OPCODE_TRUNC: return RC_OPCODE_TRUNC; */

diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.c b/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.c
index ffc9124..d593b3e 100644 (file)
--- a/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.c
@@ -60,6 +60,13 @@ struct rc_opcode_info rc_opcodes[MAX_RC_OPCODE] = {
                .HasDstReg = 1
        },
        {
+               .Opcode = RC_OPCODE_CEIL,
+               .Name = "CEIL",
+               .NumSrcRegs = 1,
+               .HasDstReg = 1,
+               .IsComponentwise = 1
+       },
+       {
                .Opcode = RC_OPCODE_CMP,
                .Name = "CMP",
                .NumSrcRegs = 3,

diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.h b/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.h
index 1c9b34d..87a2e23 100644 (file)
--- a/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.h
@@ -47,6 +47,9 @@ typedef enum {
         * dst.x = floor(src.x), where dst must be an address register */
        RC_OPCODE_ARL,
 
+       /** vec4 instruction: dst.c = ceil(src0.c) */
+       RC_OPCODE_CEIL,
+
        /** vec4 instruction: dst.c = src0.c < 0.0 ? src1.c : src2.c */
        RC_OPCODE_CMP,
 

diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c
index fced31d..05b874b 100644 (file)
--- a/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c
@@ -175,6 +175,26 @@ static void transform_ABS(struct radeon_compiler* c,
        rc_remove_instruction(inst);
 }
 
+static void transform_CEIL(struct radeon_compiler* c,
+       struct rc_instruction* inst)
+{
+       /* Assuming:
+        *     ceil(x) = -floor(-x)
+        *
+        * After inlining floor:
+        *     ceil(x) = -(-x-frac(-x))
+        *
+        * After simplification:
+        *     ceil(x) = x+frac(-x)
+        */
+
+       int tempreg = rc_find_free_temporary(c);
+       emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dstreg(RC_FILE_TEMPORARY, tempreg), negate(inst->U.I.SrcReg[0]));
+       emit2(c, inst->Prev, RC_OPCODE_ADD, inst->U.I.SaturateMode, inst->U.I.DstReg,
+               inst->U.I.SrcReg[0], srcreg(RC_FILE_TEMPORARY, tempreg));
+       rc_remove_instruction(inst);
+}
+
 static void transform_DP3(struct radeon_compiler* c,
        struct rc_instruction* inst)
 {
@@ -458,7 +478,7 @@ static void transform_XPD(struct radeon_compiler* c,
  * no userData necessary.
  *
  * Eliminates the following ALU instructions:
- *  ABS, DPH, DST, FLR, LIT, LRP, POW, SEQ, SFL, SGE, SGT, SLE, SLT, SNE, SUB, SWZ, XPD
+ *  ABS, CEIL, DPH, DST, FLR, LIT, LRP, POW, SEQ, SFL, SGE, SGT, SLE, SLT, SNE, SUB, SWZ, XPD
  * using:
  *  MOV, ADD, MUL, MAD, FRC, DP3, LG2, EX2, CMP
  *
@@ -474,6 +494,7 @@ int radeonTransformALU(
 {
        switch(inst->U.I.Opcode) {
        case RC_OPCODE_ABS: transform_ABS(c, inst); return 1;
+       case RC_OPCODE_CEIL: transform_CEIL(c, inst); return 1;
        case RC_OPCODE_DPH: transform_DPH(c, inst); return 1;
        case RC_OPCODE_DST: transform_DST(c, inst); return 1;
        case RC_OPCODE_FLR: transform_FLR(c, inst); return 1;
@@ -546,6 +567,7 @@ int r300_transform_vertex_alu(
 {
        switch(inst->U.I.Opcode) {
        case RC_OPCODE_ABS: transform_r300_vertex_ABS(c, inst); return 1;
+       case RC_OPCODE_CEIL: transform_CEIL(c, inst); return 1;
        case RC_OPCODE_CMP: transform_r300_vertex_CMP(c, inst); return 1;
        case RC_OPCODE_DP3: transform_DP3(c, inst); return 1;
        case RC_OPCODE_DPH: transform_DPH(c, inst); return 1;