summaryrefslogtreecommitdiffstats
path: root/src/mesa
diff options
context:
space:
mode:
authorMarek Olšák <maraeo@gmail.com>2010-04-18 23:40:01 +0200
committerMarek Olšák <maraeo@gmail.com>2010-04-19 00:36:20 +0200
commit394803b2cd7fd9d4c0a394eb6998ccafec19bcd3 (patch)
tree5b7712fc77b51fa3982a12caa45af2aa3e4d8521 /src/mesa
parente53a1576dde7ba9d22b48a8e4b5d41b71411ec1d (diff)
r300/compiler: lower CEIL
Diffstat (limited to 'src/mesa')
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_opcodes.c7
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_opcodes.h3
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c24
3 files changed, 33 insertions, 1 deletions
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.c b/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.c
index ffc91241ab8..d593b3e81ae 100644
--- 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 1c9b34df78e..87a2e23084c 100644
--- 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 fced31d6cb9..05b874ba7cf 100644
--- 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;