summaryrefslogtreecommitdiff
path: root/src/mesa
diff options
context:
space:
mode:
authorAndre Maasikas <amaasikas@gmail.com>2009-09-21 10:14:25 -0400
committerAlex Deucher <alexdeucher@gmail.com>2009-09-21 10:14:25 -0400
commit2d729e6e3bcb0af84790cafb9824a3937954e078 (patch)
tree93fe43707c7433a2771f667dda5fcc7d97ca6778 /src/mesa
parent3083ba38f4c884b32cd0460607b5064b6b7008d2 (diff)
r600: fix some issues with LIT instruction
- MUL_LIT is ALU.Trans instruction - some Trans instructions can take 3 arguments - don't clobber dst.x, use dst.z as temp, it'll get written correct value in last insn - respect source swizzles
Diffstat (limited to 'src/mesa')
-rw-r--r--src/mesa/drivers/dri/r600/r700_assembler.c69
1 files changed, 36 insertions, 33 deletions
diff --git a/src/mesa/drivers/dri/r600/r700_assembler.c b/src/mesa/drivers/dri/r600/r700_assembler.c
index efeccb25f1..f46bc32201 100644
--- a/src/mesa/drivers/dri/r600/r700_assembler.c
+++ b/src/mesa/drivers/dri/r600/r700_assembler.c
@@ -2024,7 +2024,7 @@ GLboolean assemble_alu_instruction(r700_AssemblerBase *pAsm)
return GL_FALSE;
}
- if (pAsm->D.dst.math == 0)
+ if (uNumSrc > 1)
{
// Process source 1
current_source_index = 1;
@@ -2880,6 +2880,11 @@ GLboolean assemble_LIT(r700_AssemblerBase *pAsm)
return GL_FALSE;
}
+ if( GL_FALSE == assemble_src(pAsm, 0, -1) )
+ {
+ return GL_FALSE;
+ }
+
/* dst.y = max(src.x, 0.0) */
pAsm->D.dst.opcode = SQ_OP2_INST_MAX;
pAsm->D.dst.rtype = dstType;
@@ -2891,11 +2896,6 @@ GLboolean assemble_LIT(r700_AssemblerBase *pAsm)
pAsm->S[0].src.rtype = srcType;
pAsm->S[0].src.reg = srcReg;
setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE);
- noneg_PVSSRC(&(pAsm->S[0].src));
- pAsm->S[0].src.swizzlex = SQ_SEL_X;
- pAsm->S[0].src.swizzley = SQ_SEL_X;
- pAsm->S[0].src.swizzlez = SQ_SEL_X;
- pAsm->S[0].src.swizzlew = SQ_SEL_X;
pAsm->S[1].src.rtype = SRC_REG_TEMPORARY;
pAsm->S[1].src.reg = tmp;
setaddrmode_PVSSRC(&(pAsm->S[1].src), ADDR_ABSOLUTE);
@@ -2909,34 +2909,47 @@ GLboolean assemble_LIT(r700_AssemblerBase *pAsm)
return GL_FALSE;
}
- /* before: dst.w = log(src.y)
- * after : dst.x = log(src.y)
- * why change dest register is that dst.w has been initialized as 1 before
- */
+ if( GL_FALSE == assemble_src(pAsm, 0, -1) )
+ {
+ return GL_FALSE;
+ }
+
+ swizzleagain_PVSSRC(&(pAsm->S[0].src), SQ_SEL_Y, SQ_SEL_Y, SQ_SEL_Y, SQ_SEL_Y);
+
+ /* dst.z = log(src.y) */
pAsm->D.dst.opcode = SQ_OP2_INST_LOG_CLAMPED;
pAsm->D.dst.math = 1;
pAsm->D.dst.rtype = dstType;
pAsm->D.dst.reg = dstReg;
- pAsm->D.dst.writex = 1;
+ pAsm->D.dst.writex = 0;
pAsm->D.dst.writey = 0;
- pAsm->D.dst.writez = 0;
+ pAsm->D.dst.writez = 1;
pAsm->D.dst.writew = 0;
pAsm->S[0].src.rtype = srcType;
pAsm->S[0].src.reg = srcReg;
setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE);
- noneg_PVSSRC(&(pAsm->S[0].src));
- pAsm->S[0].src.swizzlex = SQ_SEL_Y;
- pAsm->S[0].src.swizzley = SQ_SEL_Y;
- pAsm->S[0].src.swizzlez = SQ_SEL_Y;
- pAsm->S[0].src.swizzlew = SQ_SEL_Y;
if( GL_FALSE == next_ins(pAsm) )
{
return GL_FALSE;
}
- /* before: tmp.x = amd MUL_LIT(src.w, dst.w, src.x ) */
- /* after : tmp.x = amd MUL_LIT(src.w, dst.x, src.x ) */
+ if( GL_FALSE == assemble_src(pAsm, 0, -1) )
+ {
+ return GL_FALSE;
+ }
+
+ if( GL_FALSE == assemble_src(pAsm, 0, 2) )
+ {
+ return GL_FALSE;
+ }
+
+ swizzleagain_PVSSRC(&(pAsm->S[0].src), SQ_SEL_W, SQ_SEL_W, SQ_SEL_W, SQ_SEL_W);
+
+ swizzleagain_PVSSRC(&(pAsm->S[2].src), SQ_SEL_X, SQ_SEL_X, SQ_SEL_X, SQ_SEL_X);
+
+ /* tmp.x = amd MUL_LIT(src.w, dst.z, src.x ) */
pAsm->D.dst.opcode = SQ_OP3_INST_MUL_LIT;
+ pAsm->D.dst.math = 1;
pAsm->D.dst.op3 = 1;
pAsm->D.dst.rtype = DST_REG_TEMPORARY;
pAsm->D.dst.reg = tmp;
@@ -2948,29 +2961,19 @@ GLboolean assemble_LIT(r700_AssemblerBase *pAsm)
pAsm->S[0].src.rtype = srcType;
pAsm->S[0].src.reg = srcReg;
setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE);
- noneg_PVSSRC(&(pAsm->S[0].src));
- pAsm->S[0].src.swizzlex = SQ_SEL_W;
- pAsm->S[0].src.swizzley = SQ_SEL_W;
- pAsm->S[0].src.swizzlez = SQ_SEL_W;
- pAsm->S[0].src.swizzlew = SQ_SEL_W;
pAsm->S[1].src.rtype = SRC_REG_TEMPORARY;
pAsm->S[1].src.reg = dstReg;
setaddrmode_PVSSRC(&(pAsm->S[1].src), ADDR_ABSOLUTE);
noneg_PVSSRC(&(pAsm->S[1].src));
- pAsm->S[1].src.swizzlex = SQ_SEL_X;
- pAsm->S[1].src.swizzley = SQ_SEL_X;
- pAsm->S[1].src.swizzlez = SQ_SEL_X;
- pAsm->S[1].src.swizzlew = SQ_SEL_X;
+ pAsm->S[1].src.swizzlex = SQ_SEL_Z;
+ pAsm->S[1].src.swizzley = SQ_SEL_Z;
+ pAsm->S[1].src.swizzlez = SQ_SEL_Z;
+ pAsm->S[1].src.swizzlew = SQ_SEL_Z;
pAsm->S[2].src.rtype = srcType;
pAsm->S[2].src.reg = srcReg;
setaddrmode_PVSSRC(&(pAsm->S[2].src), ADDR_ABSOLUTE);
- noneg_PVSSRC(&(pAsm->S[2].src));
- pAsm->S[2].src.swizzlex = SQ_SEL_X;
- pAsm->S[2].src.swizzley = SQ_SEL_X;
- pAsm->S[2].src.swizzlez = SQ_SEL_X;
- pAsm->S[2].src.swizzlew = SQ_SEL_X;
if( GL_FALSE == next_ins(pAsm) )
{