preserve-emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2026-07-25 08:14:45 +02:00
Code Issues Packages Projects Releases Wiki Activity

vertexjit: Avoid switching to VFP from NEON.

Browse Source
This commit is contained in:
Unknown W. Brackets
2014-03-23 16:26:17 -07:00
parent a1301f5c4d
commit 58fe022ecd
1 changed files with 49 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files
GPU/GLES/VertexDecoderArm.cpp
+49 -28
View File
@@ -702,13 +702,15 @@ void VertexDecoderJitCache::Jit_Color8888Morph() {
VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);
VLDR(S12, tempReg2, sizeof(float) * n);
VLD1_all_lanes(F_32, Q3, tempReg2, true, REG_UPDATE);
if (first) {
first = false;
VMUL_scalar(F_32, Q2, neonScratchRegQ, QScalar(Q3, 0));
VMUL(F_32, Q2, neonScratchRegQ, Q3);
} else if (cpu_info.bVFPv4) {
VFMA(F_32, Q2, neonScratchRegQ, Q3);
} else {
VMLA_scalar(F_32, Q2, neonScratchRegQ, QScalar(Q3, 0));
VMLA(F_32, Q2, neonScratchRegQ, Q3);
}
} else {
LDRB(scratchReg, tempReg1, 0);
@@ -755,10 +757,10 @@ void VertexDecoderJitCache::Jit_Color4444Morph() {
MOVP2R(tempReg2, &gstate_c.morphWeights[0]);
MOVI2F(S13, 255.0f / 15.0f, scratchReg);
if (useNEON) {
MOVP2R(scratchReg, color4444Shift);
VLD1(I_16, D8, scratchReg, 2, ALIGN_128);
VDUP(F_32, D10, D6, 1);
}
bool first = true;
@@ -773,14 +775,16 @@ void VertexDecoderJitCache::Jit_Color4444Morph() {
VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);
VLDR(S12, tempReg2, sizeof(float) * n);
VMUL(S12, S12, S13);
VLD1_all_lanes(F_32, Q3, tempReg2, true, REG_UPDATE);
VMUL(F_32, Q3, Q3, Q5);
if (first) {
first = false;
VMUL_scalar(F_32, Q2, neonScratchRegQ, QScalar(Q3, 0));
VMUL(F_32, Q2, neonScratchRegQ, Q3);
} else if (cpu_info.bVFPv4) {
VFMA(F_32, Q2, neonScratchRegQ, Q3);
} else {
VMLA_scalar(F_32, Q2, neonScratchRegQ, QScalar(Q3, 0));
VMLA(F_32, Q2, neonScratchRegQ, Q3);
}
} else {
LDRB(scratchReg, tempReg1, 0);
@@ -836,9 +840,8 @@ void VertexDecoderJitCache::Jit_Color565Morph() {
if (useNEON) {
MOVP2R(scratchReg, color565Shift);
VLD1(I_16, D8, scratchReg, 2, ALIGN_128);
MOVI2R(scratchReg2, (char *)byColor565 - (char *)color565Shift);
VLD1(I_16, D8, scratchReg, 2, ALIGN_128);
VLD1(F_32, D10, scratchReg, 2, ALIGN_128, scratchReg2);
} else {
MOVI2F(S14, 255.0f / 31.0f, scratchReg);
@@ -855,12 +858,14 @@ void VertexDecoderJitCache::Jit_Color565Morph() {
VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);
VLDR(S12, tempReg2, sizeof(float) * n);
VMUL_scalar(F_32, Q3, Q5, QScalar(Q3, 0));
VLD1_all_lanes(F_32, Q3, tempReg2, true, REG_UPDATE);
VMUL(F_32, Q3, Q3, Q5);
if (first) {
first = false;
VMUL(F_32, Q2, neonScratchRegQ, Q3);
} else if (cpu_info.bVFPv4) {
VFMA(F_32, Q2, neonScratchRegQ, Q3);
} else {
VMLA(F_32, Q2, neonScratchRegQ, Q3);
}
@@ -914,9 +919,8 @@ void VertexDecoderJitCache::Jit_Color5551Morph() {
if (useNEON) {
MOVP2R(scratchReg, color5551Shift);
VLD1(I_16, D8, scratchReg, 2, ALIGN_128);
MOVI2R(scratchReg2, (char *)byColor5551 - (char *)color5551Shift);
VLD1(I_16, D8, scratchReg, 2, ALIGN_128);
VLD1(F_32, D10, scratchReg, 2, ALIGN_128, scratchReg2);
} else {
MOVI2F(S14, 255.0f / 31.0f, scratchReg);
@@ -933,12 +937,14 @@ void VertexDecoderJitCache::Jit_Color5551Morph() {
VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);
VLDR(S12, tempReg2, sizeof(float) * n);
VMUL_scalar(F_32, Q3, Q5, QScalar(Q3, 0));
VLD1_all_lanes(F_32, Q3, tempReg2, true, REG_UPDATE);
VMUL(F_32, Q3, Q3, Q5);
if (first) {
first = false;
VMUL(F_32, Q2, neonScratchRegQ, Q3);
} else if (cpu_info.bVFPv4) {
VFMA(F_32, Q2, neonScratchRegQ, Q3);
} else {
VMLA(F_32, Q2, neonScratchRegQ, Q3);
}
@@ -1278,6 +1284,9 @@ void VertexDecoderJitCache::Jit_AnyS8Morph(int srcoff, int dstoff) {
MOVP2R(tempReg2, &gstate_c.morphWeights[0]);
MOVI2F(S13, by127, scratchReg);
if (NEONMorphing) {
VDUP(F_32, D10, D6, 1);
}
bool first = true;
for (int n = 0; n < dec_->morphcount; ++n) {
@@ -1288,14 +1297,16 @@ void VertexDecoderJitCache::Jit_AnyS8Morph(int srcoff, int dstoff) {
VMOVL(I_16 | I_SIGNED, neonScratchRegQ, neonScratchReg);
VCVT(F_32 | I_SIGNED, neonScratchRegQ, neonScratchRegQ);
VLDR(S12, tempReg2, sizeof(float) * n);
VMUL(S12, S12, S13);
VLD1_all_lanes(F_32, Q3, tempReg2, true, REG_UPDATE);
VMUL(F_32, Q3, Q3, Q5);
if (first) {
first = false;
VMUL_scalar(F_32, Q2, neonScratchRegQ, QScalar(Q3, 0));
VMUL(F_32, Q2, neonScratchRegQ, Q3);
} else if (cpu_info.bVFPv4) {
VFMA(F_32, Q2, neonScratchRegQ, Q3);
} else {
VMLA_scalar(F_32, Q2, neonScratchRegQ, QScalar(Q3, 0));
VMLA(F_32, Q2, neonScratchRegQ, Q3);
}
} else {
LDRSB(scratchReg, tempReg1, 0);
@@ -1334,6 +1345,9 @@ void VertexDecoderJitCache::Jit_AnyS16Morph(int srcoff, int dstoff) {
MOVP2R(tempReg2, &gstate_c.morphWeights[0]);
MOVI2F(S13, by32767, scratchReg);
if (NEONMorphing) {
VDUP(F_32, D10, D6, 1);
}
bool first = true;
for (int n = 0; n < dec_->morphcount; ++n) {
@@ -1343,14 +1357,16 @@ void VertexDecoderJitCache::Jit_AnyS16Morph(int srcoff, int dstoff) {
VMOVL(I_16 | I_SIGNED, neonScratchRegQ, neonScratchReg);
VCVT(F_32 | I_SIGNED, neonScratchRegQ, neonScratchRegQ);
VLDR(S12, tempReg2, sizeof(float) * n);
VMUL(S12, S12, S13);
VLD1_all_lanes(F_32, Q3, tempReg2, true, REG_UPDATE);
VMUL(F_32, Q3, Q3, Q5);
if (first) {
first = false;
VMUL_scalar(F_32, Q2, neonScratchRegQ, QScalar(Q3, 0));
VMUL(F_32, Q2, neonScratchRegQ, Q3);
} else if (cpu_info.bVFPv4) {
VFMA(F_32, Q2, neonScratchRegQ, Q3);
} else {
VMLA_scalar(F_32, Q2, neonScratchRegQ, QScalar(Q3, 0));
VMLA(F_32, Q2, neonScratchRegQ, Q3);
}
} else {
LDRSH(scratchReg, tempReg1, 0);
@@ -1390,14 +1406,15 @@ void VertexDecoderJitCache::Jit_AnyFloatMorph(int srcoff, int dstoff) {
bool first = true;
for (int n = 0; n < dec_->morphcount; ++n) {
VLDMIA(tempReg1, false, fpScratchReg, 3);
// Load an extra float to stay in NEON mode.
VLD1(F_32, neonScratchRegQ, tempReg1, 2, ALIGN_NONE);
ADDI2R(tempReg1, tempReg1, dec_->onesize_, scratchReg);
VLDR(S12, tempReg2, sizeof(float) * n);
VLD1_all_lanes(F_32, Q3, tempReg2, true, REG_UPDATE);
if (first) {
first = false;
if (NEONMorphing) {
VMUL_scalar(F_32, Q2, neonScratchRegQ, QScalar(Q3, 0));
VMUL(F_32, Q2, neonScratchRegQ, Q3);
} else {
VMUL(S8, fpScratchReg, S12);
VMUL(S9, fpScratchReg2, S12);
@@ -1405,7 +1422,11 @@ void VertexDecoderJitCache::Jit_AnyFloatMorph(int srcoff, int dstoff) {
}
} else {
if (NEONMorphing) {
VMLA_scalar(F_32, Q2, neonScratchRegQ, QScalar(Q3, 0));
if (cpu_info.bVFPv4) {
VFMA(F_32, Q2, neonScratchRegQ, Q3);
} else {
VMLA(F_32, Q2, neonScratchRegQ, Q3);
}
} else {
VMLA(S8, fpScratchReg, S12);
VMLA(S9, fpScratchReg2, S12);